University of Southern California Dissertations and Theses

Cities on the environmental justice frontline: the intractability of hazards and the governability of amenities

(USC Thesis Other)

Cities on the environmental justice frontline: the intractability of hazards and the governability of amenities

PDF

Download a page range

Download transcript

Copy asset link

Request this asset

Transcript (if available)

Content

CITIES ON THE ENVIRONMENTAL JUSTICE FRONTLINE: THE
INTRACTABILITY OF HAZARDS AND THE GOVERNABILITY OF AMENITIES

by

Diane E. Yoder

________________________________________________________________________

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

December 2009

Copyright 2009 Diane E. Yoder

ii
DEDICATION

In memory of my beloved sister, Lark Folsom Norman.
iii
ACKNOWLEDGMENTS

Before I list the many people who inspired me throughout my dissertation
process, I would like to explain my motivation for the topic of environmental equity. I
was first introduced to this topic when I read about the Los Angeles Unified School
District’s construction of the $200 million Belmont Learning Complex in downtown Los
Angeles on an abandoned oil field replete with pockets of methane gas. At that same
time, the district was purchasing 40 acres in South Gate for the construction of two
schools—40 acres that housed foundries, a chemical manufacturer, and other industries
on soil that had tested positive for chromium, benzene, methane, and other toxic
chemicals. Both construction sites were in high-minority, low-income areas, and as
important as additional schools were, I was horrified at the thought of minority and low-
income children learning and playing on hazardous and potentially lethal land. I asked
how this could happen, and some answered, “environmental injustice.” I was drawn to
this subject because of my strong motivation for fairness in society and policy, as well as
the complexity and insidious nature of the problem. While on the surface environmental
injustice might look like housing discrimination or corporate malfeasance, it may be a
function of institutionalized racism, enduring class disparities in a capitalist society, and
unsuccessful social policies. Environmental injustice is complex and not easily solved;
any effort to understand and document it is worth undertaking.
That being said, this study would not have been possible were it not for the many
people who inspired me along the way. At the University of Southern California, I thank
my chair, Dr. Shui-Yan Tang, who instantly challenged me on the first day of class and
iv
has remained my greatest teacher, advocate, and mentor. Without his help, my
dissertation never would have materialized. I also thank my final committee members,
Drs. Lisa Schweitzer and Jefferey Sellers, for their intellect, encouragement, and their
efforts to rein in my zeal when needed. I am grateful to my previous committee members,
Drs. Sheldon Kamieniecki and Robert Stallings, for their time, energy, and help in
formulating my early ideas. Dr. John P. Wilson was instrumental in helping me
understand and gain confidence in the use of Geographic Information Systems (GIS).
Drs. James M. Ferris and Elizabeth Graddy also nurtured me through my studies, and Dr.
Graddy was instrumental in allowing me to complete my dissertation after a lengthy
absence. June Muranaka, the doctoral student services specialist, was patient and kind as
I waded through the administrative morass. I must also acknowledge Dr. Terry L.
Cooper, a trusted mentor and collaborator, whose belief in me never faltered. Finally, I
am grateful to the School of Policy, Planning, and Development and the John and Dora
Haynes Foundation for financial support.
Personally, I thank my family for their encouragement—I only wish my mother,
sister, and brother were still alive to witness my accomplishment. Similarly, I thank Dr.
James P. Lester, who was an important mentor and friend during the two years preceding
his death. Finally, this dissertation would not have been possible without the help of
Eileen Goff, who tirelessly answered my technical questions, and my friends Kathleen K.
Shannon and Jayne London, who talked me down from the ledge more than once.
To say that this dissertation would not have been completed without the help of
the above-mentioned people is a gross understatement. My future scholarship will bear
the roots of those listed here, though any shortcomings will remain mine alone.
v
TABLE OF CONTENTS

Dedication ii

Acknowledgements iii

List of Tables vii

List of Figures xiv

Abbreviations xv

Abstract xvi

Chapter 1: Introduction 1
The Context of and Rationale for Environmental Equity Studies 7
General Suppositions and Questions 20
Contributions of this Study 22
Organization of the Dissertation 23

Chapter 2: The Environmental Equity Literature 24
A Brief Overview of 25 Years 25
Methodological Problems 29
Gaps in the Environmental Equity Literature 38
Concluding Thoughts 49

Chapter 3: Research Design and Methods 51
Overall Research Design 51
Population and Sample 53
Phase I Research Methods 56

Chapter 4: Environmental Disamenities 83
TRI Facilities 90
Large-Quantity Generators 110
Conclusion 128

Chapter 5: Environmental Amenities 139
Some Preliminary Observations 139
Parks 143
Community Gardens 158
Conclusion 172

Chapter 6: Introduction to Political Analysis 183
Environmental Justice and Policymaking in Cities: The Usual Suspects 184
Theoretical Framework: Social Movements 188
vi
Research Design for Phase II 200

Chapter 7: Equity and the City 207
Equity and the City 208
Institutional Variables and Environmental Equity 217
Conclusion 236

Chapter 8: Conclusion 243
Summary of Major Empirical Findings 243
Implications and Policy Recommendations 247
Limitations and Future Research 249

Bibliography 252

Appendix 287
Procedures for Creating a GIS Map 287
Geoprocessing and Spatial Statistics Analysis 289
Regression Analysis 311
Regression Results 313
Disamenity Rankings for Medium and Small Cities 323
Amenity Rankings for Medium and Small Cities 326

vii
LIST OF TABLES

Table 2.1 Environmental Equity Studies Using Multiple Dependent Variables 27

Table 3.1 Cities Selected for Study 54

Table 3.2 Disamenity and Amenity Counts Per City 58

Table 3.3 Independent Variables Used in Phase I 62

Table 3.4 Asian Populations in Selected Cities 64

Table 3.5 Control Variables Used in Phase I 66

Table 3.6 Number of Census Tracts by City 73

Table 3.7 Hazard Models Used for Six Largest Cities 74

Table 3.8 Hazard Models Used for Six Medium-Sized Cities 75

Table 3.9 Park Models Used for Six Largest Cities 76

Table 3.10 Community Garden Models Used for Six Largest Cities 77

Table 4.1 Total TRI and LQG Sites by City 84

Table 4.2 Total Facilities and Pollution Amounts 87

Table 4.3 Cities with Hazard Mean Centers Located in Same Tract 88

Table 4.4 TRI Facility with Largest Release 92

Table 4.5 Moran’s Index for TRI Sites 93

Table 4.6 Ripley’s K Cluster Analysis for TRI Sites 95

Table 4.7 TRI “Hot Spots” or Getis-Ord Gi* Results 97

Table 4.8 TRI “Cold Spots” or Getis-Ord Gi* Results 100

Table 4.9 TRI Regression Results for Six Largest Cities 103

Table 4.10 TRI/OLS Regression Results for Austin Model III 105

viii
Table 4.11 TRI/OLS Regression Results for Boston Model I 105

Table 4.12 TRI/OLS Regression Results for Philadelphia Model III 106

Table 4.13 TRI/OLS Regression Results for San Diego Model III 106

Table 4.14 TRI/OLS Regression Results for San Jose Model III 107

Table 4.15 TRI/OLS Regression Results for Seattle Model IV 107

Table 4.16 TRI Regression Results for Five Medium-sized Cities 109

Table 4.17 LQGs with the Most Tons Managed 112

Table 4.18 Moran’s Index for LQGs 113

Table 4.19 Ripley’s K Cluster Analysis for LQGs 115

Table 4.20 LQG “Hot Spots” or Getis-Ord Gi* Results 116

Table 4.21 LQG “Cold Spots” or Getis-Ord Gi* Results 120

Table 4.22 LQG Regression Results for Six Largest Cities 121

Table 4.23 LQG/OLS Regression Results for Austin Model I 123

Table 4.24 LQG/OLS Regression Results for Boston Model III 123

Table 4.25 LQG/OLS Regression Results for Philadelphia Model III 124

Table 4.26 LQG/OLS Regression Results for San Diego Model IV 124

Table 4.27 LQG/OLS Regression Results for San Jose Model III 125

Table 4.28 LQG/OLS Regression Results for Seattle Model IV 125

Table 4.29 LQG Regression Results for Six Medium-sized Cities 127

Table 4.30 TRI/City Comparison by Region 129

Table 4-31 TRI/City Comparison by Population 132

Table 4-32 LQG/City Comparison by Region 134

Table 4-33 LQG/City Comparison by Population 136
ix

Table 5.1 Total Parks and Community Gardens by City 140

Table 5.2 Moran’s Index for Parks 145

Table 5.3 Ripley’s K Cluster Results for Parks 147

Table 5.4 Park “Hot Spots” or Getis-Ord Gi* Results 148

Table 5.5 Park “Cold Spots” or Getis-Ord Gi* Results 152

Table 5.6 Park Regression Results for Six Largest Cities 154

Table 5.7 Park/OLS Regression Results for Austin Model I 155

Table 5.8 Park/OLS Regression Results for Boston Model II 155

Table 5.9 Park/OLS Regression Results for Philadelphia Model II 155

Table 5.10 Park/OLS Regression Results for San Diego Model II 156

Table 5.11 Park/OLS Regression Results for San Jose Model I 156

Table 5.12 Park/OLS Regression Results for Seattle Model III 157

Table 5.13 Moran’s Index for Community Gardens 160

Table 5.14 Ripley’s K Cluster Analysis for Community Gardens 162

Table 5.15 Community Garden “Hot Spots” or Getis-Ord Gi* Results 163

Table 5.16 Community Garden Regression Results for Six Largest Cities 169

Table 5.17 Community Garden Regression Results for Boston Model I 170

Table 5.18 Community Garden Regression Results for Philadelphia Model III 171

Table 5.19 Community Garden Regression Results for Seattle Model IV 171

Table 5.20 Park/City Comparison by Region 174

Table 5.21 Park/City Comparison by Population 177

Table 5.22 Community Garden/City Comparison by Region 179

x
Table 5.23 Community Garden/City Comparison by Population 181

Table 7.1 The Six Largest Cities and TRI Sites 210

Table 7.2 The Six Largest Cities and LQGs 211

Table 7.3 Average Hazard Ranks for the Six Largest Cities 210

Table 7.4 Average Hazard Ranks for the Six Medium Cities 212

Table 7.5 Average Hazard Ranks for the Six Smallest Cities 213

Table 7.6 The Six Largest Cities and Parks 214

Table 7.7 The Six Largest Cities and Community Gardens 215

Table 7.8 Average Amenity Rankings for the Six Largest Cities 215

Table 7.9 Average Amenity Rankings for the Six Medium Cities 216

Table 7.10 Average Amenity Ranks for the Six Smallest Cities 216

Table 7.11 Disamenity Ranks and Political Opportunity Variables in the Six 219
Largest Cities

Table 7.12 Disamenity Ranks and Political Opportunity Variables in the Six 220
Medium Cities

Table 7.13 Disamenity Ranks and Political Opportunity Variables in the Five 221
Smallest Cities

Table 7.14 Disamenity Ranks and Political Process Variables in the Six 223
Largest Cities

Table 7.15 Disamenity Ranks and Political Process Variables in the Six 224
Medium Cities

Table 7.16 Disamenity Ranks and Political Process Variables in the Five 225
Largest Cities

Table 7.17 Amenity Ranks and Political Opportunity Variables in the Six 227
Largest Cities

Table 7.18 Amenity Ranks and Political Opportunity Variables in the Six 228
Medium Cities
xi

Table 7.19 Amenity Ranks and Political Opportunity Variables in the Six 229
Smallest Cities

Table 7.20 Amenity Ranks and Political Process Variables in the Six 231
Largest Cities

Table 7.21 Amenity Ranks and Political Process Variables in the Six 232
Medium Cities

Table 7.22 Disamenity Ranks and Political Process Variables in the Five 232
Largest Cities

Table A.1 Ripley’s K Dispersion Analysis for TRI Sites 291

Table A.2 Ripley’s K Dispersion Analysis for LQGs 292

Table A.3 Ripley’s K Dispersion Analysis for Parks 292

Table A.4 Ripley’s K Dispersion Analysis for Community Gardens 293

Table A.5 TRI “Hot Spots” or Getis-Ord Gi* Results 293

Table A.6 LQG “Hot Spots” or Getis-Ord Gi* Results 296

Table A.7 TRI “Cold Spots” or Getis-Ord Gi* Results 298

Table A.8 LQG “Cold Spots” or Getis-Ord Gi* Results 300

Table A.9 Park “Hot Spots” or Getis-Ord Gi* Results 301

Table A.10 Park “Cold Spots” or Getis-Ord Gi* Results 304

Table A.11 Community Garden “Hot Spots” or Getis-Ord Gi* Results 310

Table A.12 Community Garden “Cold Spots” or Getis-Ord Gi* Results 311

Table A.13 TRI/GWR Regression Results for Austin 313

Table A.14 TRI/GWR Regression Results for Philadelphia 313

Table A.15 TRI/GWR Regression Results for San Diego 314

Table A.16 TRI/GWR Regression Results for Seattle 314

xii
Table A.17 TRI/OLS Regression Results for Dayton Model I 314

Table A.18 TRI/OLS Regression Results for Grand Rapids Model II 315

Table A.19 TRI/OLS Regression Results for Little Rock Model I 315

Table A.20 TRI/OLS Regression Results for Norfolk Model I 316

Table A.21 TRI/OLS Regression Results for Salt Lake City Model I 316

Table A.22 LQG/GWR Regression Results for Austin 317

Table A.23 LQG/GWR Regression Results for Boston 317

Table A.24 LQG/GWR Regression Results for Philadelphia 317

Table A.25 LQG/GWR Regression Results for San Diego 317

Table A.26 LQG/GWR Regression Results for Seattle 318

Table A.27 LQG/OLS Regression Results for Dayton Model II 318

Table A.28 LQG/OLS Regression Results for Grand Rapids Model II 319

Table A.29 LQG/OLS Regression Results for Little Rock Model I 319

Table A.30 LQG/OLS Regression Results for Miami Model III 320

Table A.31 LQG/OLS Regression Results for Norfolk Model III 320

Table A.32 LQG/OLS Regression Results for Salt Lake City Model II 321

Table A.33 Park/GWR Regression Results for Austin 321

Table A.34 Park/GWR Regression Results for Boston 321

Table A.35 Park/GWR Regression Results for Philadelphia 322

Table A.36 Park/GWR Regression Results for San Diego 322

Table A.37 Park/GWR Regression Results for San Jose 322

Table A.38 Park/GWR Regression Results for Seattle 322

Table A.39 Community Garden/GWR Regression Results for Boston 323
xiii

Table A.40 Community Garden/GWR Regression Results for Philadelphia 323

Table A.41 Community Garden/GWR Regression Results for Seattle 323

Table A.42 The Six Medium Cities and TRI Sites 324

Table A.43 The Six Medium Cities and LQGs 325

Table A.44 The Five Smallest Cities and TRI Sites 326

Table A.45 The Five Smallest Cities and LQGs 326

Table A.46 The Six Medium Cities and Parks 327

Table A.47 The Six Medium Cities and Community Gardens 328

Table A.48 The Six Smallest Cities and Parks 328

Table A.49 The Six Smallest Cities and Community Gardens 329

xiv
LIST OF FIGURES

Figure 3.1 Total Disamenities and Amenities by City 60

Figure 4.1 San Diego’s Disamenities 86

Figure 4.2 TRI and LQG Mean Centers in Little Rock 89

Figure 4.3 TRI Sites by City 90

Figure 4.4 Global Clustering of Austin’s Tri Sites 94

Figure 4.5 San Jose’s TRI Hot Spots 99

Figure 4.6 LQGs by City 111

Figure 4.7 Global Clustering of Salt Lake City’s LQGs 114

Figure 4.8 Seattle’s LQG Hot Spots 118

Figure 5.1 Park and Community Garden Mean Centers in Santa Fe 142

Figure 5.2 Number of Parks by City 144

Figure 5.3 Global Clustering of Parks in Philadelphia 146

Figure 5.4 Park Hot Spots in Dayton 151

Figure 5.5 Number of Community Gardens by City 159

Figure 5.6 Global Clustering of Community Gardens in Albany 161

Figure 5.7 Garden Hot Spots in Boston 167

xv
ABBREVIATIONS

GIS Geographic Information Systems

TRI Toxics Release Inventory

LQG Large Quantity Generator

EJM Environmental Justice Movement

EPA Environmental Protection Agency

RCRA Resource Conservation and Recovery Act

EO Executive Order

TSDF Treatment, Storage, and Disposal Facility

LULU Locally Undesirable Land Use

MAUP Modifiable Areal Unit Problem

UHC Unit-Hazard Coincidence Problem

RSEI Risk Screening Environmental Indicators

SF1, SF3 Summary File 1, Summary File 3

TIGER Topologically Integrated Geographic Encoding and
Referencing system

USGS United States Geological Survey

NLCD National Land Cover Data

GWR Geographically Weighted Regression

OLS Ordinary Least Squares

AIC Akaike’s Information Criterion

VIF Variance Inflation Factor

xvi

ABSTRACT

In 1982, the predominantly African American residents of Warren County, NC,
protested for six weeks against the siting of a landfill to contain illegally contaminated
dirt. Though unsuccessful, those protests sparked the Environmental Justice Movement, a
movement dedicated to reducing the exposure of poor and minority populations to
environmental hazards. Since then, only one Executive Order has been issued, no federal
legislation has passed, and judicial efforts have been mixed. Moreover, local
governments have little guidance or financial support to address environmental inequity.
I examined the spatial proximity of environmental hazards and goods to poor and
minority people in 18 U.S. cities at the Census-tract level. Using Geographic Information
Systems, I mapped the distribution of Toxics Release Inventory facilities and large-
quantity generators of hazardous waste. I also mapped the distribution of city parks and
community gardens. My general research question was: Are minorities and poor
populations over-represented in neighborhoods where environmental disamenities
cluster and under-represented in neighborhoods where environmental amenities cluster?
Using the GIS results, I investigated the city-level institutional factors that may
contribute to improved environmental equity. Guided by the Social Movements
Framework, I examined eight propositions to answer: What are the city-level political
processes and institutional variables that are more likely to be associated with better
disamenities and amenities distributions?
xvii
The GIS results suggested that in a majority of cities, the distributions of
disamenities and amenities are not random, and often, the hazards cluster in minority
and/or poor areas, while amenities only sometimes do. Regression analyses using the
traditional explanatory variables failed to explain adequately the distribution pattern in
most cities. As well, the institutional analysis failed to show a consistent combination of
political structures or processes that might explain environmental equity, with the
exception of the combination of a strong mayor-council government and formal
neighborhood organizations. Further analysis identified the path dependency of
environmental inequity as a policy problem. Case-study analysis revealed that a history
of strong mayor-council governments, formal neighborhood organizations that work
directly with city government, and vibrant citizen participation resulted in Boston being
the study’s most equitable city in both disamenities and amenities.

1
CHAPTER ONE: INTRODUCTION

Pollution itself never shows up on death certificates. With rare
exceptions, the diseases worsened by bad air are among the most
common afflictions in the modern world: heart disease, cancer, and
asthma.
Devra Davis, When Smoke Ran Like Water

In Victorian London, the prevailing theory of how diseases, particularly cholera,
were transmitted from person to person was the miasma theory or the idea that a poisoned
vapor or mist carried dirty particulates of decomposing matter into the human lungs,
where diseases would develop and spread. In 1854, however, Dr. John Snow began an
investigation of the Soho-area cholera outbreak that would change how science
understood disease transmission. In the case of cholera, Snow and others pinpointed the
culprit: a pump spewed water containing a bacterium that leached from cesspools, which
was then carried to the homes of the poor. At that time, sewage containment and disposal
was a pedestrian and ultimately hazardous proposition in London, Paris, and other
European cities. Raw sewage was dumped into streets, into backyards, or carried from
homes into open pits. A lucrative industry was built on the disposal of sewage by “night
soil men” and on the use of sewage in production including fertilizing fields and tanning
leather. Authorities were consumed with the sewage problem as central cities flourished,
but showed little concern for human health and safety in poor neighborhoods.
While Snow’s water-pump discovery helped defeat the miasma theory, his
triumph included the idea that hazardous waste could contribute to ill health, particularly

2
in the case of poor people. Snow’s greatest contribution, however, may have been a map
dubbed “The Ghost Map” (Johnson, 2006) in which Snow charted streets, water pumps,
and the incidents of the 1854 cholera outbreak. Snow unknowingly may have been the
first to map the proximity of low-income people to hazardous waste. In other words, the
humble doctor may have inadvertently been the first to study environmental equity.
Flash forward 116 years, and a body of research on pollution’s ill effects on
human beings is started by two researchers named Lave and Seskin, who showed that air
pollution contributed to “the amount of morbidity and mortality for specific diseases”
(1970, p. 723) like bronchitis and lung cancer, and argued that by reducing air pollution
by 50%, morbidity and mortality from lung cancer and bronchitis would drop 25% and
25-50%, respectively. Lave and Seskin, like Snow before them, prompted researchers to
take a closer look at how hazards in the air, land, and water—and the proximity of
humans to those hazards—might affect public health. Since then, over 100 studies have
examined the location (and sometimes the effects) of environmental hazards in relation to
minority and/or poor populations.
Mapping the spatial proximity of environmental hazards to poor and/or minority
populations in the United States is the central focus of my study. However, I also map the
spatial proximity of environmental amenities because the environmental justice literature,
with the exception of a few studies (c.f., Wolch, Wilson, & Fehrenbach, 2005; Talen &
Anselin, 1998; Tarrant & Cordel, 1999; Porter & Tarrant, 2001), has focused on hazards
or disamenities to the near exclusion of environmental goods or amenities. I begin to fill
that gap by using Geographic Information Systems (GIS) to map the distribution of
environmental disamenities and amenities at the Census-tract level in 18 cities. The

3
disamenities are Toxics Release Inventory (TRI) facilities and large-quantity waste
generators (LQGs). The amenities are city parks and community gardens. My general
research question is: Are minorities and poor populations over-represented in
neighborhoods where environmental disamenities cluster and under-represented in
neighborhoods where environmental amenities cluster?
Why are environmental amenities so important, particularly in light of the risks
associated with environmental disamenities? First, the benefits from environmental
amenities include: contributing to positive human health outcomes (Frumkin, 2001) and
healthier ecosystems (Shutkin, 2000); fostering more collective action or associational
interest among city residents (Sellers, 2002); and even promoting “interracial interaction”
that can lead to decreased negative attitudes among diverse groups (Shinew, Glover, &
Parry, 2004). Second, the problem of environmental justice is one of power and access,
and that includes the access and the right to control land, even without ownership
(Swanston, 1999). Amenities like parks and community gardens allow access to and
control of land by those lacking property rights. They also allow individuals to explore
“our capacity for ‘self-realization’” through “social relations needed to initiate,
implement, and manage” environmental tasks (Harvey, 1996, p. 199). Finally, as city
planners and policy makers steamroll toward sustainability and green urbanism (Platt,
2004; Hudnut, 2004), those policies must also address vulnerable populations.
The amenities I examine offer positive effects for nearby communities.
Community gardens can increase property values for nearby residents (Been & Voicu,
2007); create new and sustained relationships—and a sense of community—among
diverse neighbors (Hynes & Howe, 2004); promote social capital and civic

4
environmentalism (Glover, 2004; Staeheli & Mitchell, 2007; Bowles, 1999); offer
recreational and educational opportunities (Young, 2002; L. Lawson, 2005); increase
local food supplies and beneficial nutrition (Young, 2002; Alaimo, Packnett, Miles, &
Kruger, 2008); improve urban blight (The Trust for the Public Land, 2001; Wiland &
Bell, 2006); offer a neighborhood locale for social, political, and cultural events
(Saldivar-Tanaka & Krasny, 2004); and increase the mental and physical health of
participants (Armstrong, 2000; Stuart, 2005). Community gardens also provide an
important source of sovereignty for urban residents who may not hold property rights, but
can plant and maintain their gardens, creating environmental stability and attachment to
place. Similarly, city parks can: improve property values and urban aesthetics, attracting
visitors and business (The City Parks Forum, 2002; Sherer, 2006; Johnson & Neiman,
2004); enhance productivity and cognition of nearby residents (Kuo, 2001; Gies, 2008);
improve physical and mental health of park visitors (Sherer, 2006); improve urban air
quality (Nowak, 2008; Pincetl, Wolch, Wilson, & Longcore, 2003); facilitate social
interaction and promote a sense of community (Prow, 1999; Main, 2007; Jacobs, 1961);
and decrease crime (Sampson & Raudenbush, 1999; Wolch, Wilson, & Fehrenbach,
2005). Parks have also been shown to instill in urban residents a connection to place, as
well as a concern for “protecting nature within the city,” (Ryan, 2005, p. 40).
My objective in examining 18 cities is not to develop a general theory or establish
predictive factors of environmental equity. Rather, I hope to provide a comparison of and
general explanation about environmental equity that may be tested in studies of other
cities. I use the city comparisons to reveal patterns of environmental equity and inequity,
which in turn inform my analysis of the city institutional factors that may contribute to

5
more equitable outcomes and, hence, the objectives of the Environmental Justice
Movement (EJM). In other words, I define the nature of the environmental equity
problem, and then examine the possible institutional solutions to the problem, as
explanation is the “main task of the social sciences,” (Elster, 2007, p. 9). I investigate the
political opportunity structure and process in each city to identify the governance efforts
and reveal the “hidden structure behind the structure” (Ostrom, 2005, p. xii) that drives
environmental equity in diverse cities.
Cities hold a unique position in American democracy. Aristotle first noted the
importance city governance mechanisms in terms of balancing state authority with the
needs, wants, and authority of citizens. The Greek city was quite small compared to the
urban centers of modern-day America. Still, those urban centers, like the small Greek
city, represent a microcosm or “laboratory” (Jacobs, 1961, p. 6) of American government
and the possibilities therein. Yet, these “most consistent institutional features of
democracies” are paid “scant systemic attention” by researchers (Sellers & Lidström,
2007, p. 609), particularly environmental researchers (B. Lawson, 1995; Harvey, 1996;
Schweitzer & Stephenson, 2007). Cities have an increasingly important role in setting
taxes and deciding which goods and services will be available to local residents (Dreier,
Mollenkopf, & Swanstrom, 2004). Yet, the problems of cities are “sublimated in the
American political psyche,” (Waste, 1998, p. 23). The urban city is “a tent for many
functions” (Lynch, 1960, p. 91), including the potential to effect widespread
environmental change (Beatley, 2000). Cities are currently enjoying a resurgence of
authority over their environments through land-use decisions and zoning (Judd, 1997);
waste disposal decisions, brownfields redevelopment, and ground water protection

6
(Portney, 2003; May et al., 1996); “devolved collaboration” in resource management and
pollution control (Foster, 2002); and community-based environmental protection
(Nickelsburg, 1998). This resurgence of cities is also due to the urban ecology movement
or green urbanism, in which ad hoc efforts (e.g., rehabilitation of urban wetlands and
green spaces, construction of greenways, design of “green” buildings) are promoted by
city agencies and nonprofit organizations (Platt, 2004).
This resurgence is an interesting change because for decades, environmental
policy emphasized command-and-control regulation, which was made in a top-down
fashion, with Washington, D.C., as the locus of environmental decisions affecting state
and local governments. The traditional onus for the day-to-day monitoring and protecting
the long-term health and welfare of the environment rests with the Environmental
Protection Agency (EPA), though some consider that an “unenviable role” (Eady, 2003,
p. 169). Really, states and local governments have the “unenviable role” of implementing
federal policies with little EPA oversight (Feiock & Stream, 2001) or budgetary support
(May et al., 1996). The physical and political separation of policy makers and policy
implementers impedes real progress because of economic and scale constraints at the
local level and because local governments often have different priorities and limited
resources. Cities also struggle with federal regulations requiring them to deal with
pollution that emanates from outside of their political jurisdiction, creating
“transboundary issues” (Portney, 2003, p. 78) and cross-scale complications.
Unfortunately, many modern cities have not fulfilled their environmental promise
because of “poor environmental management, destructive and unregulated commercial
and industrial practices, rampant production and disposal, inadequate public planning,

7
and a failure of urban actors to work together,” (Inoguchi, Newman, & Paoletto, 1999, p.
2). Still, cities are places in which public administrators can work more closely with
citizens to pursue “economic, social, artistic, and political interests,” (Frederickson,
Johnson, & Wood, 2004, p. 1). I would add: environmental interests. The promise of
cities and local governance mechanisms in addressing environmental inequity has been
sorely neglected in the environmental justice literature (Sze, 2007; Judd, 1997; Gelobeter,
1994), in spite of multiple analyses having been done on some cities, like New York or
Los Angeles. Though we might know about the distribution of hazards in many American
cities, we do not know why one city might have a better equity distribution than the next.
By focusing on 18 cities and their local institutional mechanisms, I address that gap.

The Context of and Rationale for Environmental Equity Studies

In spite of the environmental justice efforts I describe below, the production of
man-made environmental hazards is increasing. The United States continues to generate
huge amounts of hazardous waste and industrial effluent. According to the EPA,
American industry generates about 7.6 billion tons of industrial waste annually and
households generate 245 million tons of waste (EPA, 2007). The same year, the TRI,
which reports on some 650 chemicals used in industry, logged 4,339,463,751 pounds of
toxic releases, and this figure does not include releases by smaller operators like dry
cleaners and gas stations. That amount of hazardous waste produced “has risen
approximately 40 per cent since 1987” (Fletcher, 2003, p. 40).

8
Adverse health effects of environmental hazards have been documented,
including: shorter stature for children born and living more than 75% of their lives near
Love Canal (Paigen, Goldman, Highland, Magnant, & Steegman, 1987); clusters of
bladder cancer in residents living near contaminated wells in eight Illinois counties
(Mallin, 1990); an increased risk of birth defects in children living within one mile of 590
wastes sites in New York (Geschwind et al., 1992); and a correlation between diabetes
and particulate matter (O’Neill et al., 2005). Furthermore, approximately 85,000
chemicals are used in manufacturing in the United States, yet studies have only
investigated about 1,500 (Davis & Webster, 2002). More troubling is the fact that
“complete environmental health surveillance systems are in place for only a few of the
thousands of potentially hazardous agents,” (Cromely & McLafferty, 2002, p. 159).
As well, the 2000 Decennial Census showed that minority populations are
increasing in the United States. The percentage of minority people in each one of 50
states increased over the past 20 years, and “the number of states with 30 percent or
higher percentage of minority population doubled, from just 8 states (including the
District of Columbia) in 1980 to 17 states in 2000” (Hobbs & Stoops, 2002, p. 98). The
growth in the Hispanic population is illustrative: “In every state except Hawaii, the
percentage of the population that was Hispanic increased during the 20-year period from
1980-2000,” (Hobbs & Stoops, 2002, p. 96). Minority populations in central cities have
also grown (Pastor, 2007). From 1990-2000, the Latino population in 81 major
metropolitan areas grew 42.9%, while the Asian population grew 39.4%. Interestingly,
the growth for white Americans was a -7.7%, which Pastor deems “white flight” to rural
suburbia. This is accompanied by a rise in poverty among minorities in central cities

9
(Bishaw, 2005), though it is hard to gauge because “it is often of a different sort rather
than being a poverty of joblessness and disconnection from the labor market, it is a
working poverty in which the hours of labor are long yet the pay is low,” (Pastor, 2007,
p. 74).
The growth of industrial waste or minority populations and urban poverty do not
automatically translate into increased environmental inequity, although waste is often
generated in metropolitan areas where minority populations cluster. Still, with no specific
equity legislation, limited federal oversight, a continued focus on economic development
at the local level, and diminished potential for legal action, research and an examination
of efforts to combat environmental inequity must continue. And the role that cities play in
managing environmental hazards must be examined, not only because of the deleterious
health effects, but because opinion studies reveal that the local pollution issues are still
important to the general public (Konisky, Milyo, & Richardson, 2008).
I describe my study below, but first, I review the history of the environmental
justice movement.

Environmental Justice Efforts

Environmental justice focuses on equity, racism, and classism in environmental
decisions and outcomes. Specifically, it deals with the cultural mores, rules, and policies
(Bryant, 1995) that promote environmental equity and citizen participation in
environmental decisions. Bullard writes: “Environmental justice embraces the principle
that all people and communities are entitled to equal protection of environmental and

10
public health laws and regulations,” (1996, p. 493, emphasis original). Thus,
environmental justice is one aspect of the broader goal of social justice, in that the
differentiation in the social relation of individuals to their environment belies broader
social inequity (Figueroa, 1995). Other definitions of environmental justice name specific
hazards like risks from air, water, noise, and visual pollution, as well as access to
environmental amenities (c.f., Shrader-Frechette, 2002; Todd & Zografos, 2005).
Although these definitions are more balanced and detailed, they are not as widely cited as
Bullard’s standard. Initially, researchers spoke of “environmental racism,” which
Benjamin Chavis is credited with coining as early as 1982 (c.f., Cutter, 1995; Gauna,
1995), though Bullard’s definition is often cited: “Environmental racism refers to any
policy, practice, or directive that differentially affects or disadvantages (whether
intended or unintended) individuals, groups, or communities based on race or color,”
(1996, p. 497, emphasis original).
1
Some researchers and activists still take this hard-line
approach, but others argue against it: “The term equity, or perhaps more properly
inequity, carries considerably less emotional baggage than racism,” (Rhodes, 2003, p. 16,
emphasis original; see also Liu, 2001; Most, Sengupta, & Burgener, 2004).
I use the term “environmental equity” to refer to the distributional or outcome
equity mapped in this study. Equity is a neutral term that allows researchers and policy
makers to focus on solutions in a more objective way—or, to achieve more “policy
relevance” (Pellow, 2000)—without impugning difficult-to-prove motives. A focus on
equity also allows the researcher to consider income, gender, citizenship status, and other
sociodemographic characteristics (Phillips & Sexton, 1999; Cutter, 2006a). Thus,

1
This is toned-down from Bullard’s earlier definition: “Environmental racism defends, protects, and
enhances quality-of-life choices available to whites at the expense of blacks,” (1993a, p. 30).

11
“environmental justice” best describes the overall efforts to reduce inequity and not the
distributional outcome, though some disagree (c.f., Hamilton & Viscusi, 1999;
Scholosberg, 2007; Foreman, 1998). How did the issue of environmental justice evolve?
In the United States, the problem of vulnerable populations living near hazardous
waste came to light in the mid to late 19
th
century when Jane Addams’ worked to
improve the living, sanitary, and public health conditions in poor neighborhoods (Platt,
2005). The Hoover Administration also addressed environmental equity when it sought to
establish housing standards for poor populations living near garbage incinerators (René,
Daniels, & Martin, 2000). Research efforts to understand these problems began in the
1970s (c.f., Freeman, 1972; Asch & Seneca, 1978).
Those precursors notwithstanding, the flashpoint or birth of the broader
environmental justice movement (EJM) is widely considered to be a 1982 incident, which
took place in Warren County, NC, (c.f., Bullard, 1994; McGurtry, 1997; Cole & Foster,
2001).
2
In 1978, a truck carrying liquid polychlorinated biphenyls (PCBs) emptied its
load under the cover of darkness along a 240-mile stretch of dirt road in Warren County,
a predominantly African American area. This was repeated over two weeks. Once
alerted, the state ordered that nearly 40,000 cubic yards of contaminated dirt be contained
within a new landfill. The area’s African American residents fought the landfill, but in
1982, construction began, followed by six weeks of heated protests and arrests, which
“marked the first time African Americans had mobilized a national, broad-based group to
oppose what they defined as environmental racism,” (Bullard, 1994, pp. 5-6).

2
Some argue that Love Canal was the impetus (Fletcher, 2003). Others cite incidents that occurred after
Warren County, like the Great Louisiana Toxics March in 1989 (Allen, 2003). Kay (1994) argues that
efforts in South Central Los Angeles to fight the construction of an incinerator incited the EJM.

12
As important as Warren County was for the EJM, research and academia also
played a critical role in highlighting environmental inequity and in nurturing the
environmental justice movement, perhaps more so than in “any other broad-based social
movement in the United States,” (Cole & Foster, 2001, p. 24). Following the Warren
County protests, and partly in response to a 1990 University of Michigan conference
entitled, “Conference on Race and the Incidence of Environmental Hazards,” the EPA
established its Environmental Equity Workgroup in 1990. The workgroup’s goal was
primarily to assess the extent and health effects of environmental inequity. In 1991, the
First National People of Color Environmental Leadership Summit spotlighted
environmental equity and also helped establish a network of concerned people and
organizations (Faber & McCarthy, 2001). Following that, EPA support for environmental
equity increased and it created the Office of Environmental Equity
3
in 1992. It also
published Environmental Equity: Reducing Risk for All Communities, featuring a cover
picture of three children playing next to a refinery and calling for more research,
reporting, and risk analysis (EPA, 1992).
What is the environmental justice movement? The best description of the
movement comes from Ferris (1994), who argues that the movement emanates from
themes running through American history:

The environmental justice movement is the confluence of three of
America’s greatest challenges: the struggle against racism and poverty, the
effort to preserve and improve the environment, and the compelling need
to shift social institutions from class division and environmental depletion
to social unity and global sustainability. (p. 298)

3
The name was later changed to the Office of Environmental Justice.

13
This description includes key, but often overlooked, goal of the environmental justice
movement: to preserve the environment. This allows a focus on environmental resources
and amenities. Reducing or preventing the incidence of disamenities in poor and minority
neighborhoods is critical, but so is the goal of preserving or adding environmental
amenities in those neighborhoods. Native Americans also include environmental
amenities in their definition of environmental justice (Swanston, 1999). The movement’s
broad goal is to change the power differential in society, created by a capitalist and a
“state matters” (Dryzek, Downes, Hunold, & Schlosberg, 2003, p. vii) political and social
structure, which leaves vulnerable populations with few choices about or control over the
areas where they live and work. Excluding amenities from that struggle means that EJM
researchers and activists focus on worst-case (or least hazardous) choices for vulnerable
groups rather than best-case (or amenity rich) options.
How does the movement for environmental justice differ from the broader
Environmental Movement? The latter is ecocentric (Shrader-Frechette, 2002), based on
the idea that nature/the environment is the foundation of all living organisms and must be
protected first and foremost. Activists are focused on the protection and preservation of
the natural systems, not the relationship between people and those systems, although
intergenerational resource management is addressed (c.f., Hiskes, 2006). This desire for
preservation is driven by post-materialist values that eschew more material goals for
aestheticism, self-actualization, and quality of life (Donati, 1997). For post-industrial
societies, “environmentalism becomes a metaphor of the need to care for the collective
goods that govern social life,” (Donati, 1997, p. 169). Even the recent movements for
sustainability and alternative energy reflect postmaterialism and ecocentrism. More

14
important, the mainstream Environmental Movement does not differentiate race and class
in its objectives. The environmental justice movement hones in on the relationship
between vulnerable populations and the environment, challenging universal goals of the
broader movement for the particularistic goals of outcome and process equity, and
integrates “the social needs of human populations,” (Pellow & Brulle, 2005, p. 3).
Recalling the Civil Rights Movement, The Native Land Rights Movement, The
Community Empowerment Movement, the Labor Movement, and others (Faber, 2008),
the EJM challenges the existing power structures in the neoliberal, market-driven society.
While the rationales for cleaner air, land, and water relate to human health and safety, the
environmental justice movement is rooted squarely in the conflict of the powerless
against state-driven environmental regimes and neocorporatist factions of society (c.f.,
Bowen & Wells, 2002; Gelobeter, 1994).
Another difference between the EJM and the broader Environmental Movement is
the legislative response. A host of environmental policies have addressed the broader
concerns about environmental degradation, starting with the Air Quality Act in 1960, and
the formal legitimization of the movement with the 1969 passage of the National
Environmental Policy Act (NEPA), which established the EPA. The ensuing legislation
addressed toxics in air, water, and land and provided reporting requirements for
producers of hazards.
4
Although these acts addressed environmental hazards and

4
For example, the Resource Conservation and Recovery Act of 1976 (RCRA) and the Solid Waste
Disposal Act which it amended in its entirety gave the EPA the ability to monitor the production,
transportation, treatment, storage, or disposal of hazardous waste. Title III of the Superfund Amendments
and Reauthorization Act of 1986 (SARA) contained the Emergency Planning and Community Right-to-
Know Act of 1986, which authorizes the reporting of toxic releases under the Toxics Release Inventory.

15
generated data to inform environmental justice research, the acts did not address
environmental justice directly.
The real impetus for federal-level attention to environmental justice and a crucial
legitimizing event of the EJM was Executive Order (EO) 12898, signed by President
William Jefferson Clinton in 1994. The order was signed in spite of limited evidence and
fledgling accounts of inequity (Cutter, 2006b), but signaled that environmental justice
was a problem of major concern. Entitled “The Federal Actions to Address
Environmental Justice in Minority Populations and Low-Income Populations,” the order
called on federal agencies to incorporate environmental justice into their missions:

To the greatest extent practicable and permitted by law and consistent with
the principles set forth in the report on the National Performance Review,
each Federal agency shall make achieving environmental justice part of its
mission by identifying and addressing, as appropriate, disproportionately
high and adverse human health or environmental effects of its programs,
policies, and activities on minority populations and low income
populations in the United States and its territories and possessions, the
District of Columbia, the Commonwealth of Puerto Rico, and the
Commonwealth of the Marian islands. (EPA, 1994)

The order created the Interagency Working Group on Environmental Justice, comprised
of the heads of 18 executive agencies, to act as an information clearinghouse and to
coordinate research on “adverse human health or environmental effects,” (EPA, 1994).
Following that, the EPA created the Office of Environmental Justice to ensure that
environmental justice concerns permeate the EPA’s operations at the federal, regional,
state, and local levels. While EO 12898 marked the federal recognition of environmental
inequity, the order offered no real implementation strategy.

16
The EPA has been ineffectual in environmental equity in spite of EO 12898,
something even the EPA admits (Center for Policy Alternatives, 2007). What are the
reasons for this lack of progress? First, as important and symbolic as EO 12898 is, it does
little to encourage policymakers to address environmental equity and it provides no real
guidance for specific equity regulations, relying instead on existing toxics policies that
lack equity components. EO 12898 encourages the incorporation of environmental
justice into other policy areas, but does not explicitly mandate it. This legal ambiguity
creates problems for policymakers and agencies (Rechtschaffen & Gauna, 2002). Second,
policymakers do not know enough about the extent or impact of the spatial proximity of
minority and/or poor populations to environmental hazards (c.f., Ringquist & Clark,
1999) and are constrained by the lack of clear and irrefutable information (Bowen, 2001).
Third, the mainstream media has not promoted environmental equity like it has other
high-profile environmental issues like greenhouse gases or species protection (Shrader-
Frechette, 2002). Finally, environmental inequity cannot be solved by focusing solely on
the siting of hazards. The causes of environmental inequity are more systemic, having
roots in housing, lending, transportation, and social policy.

Congressional Efforts

In spite of the concern over toxics, congressional efforts specific to environmental
equity have been disappointing. During the 103
rd
Congress, the Environmental Justice
Act (HR 2105) was introduced by two southern lawmakers, including then-Senator Al
Gore. HR 2105 would have required the EPA to identify and monitor 100 places

17
nationwide with high levels of environmental pollution and enact restrictions on the
future establishment of pollution-producing industries. That same session, HR 1924, the
Environmental Equal Rights Act, was introduced and would have prevented new TSDFs
in minority or poor communities where they already existed. This was followed by HR
1925, the Environmental Health Equity Information Act of 1993, which attempted to
amend Comprehensive Environmental Response, Compensation, and Liability Act
(CERCLA) to require that race, ethnic, income and other demographic data be reported
for people living next to toxic sites. These bills were introduced and sent to committee,
but never enacted. “The ‘heyday’ for environmental justice in Congress,” (Ringquist &
Clark, 1999, p. 80) followed in 1993-1994 when six equity bills were introduced, though
all failed. That “heyday” was followed by three years of inaction; then introductions of
more environmental justice bills began anew, including four bills introduced in the 110
th

Congress. In all, the Congressional Record lists 23 environmental justice bills introduced
since 1992. None have made it out of committee, including two (HR 6396 and HR1602)
that targeted the areas affected by Hurricanes Katrina and Rita. One might argue that
environmental equity is not an issue that the recent Republican congress or Bush
administration would support. Similarly, since September 11, 2001, law-making related
to domestic terrorism and the wars in Afghanistan and Iraq have occupied legislative
dockets. Still, a record of 0 for 23 is dismal for any issue, and it remains to be seen how
the Obama Administration or Democrat-controlled Congress will tackle the issue.

18
Judicial Efforts

Like the legislative and executive branches of government, the judicial effort has
yielded little systemic change. Some local efforts to fight the specific location of
hazardous sites have been successful. For example, in Cox v. the City of Dallas (Civil
Action No. 3:98-CV-1763-BH), the court ruled that a mine in an African American
neighborhood had violated regulations when it tried to operate as a landfill. Several suits,
however, have not been successful (c.f., Sierra Club v. Gates, No. 2:07-cv-0101-LJM-
WGH; and Communities Against Runway Expansion, Inc. v. Federal Aviation
Administration, 355 F3d 678). Groups have sought legal redress to hold local
governments accountable for alleged environmental racism, though the efforts have
proven difficult (Checker, 2005; Hoidal, 2003). Recently, some have touted “toxic torts,”
which use civil and criminal law to punish polluters, as an option for EJM activists
(Koenig & Rustad, 2004). Unfortunately, any future judicial efforts will be complicated
by the Supreme Court’s ruling in Alexander v. Sandoval, 99 U.S.1908 (2001), which
basically limits an individual’s right of action for racial discrimination unless the
individual can prove intentional racism (Faber, 2007).

State and Local Efforts

This lack of federal success and the efforts of the Bush Administration to
decentralize environmental protection have put the onus on state and local governments
for environmental justice with little federal budgetary or implementation support. Some

19
states have “struck out on their own,” (Ringquist, 2004) and enacted environmental
equity legislation (e.g., California, Oregon, Connecticut), while some have equity
resolutions or strategic plans (e.g., Indiana, Virginia), and some have established
taskforces to address equity (e.g., Texas). Still, many states, like Michigan, Ohio, and
West Virginia have no equity-specific laws (American Bar Association, 2004).
Lack of state programs shifts the burden to local governments. While local
involvement in environmental management is necessary and attractive, leaving
environmental decision-making to local governments can also exacerbate local power
struggles, decrease representation of vulnerable populations, result in haphazard policy
solutions to trans-boundary issues that need to be addressed on a more regional or
national scale (c.f., Foster, 2002; Nickelsburg, 1998). The lack of participation in federal
decisions about environmental equity management creates a disconnect between federal
goals and local efforts (Field, 1998; DeWitt, 2004). Cities often are under severe
budgetary constraints and courting pro-growth partnerships, thus they often do not have
the fiscal resources to devote to environmental equity (Portney, 2003). Still, some cities
manage to do so. For example, San Diego sponsors the Lead Safe Neighborhoods
Program, a lead awareness, training, compliance, and abatement program for low-income
neighborhoods. Local efforts, however, are often ad hoc with limited effectiveness.

Nonprofit Efforts

Many small nonprofit groups combat localized incidents of environmental justice
and it is at the local level where many of the EJM’s “clearest victories” have occurred

20
like closing landfills, preventing the expansion or siting of new hazards, and assisting in
relocating residents from hazardous neighborhoods (Pellow & Brulle, 2005). One
example is the Chelsea Creek Community Based Comparative Risk Assessment
organization that monitors pollution and its effects in Chelsea and East Boston, low-
income and diverse (including a large immigrant contingent) areas, and recommends
solutions to government officials. These nonprofits are locally based, activist groups that
usually start as grassroots efforts to combat single issues (Lichterman, 1996). Unlike the
more well-known and well-funded mainstream national environmental organizations
(e.g., Natural Resources Defense Council, Sierra Club), the smaller activist organizations
or social movement organizations are small and often struggle for financial and member
support, thus their results of these efforts have been mixed.
Government and nonprofit efforts to address environmental equity have been
mixed at best. It has been 27 years since the Warren County incident, but progress in
environmental justice has been disappointing.
In the next section, I provide an overview of my study and hypotheses.

General Suppositions and Questions

This study proceeds in two phases. In Phase I, I examine the distribution of
environmental disamenities and amenities in poor and minority neighborhoods by
mapping the distribution of TRI facilities and LQGs, as well as city parks and community
gardens at the Census-tract level. The Census variables I use are consistent with the
environmental equity literature and are discussed in Chapter 3. The conceptual goals of

21
Phase I are to examine the spatial distribution of hazards and conduct proximity analysis
to address the following justice-based hypotheses: (1) The greater the percentage of poor
and/or minority residents in a tract, the greater the number of disamenities and the more
unequal the distribution; and (2) The greater the percentage of poor and/or minority
residents in a tract, the fewer the number of amenities and the more unequal the
distribution. To address these questions, I answer the following questions (adapted from
Mitchell, 2005):

• What is the distribution of the disamenities and amenities?
• Do disamenities or amenities cluster?
• Where do disamenities or amenities cluster?
• Is a sociodemographic pattern evident in the distribution of disamenities and
amenities?

Phase II of the study turns to the political institutional factors that might influence
the distribution of amenities and disamenities. Phase II is designed to answer the
question: What are the city-level political processes and institutional variables that are
more likely to be associated with better disamenities and amenities distributions?
To answer this question, I use the political process approach of the Social
Movements Framework and offer eight propositions, investigating the opportunity
structure of each city. I focus on factors such as the form of government, election
procedures, which agencies or departments are responsible for environmental justice, if

22
any regulations or executive orders specifically addressing environmental justice exist,
and the city’s political majority.

Contributions of this Study

The first contribution of my study is the mapping two disamenities nationally in
18 carefully selected cities that represent geographic, population, government, and
economic diversity at the Census-tract level. Though some earlier studies of
environmental equity examined cities nationwide, several used zip-code level data and
most did not use the advanced GIS mapping techniques and spatial statistics available
today. The second contribution I make is to map two environmental amenities. As stated
above, very few equity analyses focus on environmental amenities; yet, as important as
the concern that vulnerable populations face unequal risk burdens from hazards, those
same populations may face a dearth of the benefits that accrue from environmental
amenities. The third contribution I make is to focus on the city institutional factors that
may improve or impede environmental equity. The fourth contribution I make is an
attempt to integrate the Social Movements Framework in conjunction with GIS mapping
analysis. Finally, I begin to fill a gap in the Public Administration literature. While
several policy and urban affairs journals have published studies, not many public
administration journals have. My results may help influence the public administration
debate by contributing to the discourse on environmental equity and local governance
efforts within the field (Strong & Hobbs, 2002).

23
My research will be of interest to academics in the environmental policy area,
especially those who work in the area of environmental justice, environmental health, and
sustainability. Researchers studying city policymaking will be interested, as will city
officials who are responsible for environmental affairs. Those nonprofit organizations
that work in local environmental issues and justice advocacy also will be interested.

Organization of the Dissertation

This dissertation is organized into eight chapters. I begin with a thorough review
of the environmental equity literature in Chapter 2. In Chapter 3, I outline my research
design and methods for Phase I. Chapters 4 and 5 present the results of the GIS mapping
of environmental disamenities and amenities, respectively. In Chapter 6, I provide a brief
literature review of the city policymaking and Social Movements literatures, and outline
my research design and methods for Phase II. Chapter 7 presents the results of the
institutional analysis, and Chapter 8 provides a summary of the dissertation along with
implications for policymakers and suggestions for future research.

24
CHAPTER TWO: THE ENVIRONMENTAL JUSTICE LITERATURE

The legitimacy of the grievances advanced by the environmental
justice movement, the power of its claims on the political process,
and the justification for continued policy change to address
perceived environmental inequities all hinge on resolving the current
uncertainty regarding the existence of environmental inequities.
Evan Ringquist, Assessing Evidence of Environmental Inequities

Three seminal, but flawed, works published after the 1982 Warren County
incident coalesce as the foundation of the environmental justice literature. Bullard’s 1983
qualitative study of Houston’s solid waste facilities showed that a disproportionate
number of the facilities were sited in predominantly black (versus non-black)
neighborhoods. Among other problems, the study lacked control variables and did not
test for income effects. That same year, the General Accounting Office (1983) examined
four hazardous waste landfills in the Southeastern U.S. (using non-random sampling) and
described the demographic characteristics of the Census tracts around the sites. The study
found that three of four sites were located in or near predominantly black tracts (with
moderate income effects), but due to its design, limited sample, and lack of control
variables, the results were not generalizable. Many people are familiar with the 1987
UCC study Toxic Wastes and Race in the United States, which described the proximity of
TSDFs to U.S. minority populations at the zip-code level. The UCC found race effects
and argued that race was the best predictor of the hazard location. Many criticized these
studies for their qualitative and case-study methods (c.f., Kamieniecki & Steckenrider,
1997) and weak designs (c.f., Bowen, 2001). Yet, these works captured people’s attention

25
and ignited researchers, and a flurry of studies followed. Over 100 studies have been
published since, and close to 85% of those studies support the environmental equity
hypothesis (i.e., that environmental disamenities are over-represented in minority and/or
poor neighborhoods). One might think this constitutes a fairly cohesive body of evidence.
Upon closer examination, however, the evidence is neither clear nor convincing.

A Brief Overview of 25 Years

Researchers began examining the distribution of air pollution in the 1970s (c.f.,
Lave & Seskin, 1970; Freeman, 1972; Asch & Seneca, 1978), but environmental equity
as a field began to take shape in the late 1980s. Close to 85% of the 100+ studies I
reviewed support the environmental equity hypothesis. For example, a majority of TRI
facility and/or emissions studies have shown income and/or race effects (Burke, 1993;
Glickman, 1994; Glickman & Hersh, 1995; Bowen, Salling, Haynes, & Cyran, 1995;
Perlin, Setzer, Creason, & Sexton, 1995; Pollack & Vittas, 1995; Kreisel, Centner, &
Keeler, 1996; Cutter & Solecki, 1996; Ringquist, 1997; Brooks & Sethi, 1997;
Chakraborty & Armstrong, 1997; Neumann, Forman, & Rothlein, 1997; Downey, 1998;
Perlin, Sexton, & Wong, 1999; Morello-Frosch, Pastor, & Sadd, 2001; Morello-Frosch,
Pastor, & Sadd, 2002; Dolinoy & Miranda, 2004; Pastor, Sadd, & Morello-Frosch,
2004a; Pastor, Sadd, & Morello-Frosch, 2004b; Mennis & Jordan, 2005; Pastor, Morello-
Frosch, & Sadd, 2005; Pastor, Morello-Frosch, & Sadd, 2006; Morello-Frosch & Jesdale,
2006; Sicotte & Swanson, 2007; Maantay, 2007). A few studies found no race or income
effects (Hersh, 1995; Krieg, 2005), while other studies were mixed (Mitchell, Thomas, &

26
Cutter, 1999; Tiefenbacher & Hagelman, 1999; Ash & Fetter, 2004; Holmes, Slade, &
Cowart, 2000; Haynes, Lall, & Trice, 2001; Boone, 2002). Several studies showed race
effects and found income to be moderate (Shapiro, 2005); curvilinear (Sadd, Pastor, Boer,
& Snyder; 1999a; Daniels & Friedman, 1999); or mixed (Arora & Cason, 1999).
Some researchers examined TSDFs and found support for the environmental
equity hypothesis (Bullard, 1990; Hamilton, 1993; Adeola, 1994; Hamilton, 1995; Boer,
Pastor, Sadd, & Snyder, 1997; Davidson & Anderton, 2000; Pastor, Sadd, & Hipp, 2001;
Black & Stewart, 2003; Saha & Mohai, 2005). Other studies—some done in an effort to
test the seminal works in the field—showed weak, mixed, or no support (Anderton,
Anderson, Oakes, & Fraser, 1994; Anderton et al., 1994; Been, 1994; Oakes, Anderton,
& Anderson 1996; Markham & Rufa, 1997; Bowman & Crews-Meyers, 1997; Liu, 1997;
Been & Gupta, 1997; Atlas, 2002a, 2002b).
Studies using multiple dependent variables, including TRI facilities and
emissions, TSDFs, Superfund sites, air pollution, and other hazards present a different
picture. Table 2.1 highlights these studies and their results.
Recently, researchers have used more innovative dependent variables to test the
environmental equity hypothesis. Shaikh and Loomis (1999) examined siting decisions
for sources of criteria air pollutants
5
revealing race and income effects. Stretesky and
Lynch (1999) studied accidental chemical releases in one Florida county and found race
and income effects in bivariate analyses, but only income was significant in multivariate
analyses. Graham, Beaulieu, Sussman, Sadowitz, and Li (1999) showed that coke
6
plants

5
Criteria air pollutants are carbon dioxide, nitrogen dioxide, sulfur dioxide, ground-level ozone, lead, and
particulate matter (i.e., PM
10
and PM
2.5
), and have established National Ambient Air Quality Standards.
6
Coke is a byproduct of coal destruction and is used in smelting iron ore.

27
were located closer to low-income populations and Hispanic populations. Stretesky,
Johnston, and Arney (2003) showed that large-scale hog operations were placed and/or
expanded in predominantly black communities (for other hog-operations studies, see
Taquino, Parisi, & Gill, 2002 and Wilson, Howell, Wing, & Sobsey, 2002).

Table 2.1
Environmental Equity Studies Using Multiple Dependent Variables
Dependent Variables Race
Effects
Income
Effects
Napton & Day, 1992 Five types of petrochemical sites No No
Cutter, Holm, & Clark, 1996 TSDFs, TRI & Superfund sites No Mixed
Greenberg & Cidon, 1997

Superfund sites, sewage processors, trash-to-
steam incinerators, recycling plants
Yes Yes
McMaster, Leitner, &
Sheppard, 1997
TRI sites/emissions, Superfund sites, LULUs Mixed Yes
Bowman & Crews-Meyer,
1997
TSDFs, waste generators, other LULUS No No
Hird & Reese, 1998 TSDFs, TRI sites, industrial air & water
emissions, air & water quality
Yes Mixed
Stretesky, 1998 TRI & Superfund sites, TSDFs Yes No
Sheppard, Leitner, McMaster,
& Tian, 1999
TRI, Superfund & Petrofund sites Yes Yes
Lester, Allen, & Hill, 2001 TRI sites, Superfund sites, air pollution,
hazardous & solid waste
Yes Mixed
Fricker & Hengartner, 2001 Incinerators, bus garages, landfills, TRI sites,
TSDFs, sewage plants
Mixed Mixed
Harner, Warner, Pierce, &
Huber, 2002
TSDFs, TRI sites, Superfund sites Yes Yes
Faber & Krieg, 2002 TSDFS, LULUs, Superfund sites Yes Yes
Stretesky & Lynch, 2002 TSDFs, TRI, & Superfund sites Yes Mixed
Bolin et al., 2002 Superfund & TRI sites, TSDFs, LQGs Yes Yes
Grineski, Bolin, & Boone,
2007
Six air pollutants Mixed Yes

Others have studied electric power plants (Touché & Rogers, 2005), illegal drug
markets and other neighborhood disamenities (Ford & Beveridge, 2004), alcohol outlets
(Romley, Cohen, Ringel, & Strum, 2007), and noise pollution (Chakraborty, Schweitzer,
& Forkenbrock, 1999; Sobotta, Cambell, & Owens, 2007) showing race and/or income
effects. Transportation-related environmental justice studies have shown race and/or

28
income effects (c.f., Kennedy, 2004; Haynes, 2004; Jacobson, Hengartner, & Louis,
2005; Schweitzer, 2006). Konisky (2009) examined 15 years of enforcement of the Clean
Air and Clean Water Acts and showed that counties with high concentrations of poor
and/or minority populations conducted fewer enforcement actions.
One might think these studies constitute a fairly cohesive body of evidence to
inform policy makers. Though recent studies are quite robust in their methods, upon
closer examination of the literature overall, the evidence is neither clear nor convincing.
In the first 15 years, researchers were conservative in the hazards they analyzed
(Gelobeter, 1994; Pinderhughes, 1997; Maantay, 2002), relying mainly on TRI sites and
TSDFs in an effort to create parsimonious models. Some early studies relied on simple
methods, and many suffered from aggregation bias. Researchers often tested only African
American populations or used “minority” versus “nonminority” distinctions (c.f., Lavalle
& Coyle, 1992; Hamilton, 1993; Liu, 1996; McMaster, Leitner, & Sheppard, 1997),
failing to examine race ethnicity thoroughly and failing to account for density variations
in minority populations (Kamieniecki & Steckenrider, 1997; Lester & Allen, 1999).
Similarly, the majority of studies focused on single hazards, namely TRI sites and/or
emissions. This may be due to the availability of TRI data, but as environmental equity
studies in transportation and urban planning have shown, non-point source pollutants are
a critical aspect of a hazardscape. Non-point source pollution emanates from moving
sources like trucks, railcars, automobiles, and heavy equipment; though hard to measure,
it merits further examination (Cromley & McLafferty, 2002). In spite of the recent
innovative dependent variables and advanced mapping techniques used, several
methodological problems continue to vex researchers.

29
Methodological Problems

The seminal works in the field were severely criticized for weak designs and
methods, as were many subsequent studies. One of the key problems in these early
studies was the Modifiable Areal Unit Problem (MAUP), which means that researchers
can obtain different results from the same data and universe by varying the unit of
analysis. This problem deeply affects the environmental equity literature because
“conclusions in geographic studies are highly sensitive to the scale and the zoning
scheme (areal boundaries) used in the analysis,” (Sui, 1999, p. 44). That is, counties v.
zip codes v. tracts v. block groups can yield different results from the same data. In their
study of TRI emissions, Bowen, Salling, Haynes, and Cyran (1995) noticed that statewide
race effects in Ohio disappeared in a tract-level analysis of Cuyahoga County. Some
researchers have specifically addressed this issue of scale dependency by testing counties,
tracts, and block groups in a single location. For example, Cutter, Holm, and Clark
(1996) found support for race
7
and income effects at the county level in South Carolina,
but at the tract and block-group levels, both effects disappeared. McMaster, Leitner, and
Sheppard (1997) also saw race effects disappear at the block-group level in their study of
TRI sites in the Twin Cities. In a study of hog farms, Taquino, Parisi, and Gill (2002)
found varying effects using zip codes, tracts, block groups, and communities. After
testing multiple units of analysis on TRI facilities, Sui (1999) warned, “It is possible to
find almost any desired results simply by reaggregating the data to different scales and

7
Interestingly, in their study, white, middle- and upper-income populations were closer to environmental
hazards than rural, minority, and low-income populations.

30
areal-unit boundaries,” (p. 50). Thus, a key question in environmental equity is: “How
big is your backyard, or what is the appropriate geographic scale,” (Sui, 1999, p. 52).
With larger geographic units of analysis, the population is not randomly or evenly
distributed across that broad unit. Therefore, even when an environmental hazard is
located within a zip code, for example, researchers cannot assume equal exposure of all
populations to those hazards. The alternative is to use a smaller, disaggregated unit (like a
Census tract or block group) to better represent “that the entire population of that unit has
been equally exposed to a given externality,” (Most, Sengupta, & Burgener, 2004, p.
579). Researchers who used large units of analysis like zip codes (c.f., United Church of
Christ, 1987; Hamilton, 1995; Oakes, Anderton, & Anderson, 1996; Ringquist, 1997;
Shaikh & Loomis, 1999) and counties (c.f., Hamilton, 1993; Hird, 1993; Bowen, Salling,
Haynes, & Cyran, 1995; Daniels & Friedman, 1999) decreased the significance and
power of their explanatory variables.
Smaller geographic units are preferred for equity studies (c.f., Anderton,
Anderson, Oakes, & Fraser, 1994; Anderton, 1996; Shrader-Frechette, 2002; Ash &
Fetter, 2004; Schweitzer & Stephenson, 2007). The Census tract is the most attractive
geographic area because: (a) using smaller units like block groups or blocks creates data
gaps due to confidentiality concerns (i.e., lack of income data at block levels); (b) hazard
effects are likely to be larger than the block or block-group level; (c) the geographic area
of blocks tends to vary widely; and (d) blocks and block groups change over time (Been,
1994). Some prefer block groups arguing that they offer a more accurate snapshot of a
neighborhood because they are based on relatively homogenous and small groups
(Baybeck & DeLorenzo, 2001). However, such a small unit may misrepresent effects

31
when considering the larger population. A Census block is an even smaller unit, but
income data are not recorded at the block level. Some contend that administrative
boundaries like Census tracts do not represent neighborhoods or communities very well.
Communities are an attractive unit of analysis because the community “holds both legal
and social authority to raise concern about environmentally controversial facilities,”
(Taquino, Parisi, & Gill, 2002, p. 298). However, researchers have yet to settle on a
definition of community that can theoretically and statistically represent a neighborhood
because a superior delineation of community is elusive (Ringquist, 2005).
The focus on MAUP sometimes distracts from the more important geographical
phenomenon of spatial autocorrelation: “Spatial data is by its very nature not randomly
distributed, as traditional statistical approaches require,” (Maantay, 2007, p. 33). When
time and resources permit, multi-scale analysis is an excellent option to understand the
differences “across scales,” (Mennis, 2002; see also Rhodes, 2003). I chose tracts because
they are the standard unit of analysis and avoid the data problems of smaller Census
units, and because multiple levels of analysis in an 18-city study would prove unwieldy.

The Unit-Hazard Coincidence Problem

Many early equity studies (c.f., Bullard, 1990; Napton & Day, 1992; Zimmerman,
1993; Adeola, 1994) focused on the presence or absence of hazards in a geographically
defined place. The unit-hazard coincidence (UHC) method identifies environmental
hazards (e.g., TRI facilities) within defined geographic units (e.g., Census tracts). The
racial and income characteristics of those units are then defined and hazard-containing

32
units are compared with non-hazard units, thereby identifying the spatial coincidence of
the hazard with minority and/or poor populations. While informative, this method failed
to identify the populations closest to the hazard and thus, their potential exposure. That is,
the UHC method served as a weak proxy for risk and failed to establish a more precise
portrait of the affected population. Anderton (1996) and Glickman and Hersh (1995)
identified this problem early, and many continue to admonish those using this method
(c.f., Bevc, Marshall, & Picou, 2007; Mohai & Saha, 2006; Ash & Fetter, 2004;
Davidson, 2003; Foreman, 1998; Cutter, Holm, & Clark, 1996).
With the help of advanced mapping and geoprocessing techniques in GIS,
researchers have developed better proximity methods for identifying the populations
living closest to environmental hazards. In a path-breaking study of chronic and acute
airborne releases in Allegheny County, PA, Glickman (1994) revealed the potential of
GIS. Using GIS to construct circular buffers of .5, 1, and 2 miles around polluting
facilities, he compared the demographics of the residents within the buffers to the
demographics of those not in the buffers. His results showed neither race nor income
effects. In a given GIS, buffers can be drawn around: points, such as hazardous facilities;
lines, such as highways or railroads; or polygons, such as Superfund sites (Maantay &
Ziegler, 2006). Circular buffers are the most common (c.f., Heitgard, Burg, & Strictland,
1995; Neumann, Forman, & Rothlein, 1998; Perlin, Sexton, & Wong, 1999; Liu, 2001;
Atlas, 2002a, 2002b), but plume buffers can also be used to map air pollution (c.f.,
Glickman & Hersh, 1995; Davidson, 2003; Maantay, 2007). Chakraborty and Armstrong
(1997) demonstrated that plume buffers resulted in a more accurate analysis of the
populations around TRI facilities. By using buffer analysis, researchers can also use

33
techniques like aerial weighting to interpolate the populations within the buffer: “This
implies that for spatial units partially contained within the buffer zone, a fraction of the
total counts is used, based on the proportion of the unit’s area that is contained within the
buffer,” (Chakraborty, Schweitzer, & Forkenbrock, 1999, p. 243). They argue that this
helps to remedy the assumption of an evenly distributed population within the geographic
unit. After all, a GIS map is a social construction (Maantay, 2002, 2007).
Researchers are still refining mapping techniques. Dasymetric mapping “refers to
a process of disaggregating spatial data to a finer unit of analysis, using additional (or
‘ancillary’) data to help refine locations of population,” (Maantay, Maroko, & Herrmann,
2007, p. 77). Dasymetric mapping accounts for land or administrative factors that might
affect population distribution like tax or cadastral data. Land characteristics like water
bodies or land-use boundaries can also be used. Mennis (2002) used dasymetric mapping
to demonstrate race and income effects in his study of TSDFs and TRI facilities in five
southeast Pennsylvania counties. Others prefer adjacency analysis, which goes beyond
the administrative boundary to define the contours of the entire affected area (c.f., Most,
Sengupta, & Burgener, 2004), but Liu (2001) argues that it inhibits comparisons because
the contours of the affected areas cannot be controlled and are mostly irregular.
Refined methods are also needed because pollution is dynamic; air pollution has
varying trajectories, but so do groundwater and soil contamination. With no fixed
boundary, pollution can only be wholly contained in a hermetically sealed container. To
accommodate this, Milman (2006) used pollution concentration levels to identify the
populations affected by emissions from three power plants and compared those results
with proximity-analysis results. By accounting for the dispersion of emissions, weather

34
patterns, and emission rates, she captured greater numbers of minorities and poor people
by mapping dispersion than by simply drawing circular buffers around plants. However,
she noted the difficulty in studying large numbers of facilities when using this kind of
detailed analysis. Another difficulty lies in mapping “the changing geography of
hazards,” (Cromley & McLafferty, 2002, p. 167). Industrial processes “involve assembly
of raw materials from many sources and distribution of finished products and waste
materials to many locations,” and thus, “hazard geography is not static,” (Cromley &
McLafferty, 2002, p. 167). In spite of the need for new techniques, most environmental
equity researchers agree that proximity methods, like those used in this study, are
superior to UHC methods (c.f., Mohai & Saha, 2006; Maantay, 2002) and that GIS has
significantly improved researchers’ ability to understand the hazardscape.

The Time-order Relationship Problem

The environmental equity literature boasts few longitudinal studies, limiting the
understanding of the time-order dynamics of hazard proximity (Helfand & Peyton, 1999;
Mitchell, Thomas & Cutter, 1999; Szazs & Meuser, 2000; Krieg, 2005) and limiting
predictive validity (Bowen, 1999). Some argue that determining a causal order of
minorities and hazards based on cross-sectional, quantitative methodologies alone is
reductionist, narrowing racism to “a discrete and hostile act,” (Pulido, 2000, p. 17). Noted
early on (c.f., Clark, Lab, & Stoddard, 1995; Pulido, Sidawi, & Vos, 1996), time-order
inconsistencies abound, partly due to the short timeframe of environmental data and
partly due to the prevalence of cross-sectional research designs. As a result, no one can

35
say for sure if the hazard or the minority and poor populations came to the neighborhood
first. Indeed, the competing hypothesis in environmental equity contends that minority
and low-income populations moved to already hazard-prone areas. This hypothesis was
first tested in Been’s challenge to the 1983 GAO and Bullard studies. She studied LULUs
in Alabama, South Carolina, North Carolina and Houston, TX, hypothesizing that market
dynamics resulted in siting inequities: “Such factors as poverty, housing discrimination,
and the location of jobs, transportation, and other public services may have led the poor
and racial minorities to ‘come to the nuisance’…because those neighborhoods offered the
cheapest available housing,” (Been, 1994, p. 1385). Her results showed that black and
poor populations were not disproportionately large at the time of the incinerator and
landfill sitings, even though they later were. A later study of 544 TSDFs showed similar
results for African Americans, but also revealed that poor Hispanics were present at siting
in the Southwestern U.S. (Been & Gupta, 1997). Alternative explanations for the
proximity of vulnerable populations to hazards include: “postwar social arrangements and
power relations” (Hurley, 1995, p. 180; see also Hurley, 1997); white flight (Hersh,
1995); neighborhood change due to “a mix of social, economic, institutional, and
environmental dynamics” (Liu, 1997, p. 655); zoning decisions dating back to the 1920s
(Boone & Modarres, 1999); normal socioeconomic processes, including “economic
boosterism, unregulated development, and racial and ethnic differences in education,
occupation, and income,” (Szazs & Meuser, 2000, p. 602); and residential segregation
(Boone, 2002).
In two studies, however, longitudinal results were not so quick to dismiss racism.
Examining the siting of high-capacity TSDFs between 1970 and 1990 in Los Angeles,

36
Pastor, Sadd, and Hipp (2001) provided a rigorous analysis of minority “move-in.” Using
bivariate analysis and logit regression to measure the demographic changes in tracts, their
results ran contrary to the idea that minorities moved to the locale after the TSDF was
sited: “Controlling for other factors, minorities attract TSDFs but TSDFs do not generally
attract minorities,” (2001, p. 18). Heeding the call for “historical geography” (Pulido,
1996b, 2000), Bolin, Grineski, and Collins (2005) examined the history of commercial
and industrial zones in South Phoenix and concluded that the industrialization and
accompanying transportation corridors were placed in minority communities, but not
simply because of racist intent: “acts of omission, such as failing to provide urban
infrastructure or not enforcing housing codes, have been as important in the development
of environmental inequalities in Phoenix as have been acts of commission,” (p. 166).
Historical studies can capture migration or “the process that results in the permanent or
semi-permanent relocation of the home,” (Cromley & McLafferty, 2002, p. 3). Historical
investigations are costly and time-prohibitive, but their importance cannot be understated.
Still, as Harvey (2000) notes, geographies “are perpetually being reproduced, sustained,
undermined, and reconfigured by political-economic and socio-ecological processes
occurring in the present,” (p. 78). Thus, cross-sectional studies like this one are critical.

Measuring and Assessing Risk

Most environmental equity studies use spatial proximity as a proxy for risk. Few
researchers attempt to assign risk though many acknowledge this is a serious problem
that threatens the face validity of most research designs (Bowen, 1999; Bowen & Wells,

37
2002). Researchers have yet to identify a methodological solution. The quantitative and
qualitative methodologies for assessing risk probabilities can be marred by value-laden
cognitive processes. Slovic argues, “human beings have invented the concept of risk to
help them understand and cope with the dangers and uncertainties of life,” (1997, p. 280,
emphasis original). Sociologists suggest that risk is rarely evaluated rationally (Clarke,
1989; Stallings, 1991). Still, policy makers need to understand risk because it “could have
negative effects on economic development possibilities in poorer, minority
neighborhoods,” (Sadd, Pastor, Boer, & Snyder, 1999b, p. 136; see also Bowen, 2001).
What methods measure risk? Scientific research in the form of animal studies,
cellular research, epidemiological studies, and computer simulations can help, but few
such studies exist in the environmental equity literature. They are difficult and expensive
because to identify risk probabilities, researchers have to understand not only the toxic
chemicals involved, but also the exposure routes, the concentrations of those hazards,
their toxicity levels in different media like air or water, and the duration of time they exist
in those media (Rhodes, 2003; Davidson, 2003; Cutter, Scott, & Hill, 2002; Bowen &
Wells, 2002; Bowen, Salling, Kingsley, & Cyran, 1995; Hamilton, 1999).
Some researchers have incorporated risk measurements in their equity studies
(c.f., Neumann, Forman, & Rothlein, 1997). One measure is the EPAs Risk Screening
Environmental Indicators (RSEI) model (c.f., Shapiro, 2005; Sicotte & Swanson, 2007).
RSEI is a database that combines an air dispersion model and hazard exposure estimates
to determine how likely it is that air pollutants will reach nearby residents based on
volume, toxicity, decay and solubility, and dose estimates (Sicotte & Swanson, 2007).
Others have incorporated cancer risk estimates in their studies (c.f., Morello-Frosch,

38
Pastor, & Sadd, 2001; Morello-Frosch & Jesdale, 2006; Pastor, Morello-Frosch, & Sadd,
2005). These studies represent solid attempts to address risk measurement problems,
though much work remains to be done in this area.

Gaps in the Environmental Equity Literature

The amenities gap is a serious one in the literature. Only a few researchers have
examined environmental “goods,” in spite of the benefits that accrue to populations living
and working near amenities (c.f., Been & Voicu, 2007; Hynes & Howe, 2004; Young,
2002; Kuo, 2001; Nowak, 2008). Talen and Anselin (1998) studied the distribution of
public playgrounds, although their results did not show any inequity. Tarrant and Cordel
(1999) mapped recreation sites in the Chattahoochee National Forest and all Census
block groups within one mile of those sites. Their analysis showed income effects, but
they framed the study in terms of those recreation sites having environmental and social
costs and thus, being undesirable land uses. In a limited survey of New Jersey residents,
Greenberg and Renne (2005) focused on neighborhood “walkability,” a very interesting
idea, but a poorly done study. Morland and Wing (2007) examined minority and low-
income populations’ access to healthy food. Not surprisingly, minority neighborhoods
had a dearth of healthy food outlets compared to non-minority neighborhoods. Wolch and
colleagues mapped park access in Los Angeles, finding that parks are less accessible (i.e.,
within ¼ mile walking distance) to minority and low-income residents than to white and
more affluent residents (c.f., Wolch, Wilson, & Fehrenbach, 2005; Pincetl, Wolch,
Wilson, & Longcore, 2003). Park access has also been studied by the Trust for the Public

39
Land (c.f., Harnik, 2003). Sellers (2002) included environmental amenities in a
comprehensive study of local governance structures in Germany, France, and the United
States, finding that service-oriented cities offered more environmental amenities than
declining manufacturing cities and that elite neighborhoods enjoyed more environmental
quality benefits. A previous, cross-national study supported the hypothesis that cities with
larger numbers of managerial and professional workers inure more environmental
amenities (Sellers, 1999). More studies are sorely needed to fill the amenities gap
(Swanston, 1999; Lazarus, 1993; Floyd & Johnson, 2002), which is why I chose to
include two amenities my study.

The Theory Gap

The environmental equity literature is more of a concept literature than a
theoretical one. Researchers have yet to converge around a central theory of
environmental equity. Ironically, in a field that heralds distributive and procedural justice
and has produced over 100 studies, a dominant theory has not emerged. Schweitzer and
Stephenson (2007) argue that studies of urban inequality have long tackled questions
about the function of racism and classim in urban problems—the theories of which
environmental justice researchers may simply be “decanting” into new studies (p. 328).
Others (c.f., Liu, 2001) have also noted the lack of a unified or new theory of
environmental justice. Still, many of the theories that inform environmental justice
studies are quite useful.

40
Westra defines justice “as the highest virtue, including in its reach both law and
morality. Justice is a principle and a concept that permits us to assess the cogency and the
thoroughness of laws and judicial decisions by how they compare to that standard” (2006,
p. 138). That is, justice cannot be based on characteristics like social status, ethnicity,
race, etc. Most equity researchers focus on distributive or procedural justice, but
advocates of a broader framework argue that studies should include all distributive,
procedural, and organizational concepts of justice (Peña, 2005; Capek, 1993).
Researchers should consider intergenerational justice as well (c.f., Clayton, 2000; Hiskes,
2006; Cutter, 2006a; Westra, 2006).
Schlosberg (2004, 2007) advocates the Rawlsian concept of justice, “or, more
properly, the rules that govern a just distribution of social, political, and economic goods
and bads,” (2007, p. 13). Rawls’ Kantian-based theory of “justice as fairness” is an
obvious choice for the equity literature. His argument that the “least advantaged” or
“persons whose family and class origins are more disadvantaged” (Rawls, 1980, p. 202)
should enjoy the same primary social goods as others can be extended to include freedom
from environmental harm, even if that goal means setting up a policy system of
differences to benefit the least advantaged. This scheme is derived from the hypothetical
or “original position,” or from behind a “veil of ignorance” whereby one’s place in
society is not known a priori and decisions are made in an objective, rational way
(Rawls, 1971). Justice as fairness, Rawls argues, is “not metaphysical or epistemological”
(1985, p. 230). Rather, it is a political intellection, whereby “social unity” among citizens
is “not founded on their all affirming the same conception of the good, but on their
publicly accepting a political conception of justice to regulate the basic structure of

41
society,” (1985, p. 249). That basic structure includes access to environmental amenities
and freedom from disamenities. The idea of a “politics of difference” (Sandercock, 1998,
p. 185) complements Rawls’ basic ideas about justice. Sandercock argues that some
groups must be treated differently in order to compensate for their victimization by
oppressive social and economic forces. Whereas some would say that “justice is
essentially a matter of impartiality” (Dryzek & Goodin, 1986, p. 4), Harvey (1996)
argues that an egalitarian approach to environmental justice is nearly impossible given
the power dynamics in society. Racism is one such dynamic.
The earliest equity studies focused on “environmental racism,” narrowing the
burgeoning field’s theoretical lens. Institutional racism—the idea that society’s
institutions are imbued with racism that may be neither intentional nor obvious—was
central to many early studies (c.f., Gelobeter, 1992; Bullard, 1993a; Phillips, 1995; Szasz
& Meuser, 2000) and endures in more recent studies (c.f., Pellow, 2007). This focus
sometimes resulted in a leap to intentional racism, as researchers questioned racist intent
in siting (c.f., Pinderhughes, 1996; Novotny, 2000) and the intentional exclusion of racial
minorities in hazards decisionmaking (c.f., Rechtschaffen & Gauna, 2002). These studies
lacked an attendant discourse on race theory, thus “fetishizing” race, rendering it “no
longer politically meaningful,” (Pulido, 1996a, p. 148). Racism was “appropriated by
academics, who by the process of operationalization and adherence to methodological
rigor have transformed an inherently complex and contradictory ideology and set of
practices (racism) into an either/or situation,” (p. 148). Some suggested that equity
researchers use Critical Race Theory (Pulido, 2000), which “pays particular attention to
the roles that race, racism, and nativism play in the formation of legal norms and the

42
administration of justice,” (Yamamoto & Lyman, 2001, pp. 341-342). Similarly, Pulido
(2000) offered “white privilege,” which focuses not on discrimination aimed at people of
color, but on the lack of discrimination endured by white people. Few have heeded her
call (c.f., Boone, 2002; Sicotte & Swanson, 2007; Grineski, Bolin, & Boone, 2007).
Institutional racism and white privilege assume place inequalities, yet few equity
studies invoke theories of residential segregation (for exceptions see Liu, 1997; Boone,
2002; Schweitzer & Stephenson, 2007). Residential segregation is “the principal
organizational feature of American society that is responsible for the creation of the
urban underclass,” (Massey & Denton, 1993, p. 9). Place inequality and segregation are
critical components of power, which is socially and politically constructed, leaving
minority and/or poor populations with limited choices, less sovereignty, and fewer
privileges than their white and/or more affluent counterparts (Squires & Kubrin, 2005;
Swanston, 1999; Massey, 1994; Harvey, 1989). This “geography of exclusion,” limits the
spatial choices of vulnerable populations because “power is expressed in the
monopolization of space,” (Sibley, 1995, p. ix). Limited choices carry “concentration
effects,” including inequitable access to healthcare, education, employment, financial
services, and home ownership (Squires & Kubrin, 2005), as well as unequal access to a
clean residential environment, which can greatly influence health outcomes (Middleton,
2003). Theories of place inequalities are critical to understanding the “social vulnerability
of places” (Cutter, Boruff, & Shirley, 2003, p. 243; see also Cutter, 2006b) and the
residents therein. Acknowledging that the housing system in the U.S. “traps” minorities
in hazardous areas (Bullard, 1990; Hite, 2000) and that capitalism has race effects that

43
perpetuate spatial inequality (Heiman, 1996) is important, but equity studies need to
incorporate and test segregation theories (c.f., Boone, 2002; Lopez, 2002).
Place is an important concept to consider, particularly in a field that focuses on
the hazardous places where people live. The founding of places and the places themselves
shape their residents’ identities (Cannavò, 2007). The choice of residence, however
empowered or disempowered, represents a chance for people to control their identities.
Living in a hazardous area tyrannically shapes residents and harshly limits their self-
determination, demeaning already-vulnerable populations. Moreover, a place is a social
construct, “an aggregation of things and relationships—human and nonhuman, social and
ecological—that are tangibly cohering,” (Cannavò, 2007, p. 20). That cohesion can be
strong when people form attachments to place: “Such place-based emotional bonds are
essential to personal and social community facets of identity,” (Long & Perkins, 2007, p.
566). The physical environment of that place can nurture or harm the social psychology
of the residents therein. Research has shown that “physical incivilities” (e.g., trash,
dilapidated buildings, graffiti) can lower place attachment (Brown, Perkins, & Brown,
2003) as social psychology is bound to place. As well, community attachment is
determined both by social ties and the natural environment in which the community
exists (Brehm, Eisenhauer, & Krannich, 2006). Thus, the psychological impact of a
hazardous site could negatively affect both place attachment and community
attachment—something that few equity researchers consider. Conversely, environmental
amenities have been shown to foster attachment. For example, MacArthur Park was
shown to be an important source of place attachment and of “continuity of identity” for
Latino immigrants in Los Angeles (Main, 2007). Thus, social attachment and feelings of

44
community may be enlarged by the presence of amenities, but these concepts are rarely
addressed by equity researchers.

Searching for a Theory

Some equity researchers have used alternative theoretical frameworks. New
Institutional Economics provided the basis for Hamilton’s TSDF studies (1993, 1995,
1999). Arguing that institutional racism was central, but that location choices are
controlled by government and corporate elites, Stretesky and Lynch (2002) suggested a
“political economy of race.” Sze has argued that critical studies like ecofeminism “can
deepen our understanding of the origins of environmental racism,” (2002, p. 166). These
are welcome alternatives, but the lack of a coherent, specific environmental justice theory
opens the field to criticism: “Academics who write about environmental justice are, as a
rule, strikingly unconcerned with the abundant (and ancient) scholarly and philosophical
literature addressing equity and justice,” (Foreman, 1998, p. 9). Moreover, Pulido argues
that “it is essential that [environmental justice] researchers begin to situate their work in
terms of a larger sociospatial dialectic,” (2000, p. 33). I chose the social movements
framework for my study to broaden the theoretical debate in the field.

The Comparability Gap

One reason federal policy makers may shy away from addressing the distribution
of environmental hazards may be the lack of comparability across studies. This lack

45
severely restricts the ability of researchers to generalize findings (Cutter, 2006c). Several
reasons explain this: researchers have yet to agree on a standard research design or
methodology; the universe for studies ranges from city to county to state to region to
nation, and focuses predominantly on urban areas; researchers disagree on the appropriate
unit of analysis; researchers use different dependent variables and myriad independent
variables; and control populations are often overlooked or only casually examined. Also,
researchers have yet to develop an index of environmental equity (for exceptions, see
Harner, Warner, Pierce, & Huber, 2002; Warner, 2001), though one would be hard to
construct (c.f., Hayward, Fowler, & Steadman, 2000). Finally, sophisticated GIS analysis
has inadvertently limited the universe for individual studies: by affording researchers the
precision and ability to map complex hazards with endless sociodemographic and
geographic details, studies of multiple locations quickly become too laborious.
Why is comparability important? First, the federal government is the locus
environmental equity, so researchers must develop broad recommendations to inform
policy makers and administrators. Even multiple studies done on a single city or region
are not enough. After several in-depth studies on Southern California cities, Pastor and
colleagues warn that “it is unclear whether study results from southern California can be
meaningfully generalized to other regions in the United States,” (Morello-Frosch, Pastor,
Porras, & Sadd, 2002). Second, equity studies are too often “liable to the criticism of
sampling on the dependent variable in the choice of study area,” (Baden, Noonan, &
Turaga, 2007). Studies that can reveal patterns of inequity, or lack thereof, across a
diverse and random sample of cities and dependent variables are more likely to be viewed

46
as reliable. Third, comparisons are needed to generate additional hypotheses that can be
tested more widely and may in turn contribute to a theory of environmental justice.
About 25% of existing studies is national in scope, yet the studies have little in
common outside of a race and/or income hypothesis. The dependent variables and units
of analysis vary making comparisons of results impossible. For example, about one third
of the national studies have examined TSDFs using units of analysis ranging from zip
codes (c.f., UCC, 1987; Hamilton, 1993; Hamilton, 1995) to tracts (c.f., Anderton,
Anderson, Oakes, & Frasier, 1994; Anderton et al., 1994; Oakes, Anderton, & Anderson,
1996; Been & Gupta, 1997; Markham & Rufa, 1997; Davidson & Anderton, 2000) to
counties (c.f., Bowman & Crews-Meyer, 1997; Hird & Reese, 1998; Lester, Allen, &
Hill, 2001). Studies of TRI facilities or emissions similarly used counties (Perlin, Setzer,
Creason, & Sexton, 1995; Hird & Reese, 1998; Arora & Cason, 1999; Lester, Allen, &
Hill, 2001; Daniels & Friedman, 1999), zip codes (Ringquist, 1997; Brooks & Sethi,
1997; Krieg, 2005), and tracts (Morello-Frosch & Jesdale, 2006).
Researchers have conducted multiple studies on the same location, yet comparing
across these studies is difficult. Consider the eight studies on the City of Los Angeles and
the surrounding areas.
8
Five studies examined TRI facilities or emissions (Burke, 1993;
Pulido, Sidawi, & Vos, 1996; Sadd, Pastor, Boer, & Snyder, 1999a; Morello-Frosch,
Pastor, & Sadd, 2002; Pastor, Sadd, & Morello-Frosch, 2004b), but two focused
specifically on the geographic area of LAUSD (Morello-Frosch, Pastor, & Sadd, 2002;
Pastor, Sadd, & Morello-Frosch, 2004a). Expanding the universe, Burke (1993) studied
Los Angeles County, as did Boer, Pastor, Sadd, and Snyder (1997). These studies present

8
This figure does not include studies done on the South Coast Air Basin (c.f., Brajer & Hall, 1992;
Morello-Frosch, Pastor, & Sadd, 2001; Schweitzer, 2006) or statewide or national studies.

47
a rich picture of environmental equity in the Los Angeles area, but none attempts to
replicate the other studies, nor have researches tested these designs in other metropolitan
areas. A similar argument can be made for the studies done on Phoenix (c.f., Bolin et al.
2002; Bolin, Grineski, & Collins, 2005; Grineski, Bolin, & Boone, 2007).
Some researchers have “re-tested” previous study areas. For example, Been
(1994) designed her study of TSDFs in Houston, TX, and Alabama, South Carolina, and
North Carolina to examine the results of Bullard’s 1983 study and the GAO’s 1983 study.
Yandle and Burton (1996), Cutter and Solecki (1996), and Liu (1997) also chose
Houston, TX for their studies to test Bullard’s earlier work. Other researchers similarly
chose the Southeast areas that the GAO study examined (c.f., Bowman & Crews-Meyers,
1997; Mitchell, Thomas, & Cutter, 1999). These studies are important, but do not provide
comparability on a national scale, which is why I examine 18 cities nationwide.

The Health Effects Gap

Many early studies of environmental equity augured grave danger to the health of
minority and poor populations living in close proximity to environmental hazards (c.f.,
Pinderhughes, 1996). Many researchers advocate more studies to examine health policy
and environmental equity (c.f., Vrijheid, 2000; Bowen & Wells, 2002; Cutter, 2006a;
Buzzelli, 2007). Thus far, in the equity literature, spatial coincidence and even proximity
analysis have proven inadequate proxies for actual health effects from ecotoxicants
(Institute of Medicine, 1999). Some equity studies have shown correlations between
adverse respiratory effects and air pollutants (from stationary and mobile sources) in poor

48
and/or minority populations (c.f., Prakash, 2007; Ayers, 2002; Pastor, Morello-Frosch, &
Sadd, 2006; Morello-Frosch, Pastor, & Sadd, 2001, 2002) and some studies have tied
these respiratory effects to decreased academic performance (c.f., Pastor, Sadd, &
Morello-Frosch, 2004a). Maantay (2007) studied asthma rates and air pollution in the
Bronx, demonstrating that 88% of the people living within .25- and .5-mile buffers were
minority, 33% were poor, and asthma hospitalizations were 30% more likely to occur
within those buffers. Some have examined reproductive issues. Frazier, Margai, and
Tettey-Flo (2003) found higher lead levels in blood, low birth weight, and cancer in poor
and minority children living near environmental hazards.
Others have been critical of these efforts. René, Daniels, and Martin (2000) cite
the formidable challenge to obtain exact dosage and exposure of those populations to the
environmental hazard (see also Vrijheid, 2000), and suggest that epidemiological studies
must be done at the cellular and molecular level. Similarly, Gragg, Gasana, and Christaldi
(2002) write, “The principal problem of the environmental justice movement is the lack
of biologically based scientific evidence that supports the claims of adverse health effects
caused by exposure to environmental contaminants from nearby environmental hazardous
sites,” (p. 290). Cutter (2006c) argues, “One of the biggest stumbling blocks to this line
of enquiry is the lack of a consistent measure of toxicity that the social science
community can use in comparing risk,” (p. 267). Toxicologists and epidemiologists have
demonstrated in birds, fish, and invertabrates the adverse effects of ecotoxicants like
dioxins, which “induce harmful effects on many organ systems,” (Blanco & Cooper,
2004, pp. 465-466). Still, understanding hazard effects on humans remains elusive.
Epidemiological studies require vast resources and time, and they are “difficult and

49
protracted” (Davis & Webster, 2002, p. 25; see also Liu, 2001; Rhodes, 2003). Sadly,
some health affects found in poor populations may be “the inevitable and natural, if
unwanted, by-products of a modern industrial society,” (Hofrichter, 2000, p. 2). Smoking,
lifestyle choices, and alcohol consumption often confound epidemiological studies (Petts,
2005). Without an understanding of exposure, epidemiology, and toxicology—something
that may be beyond the expertise of most equity researchers, including this one—
policymakers are stymied by uncertainty and unable to make informed policy choices.

Concluding Thoughts

Though a majority of studies support the environmental equity hypothesis, the
evidence is less than conclusive, something that most reviews of the literature admit (c.f.,
Pellow; 2004; Bowen, 2001; Davidson; 2003; Black & Stewart; 2003; Schweitzer &
Stephenson, 2007). Some argue the evidence is overwhelming (c.f., Novotny, 1998;
Girdner & Smith, 2003), and research design and methods have improved markedly over
the simplistic and faulty designs of the seminal and early works in the field. Still, the
extant corpus belies a lack of a specific environmental justice theory, problematic
methodology, inadequate measures of risk and health affects, few longitudinal studies,
and a lack of generlizability and comparability across studies. Moreover, additional areas
must be mapped, diverse hazards and amenities must be studied, and additional
vulnerable populations must be included, like women and immigrants (Cutter, 2006a;

50
Pellow & Park, 2002) and the elderly. Environmental equity research is rich and
complex, yet unfinished and necessary.

51
CHAPTER THREE: RESEARCH DESIGN AND METHODS

Accordingly, absent research guided by reasonably high standards of
social science method, regardless of laudable concepts of fairness
contained in government policy or administrative decisions, efforts to
solve the problem or improve the situation are likely to be
counterproductive if not massively destructive.
William Bowen, Environmental Justice Through Research-based
Decision-making

The chapter explains my overall research design, as well as the specific methods
used in the GIS analysis of environmental equity (i.e., Phase I of the study). I discuss the
methods used for the institutional analysis in Chapter 6. Here, I describe the study’s
design and intent, followed by a description of the population and sampling methods. I
then specify the research questions, variables, data collection and analyses for Phase I.

Overall Research Design

I used a comparative case-study approach to examine environmental equity in the
United States. In this sense, my study is an exploratory examination of how cities differ
in their current levels of environmental equity and how a city’s institutional context may
contribute to a more or less equitable distribution of environmental amenities and
disamenities. By comparing 18 cities, I hoped to reveal patterns of institutional variables
that might describe why one city has a more equitable distribution than another. For this
type of empirical investigation, the case-study method was appropriate (Yin, 2003; King,
Keohane, & Verba, 1994; Gering, 2004). Critics of this method cite a lack of rigor on the

52
part of the researcher, difficulty in generalizing results, and difficulties in synthesizing
the results in a concise way (Yin, 2003). Critics of the multiple case-study design argue
that doing more than two to three cities becomes “unwieldy” (De Socio, 2007). However,
the use of multiple cases may yield more compelling results than single case studies (Yin,
2003); allows the researcher to control contextual variables across cases to understand
patterns at work (Sartori, 1991; Pierre, 2005); allows a researcher “to assimilate and
differentiate” different entities (Sartori, 1991, p. 246); and can help researchers confirm
results of a phenomenon in myriad locations (Durkheim, 1982).
In addition, I chose a comparative multi-case study design because most case
studies in environmental justice focus on a single city or metropolitan area. Similarly,
Pierre argues that “urban politics has largely remained dominated by single-case studies
supplemented with the odd two-case comparison,” (2005, p. 447). Though more multi-
case studies appear in recent literature, urban research “can be idiosyncratic, focusing on
problem solving in specific localities,” (Andranovich & Riposa, 1993, p. 47). Still,
comparing cities and drawing generalizations based on the limited number of cities is
difficult. The unique historical development of cities in the United States (from the early
settlers in the Northeast to the Southern and Westward expansions) and the diverse
geography of the continental U.S. contribute to this difficulty. Platt argues, “[E]ach major
city also has developed its distinct landscape, traditions, and character, reflecting the
diverse influences of demographics, culture, physical site, politics, economy, and law,”
(1996, p. 128). While historical case-studies of 18 cities is beyond the scope of this
dissertation, this preliminary comparative analysis may yield potentially important
patterns for future research in both environmental justice and urban politics.

53
As stated in Chapter 1, my study involved two phases. First, I examined the
distribution of two environmental disamenities and two amenities at the Census-tract
level in 18 cities to answer the following question: Are minorities and poor populations
over-represented in neighborhoods where environmental disamenities cluster and under-
represented in neighborhoods where environmental amenities cluster? In Phase II, I
used the political process approach of the Social Movements framework to answer the
following question for each city: What are the city-level political processes and
institutional variables that are more likely to be associated with better disamenities and
amenities distributions? I describe the population and sampling method below. Following
that is an explanation of the methods, variables, and analyses of Phase I of my study.

Population and Sample

Two dangers in comparative and case-study research are: (1) to select cases on the
dependent variable and (2) to select cases familiar to the researcher (Peters, 1998). The
former constitutes one of the “moral sins that can afflict comparative research” and the
latter can lead to cases that are weak and “are not necessarily the best for testing
theories,” (Peters, 1998, p. 31 and p. 51). Further, a comparative study must include
myriad cases that are “selected in a purposeful and not random fashion,” (Pierre, 2005, p.
454). Given those guidelines, my hope for this study was to bring new light to
environmental equity by examining 18 cities that had not been studied previously (see
Table 3.1), though I did not achieve complete originality in my sample. I selected the

54
cities non-randomly, because case studies rarely use traditional sampling logic (Yin,
2003), and based on four criteria.

Table 3.1
Cities Selected for Study
City Population Primary Economy Form of Government

Rust Belt
Philadelphia 1,517,550 Manufacturing Mayor/Council
Boston* 589,141 Education Mayor/Council
Grand Rapids 197,800 Manufacturing Council/Manager
Dayton 166,179 Industrial Council/Manager
Decatur IL 81,860 Industrial Council/Manager
Albany* 95,658 Government Mayor/Council

Sun Belt
San Diego 1,223,400 Defense Mayor/Council
†

Austin* 656,562 High Tech Council/Manager
Miami 362,470 Corporate Mayor/Commission
Little Rock* 183,133 Diversified Mayor/Council
†

Santa Fe* 62,203 Science/Tech Mayor/Council
Flagstaff 52,894 Tourism Council/Manager

Other
San Jose 894,943 High Tech Council/Manager
Seattle 563,374 Fortune 500 Mayor/Council
Norfolk** 234,403 Defense Council/Manager
Salt Lake City* 181,743 Service Mayor/Council
Boulder 94,673 High Tech Council/Manager
Charleston WV* 52,421 Government Mayor/Council
*State capitol **Not a county seat
†
Previously a council/manager city

First, I selected cities with varying populations based on the 2000 U.S. Decennial
Census. I chose six big cities (population > 500,000); six medium cities (population
100,000-499,999); and six small cities (population 50,000-99,999). Previous
environmental studies have focused predominantly on large- and medium-sized cities,
and I wanted to examine smaller cities to see how size might matter. As well, a challenge
in environmental justice is to examine insidious cases, which may be in smaller cities.

55
Second, I wanted to create a good geographic representation of the United States
to understand how environmental equity might vary in geographic regions. The primary
geographic distinction I used was Rust Belt versus Sun Belt. I selected six cities each
from the Rust Belt and Sun Belt, selecting the remaining six from geographically unique
places like Appalachia, the Pacific Northwest, the Atlantic seaboard, and the Rocky
Mountains. The distinction between the Rust Belt (sometimes called the Frost Belt) and
the Sun Belt is one that distinguishes between the old industrial cities of the Northeast
and the newer cities that experienced rapid population growth starting in the 1950s
(Peterson, 1995). The geographic demarcation of the two is often considered the 37
th

parallel, which runs roughly from North Carolina to California. All 15 states below the
37
th
parallel are considered to be Sun Belt states, though technically just the southern tip
of Nevada and Southern California are included. The Rust Belt does not reach from East
to West; rather, 14 states make up the Rust Belt and they are primarily in the Northeast
with a few states in the Upper Midwest (Judd & Swanstrom, 2008). The Sun Belt states
tend to be “highway cities,” characterized by suburbanization, industrial parks, reformed
governments, and development-oriented (Peterson, 1995; Nicolaides, 2003). Sun Belt
cities are characterized by diverse racial and ethnic populations and, in particular, large
Hispanic populations (Hero, 2005). Sun Belt economies are generally less industrial and
more service-oriented (Hero, 2005). Conversely, Rust Belt cities are generally higher-
density cities with their populations located in central cities (Dreier, Mollenkopf, &
Swanstrom, 2004). They tend to be heavily industrialized or transitioning economies with
a strong industrial history, and characterized by population decline in the decades after

56
World War II. These cities tend to have higher poverty and crime rates, and local
governments entrenched in bureaucratic and political patterns (Peterson, 1995).
Though Rust Belt cities have predictable economies, my third selection criterion
was a city’s primary economy. I opted for diverse economies that included
manufacturing, high tech, government, tourism, and service. In the case of a few cities,
their primary economy changed. For example, San Jose, CA, was once an agricultural
economy, but today it is firmly a high-tech economy. Likewise, Boston was a
manufacturing town, but has transitioned to an education and research economy. A city’s
economy can affect the environmental outcomes of local governance policy, beyond the
simple economic determinism arguments of city politics (e.g., Peterson, 1981). For
example, Sellers (2002) found that cities with service-oriented economies promoted more
environmental improvement in their local policies than manufacturing-oriented cities.
The final criterion I used to select cities was form of government. I hoped to
divide each grouping (e.g., Rust Belt, Sun Belt, Other) evenly among mayor/council and
council/manager forms by population size. That proved difficult, but I selected 10
mayor/council and 8 council/manager cities with fairly equitable regional distribution. In
all, it proved quite challenging to select 18 cities according to my four criteria, but I am
confident that my sample represents a diverse and interesting collection of cities.

Phase I Research Methods

The conceptual goals of the equity analysis are to examine the spatial distribution
of hazards and amenities, and conduct proximity analysis to address the following

57
justice-based hypotheses: (1) The greater the percentage of poor and/or minority
residents in a tract, the greater the number of disamenities and the more unequal the
distribution; and (2) The greater the percentage of poor and/or minority residents in a
tract, the fewer the number of amenities and the more unequal the distribution. The unit
of analysis for Phase I is the Census tract. To examine the spatial distribution, I used GIS
mapping technology and spatial statistics to create and analyze each city individually.
When GIS mapping and spatial statistics are used concurrently, researchers are able to
investigate the location of events (in this case, amenities and disamenities) and identify
patterns and clusters of those events. Environmental equity involves spatial and
nonspatial data, which create complexities for researchers especially in dense urban
areas: “The multiplicity of data—spatial and nonspatial—that describes an urban area and
its built and natural environment is exceedingly difficult to understand and manage
without GIS,” (Maantay & Zeigler, 2006, p. 11).

Variables and Data Collection

The dependent variables for Phase I include two disamenities and two amenities.
The disamenities are: (1) TRI facilities (with stack and fugitive emissions > 0) for
reporting year 2006; and (2) large-quantity waste generators (those generating 1,000
kilograms or more of hazardous waste monthly, those generating more than 1 kilogram of
acutely hazardous waste monthly, or a per month “acute spill” greater than 100
kilograms) for reporting year 2005. The amenities are community gardens and city parks.

58
Table 3.2 lists the nearly 3,000 disamenity and amenity sites discovered. Figure 3.1
arrays the distribution of amenities and disamenities graphically.

Table 3.2
Disamenity and Amenity Counts Per City
City TRI Facilities LQGs City Parks Community
Gardens

Rust Belt
Philadelphia 33 39 79 100
Boston 6 26 249 148
Grand Rapids 36 27 76 7
Dayton 25 25 33 18
Decatur IL 6 6 40 7
Albany 3 12 35 20

Sun Belt
San Diego 28 97 73 12
Austin 19 21 144 7
Miami 13 29 39 22
Little Rock 10 12 52 9
Santa Fe 0 0 26 6
Flagstaff 1 1 23 4

Other
San Jose 22 58 146 19
Seattle 31 77 328 56
Norfolk 7 8 32 6
Salt Lake City 34 19 65 7
Boulder 9 8 78 8
Charleston WV 2 6 17 2

For the dependent variables, I downloaded the following TRI data from the EPA’s
Envirofacts Warehouse: facility ID, name, address, latitude, longitude, and total on- and
off-site releases in pounds. I downloaded the following LQG information from the Right-
to-Know Network and cross-referenced it with the RCRA data in the EPA Envirofacts
Warehouse: EPA ID, name, address, latitude, longitude, and tons managed. I verified
addresses using Google Maps Street View and Google Earth. GIS maps are most accurate

59
when points are geolocated (i.e., located via latitude and longitude
9
). Thus, I cross-
referenced (i.e., geocoded) EPA addresses using Google Earth to ensure the accuracy of
the EPA latitude and longitude. In some cases, like Decatur, the Google Earth resolution
was bad in various parts of the city. When that occurred, I used a combination of Google
Earth and geocoding software (www.batchgeocode.com) to geocode addresses.
A leading complaint about EPA data is the “locational inaccuracies associated
with the geographic coordinates of facilities,” (Sadd, Pastor, Boer, & Snyder, 1999b, p.
109; see also Scott & Cutter, 2006; Perlin, Sexton, & Wong, 1999; Maantay 2007;
Stretesky, 1998). Researchers expend great resources geocoding because of incorrect
addresses, wrong latitude and longitude coordinates, and addresses that identify
administrative buildings rather than the buildings where chemicals are used (Bolin et al.,
2002; Scott, Cutter, Menzel, Ji, & Wagner, 1997). For example, in the downloaded LQG
data on San Jose, CA, 27 out of 58 facilities had bad or missing addresses or latitude and
longitude coordinates. This required me to search all EPA databases by handler ID, as
well as confirm the locations via Google Maps for this missing 47% of facilities.
Following that, I geocoded the facilities via Google Earth.
The amenity locations were more difficult and time-consuming to identify. To
find city parks, I first consulted each city’s website to determine if city parks were listed.
In many cases, both the names and the street addresses of the parks were listed on the
website. Where only names and vague locations were available, I contacted each city’s
appropriate department to request a list of parks. In some cases, I also conducted a search
of the World Wide Web (e.g., visitor information, chambers of commerce, nonprofit

9
I used the Decimal Degrees format for latitude and longitude.

60
organization sites) to complete the list of parks. I used Google Maps, including Street
View, to clarify or specify park addresses and then used Google Earth to determine the
latitude and longitude for each park.

Figure 3.1
Total Disamenities and Amenities by City

Total Number of Sites
0
50
100
150
200
250
300
350
400
450
Albany
Austin
Boston
Boulder
Charleston
Dayton
Decatur
Flagstaff
Grand Rapids
Little Rock
Miami
Norfolk
Philadelphia
Salt Lake City
San Diego
San Jose
Santa Fe
Seattle
Disamenities
Amenities

The community gardens were even more difficult to identify. I first consulted
each city’s website to determine if a list of community gardens was available. Where no

61
listing was provided, I contacted the city’s appropriate department to obtain a list. I then
used resources on the World Wide Web to identify nonprofit organizations, like the
American Community Gardening Association, to identify additional community gardens.
I also consulted County Extension office websites (and followed up with phone calls
where necessary) for each city to identify additional locations. In several cases, only a
street name or street corner was provided as an address, which meant I had to make
visual inspections of each garden using Google Maps Street View. Following that, I
geocoded each community garden using Google Earth.
Three caveats are in order. First, according to the reporting years used, Santa Fe
has no disamenities as I defined them. Santa Fe does have TRI facilities, but those
facilities either did not report to the EPA or had zero releases in 2006. Similarly, Santa Fe
has no LQGs. I examined prior years’ data, but was unsuccessful in identifying facilities
that met my specifications. Thus, Santa Fe was dropped from the disamenities analysis.
Second, in some cases (e.g., Miami), sites were located just outside the city boundary, but
within one-half mile. These sites were included in all but the regression analyses because
of their potential effects on nearby populations. Third, Philadelphia has a great tradition
of community gardens. Unfortunately, the city, nor the American Community Gardening
Association or local nonprofits have a comprehensive list of the gardens. I located and
verified 100 gardens; but some people claim that city has nearly double that amount.
Additional gardens may be “guerilla gardens,” which are started without permits or
property rights and thus, are not being tracked. I am confident that I located the
community gardens on record, but future field work may yield additional sites.

62
Independent Variables

The independent variables I used in this study are the standard independent
variables used in the Environmental Justice literature. They include racial and ethnic
categories, population, income, wealth, and education variables as specified below. For
the independent variables, I used data from the 2000 U.S. Decennial Census, downloaded
from the American FactFinder service of the Census website (www.census.gov). Once
downloaded in .csv format, I sorted and coded the data accordingly, before adding them
to the GIS maps.

Table 3.3
Independent Variables Used in Phase I

Name

Description

Source

PCT_WHITE
PCT_BLACK
PCT_HISPAN
PCT_OTHER
a

PCT_ASIAN
b
PCT_MIN

Race/Ethnicity

SFI, Hispanic or Latino by Race (Total
Population), P8

PCT_TENURE Percentage of people living in
neighborhood 5+ years

SF1, Tenure—Owner Occupied Housing
Units, H4

PCT_MFG Percentage of males and
females employed in
manufacturing
SF3, Sex by Industry for the Employed
Civilian Population 16+ Years, P49

MED_INC_99 Median income SF3, Median Household Income in 1999
(Dollars), P53
OO_HS_VAL Median value of owner-
occupied housing
SF3, Median Value (Dollars) for All
Owner-Occupied Housing Units, H85
PCT_ED_BA Percentage of population 18
years or older with a BA or
better
PCT25-Sex by Age of Educational
Attainment for the Population 18+ Years

a
In all but five cities, this variable includes the Census categories: Asian, Native Hawaiian/Pacific Islander,
American Indian/Alaskan Native, and Some Other.
b
This variable is used in only five cites, as explained below.

63

The race and ethnicity categories I included in this study are white
(PCT_WHITE), black (PCT_BLACK), Hispanic (PCT_HISPAN), other (PCT_OTHER),
and percent minority (PCT_MIN). In all cities, except Boston, Philadelphia, San Diego,
San Jose, and Seattle, “other” includes the Census categories Asian, Native
Hawaiian/Pacific Islander, American Indian/Alaskan Native, and Some Other. However,
in Boston, Philadelphia, San Diego, San Jose, and Seattle, Asian is treated as a separate
category (PCT_ASIAN). Asian populations are relatively small in comparison to other
racial and ethnic minorities in all cities except Seattle, where Asians are the leading racial
minority. Though some previous studies have included Asians as a separate category
(c.f., Pastor, Sadd, & Morello-Frosch, 2004b; Morello-Frosch, Pastor, Porras, & Sadd,
2002), no consistent or definitive method for including Asians as a separate category
exists. To determine when to include Asians as a separate category in my study, I used
the guidelines provided for defining minority populations in conjunction with EO 12898:

Minority populations should be identified where either: (a) the minority
population of the affected area exceeds 50 percent or (b) the minority population
percentage of the affected area is meaningfully greater than the minority
population percentage in the general population or other appropriate unit of
geographic analysis. (Council on Environmental Quality, 1997, p. 25)

In my study, Boston, San Diego, San Jose, and Seattle had multiple tracts where the
Asian population exceeded 50% or was greater than the overall minority population in
the city (see Table 3.4). In Philadelphia, Asians exceeded the overall minority population
in only one tract; however, in that tract, 80% of the population was Asian and merits

64
further examination. Interestingly, these five cities are five of the six largest cities in my
study. Austin completes that list, but the Asian population (7.47% overall) did not exceed
the city’s overall minority population (41.82%) in any tract.

Table 3.4
Asian Populations in Selected Cities
Overall Minority
Population
Overall Asian
Population
Tracts Where Asians Exceed Overall
Minority Population
Boston 47.43% 7.47% 2
Philadelphia 55.91% 4.42% 1
San Diego 46.50% 12.76% 7
San Jose 58.12% 25.14% 24
Seattle 32.27% 13.17% 25

I included two wealth and income variables. The standard measure for income is
median household income (MED_INC_99), which I included. However, to better reflect
a person’s wealth, I included the median value of owner-occupied housing
(OO_HS_VAL) and the percentage of people working in manufacturing jobs
(PCT_MFG). I included two variables that help reflect a person’s empowerment and
attachment to their neighborhood. The first, PCT_ED_BA is the percentage of the
population 18 years or older who hold a bachelor’s degree. A person’s education level
may be reflective of how willing they are to participate in political actions. The second,
PCT_TENURE, reflects the percentage of people in owner-occupied housing who have
lived in that housing for five or more years. The more time people have lived in a
neighborhood, the more willing they may be to engage in civic actions as they may feel
more attachment to place.

65
Control Variables

I employed two types of control variables (see Table 3.5). The first is population
density. ArcMap calculates the shape area of each Census tract in meters. For population
density, I first divided the shape area by 10,000 to determine the hectares of each Census
tract and then divided the total population of each tract by its measurement in hectares.
The second type of control variable I used is a land use variable, for which I selected the
United States Geographic Service (USGS) National Land Cover Data (NLCD). This
dataset, completed in 2000, is based on extensive aerial photography, Landsat Thematic
Mapper imagery, and other data sources, with land uses coded according to the Anderson
Classification System (Ralston, 2004). Specifically, I used codes 21 (low-intensity
residential); 22 (high-intensity residential); and 23 (commercial/industrial/transportation).
The variable is operationalized as the percentage of each tract devoted to each of the
three types of land use.

Geographic Data

I constructed each GIS map using geographic data available from the U.S. Census
Bureau’s Topologically Integrated Geographic Encoding and Referencing (TIGER)
system. TIGER shapefiles include administrative boundaries, as well as rivers, roads, and
railroads. I downloaded the boundaries and the Census tracts for each city from
http://www.census.gov/geo/www/tiger/tgrshp2007/tgrshp2007.html. TIGER shapefiles
are only boundary files; they contain no Census data (e.g., race, income).

66

Table 3.5
Control Variables Used in Phase I

Name

Description

Source

POP_DENSITY

Population density

US Census, SF1, Total Population, P1

PCT_LDCV21
PCT_LDCV22
PCT_LDCV23

Land Use
Land Use
Land Use

USGS NCLD, Gridcode 21
USGS NCLD, Gridcode 22
USGS NCLD, Gridcode 23

Geographic and Spatial Statistical Analyses

In this section, I describe the basic geographic distribution methods and spatial statistics I
used in the study. A more thorough and technical explanation of exactly how I
constructed the GIS maps and conducted the analyses is in the Appendix. For each of the
18 cities, I constructed a base map using ArcCatalog and ArcMap
10
in conjunction with
Microsoft Excel. Each city map contains the city boundaries, Census tracts, the
environmental disamenities and amenities, the USGS land cover data, and the Census
variables (all in vector format). Each map was made from a single template so that all 18
maps are the same, with the exception of Santa Fe because it contains no hazard data.
GIS maps have tremendous data capacity, so one map can be constructed for a city and
multiple geographic and statistical analyses can be run from the base map.

10
ArcCatalog is a filing and organization program, while ArcMap is a map-processing program. Both are
contained within ArcView 9.3, which is part of ArcGIS, widely considered to be the leading GIS software
and produced by ESRI (Environmental Systems Research Institute) in Redlands, CA.

67
The goal for the mapping and spatial statistical analyses was to understand the
following: (1) What is the distribution of the disamenities and amenities (i.e., features);
(2) Is a pattern evident in the distribution of these features vis-à-vis the sociodemographic
data; (3) Do the features cluster; and (4) Where do features cluster (Mitchell, 2005)? In
addition to visualization of these phenomena using chloropleth and density maps, these
questions can be answered by specific geoprocessing and spatial statistics. Before
describing these, it is important to understand that spatial data often violate common
assumptions of nonspatial data. First, spatial data usually exhibit spatial autocorrelation,
based on Tobler’s First Law of Geography, which states that everything is related to
everything else, but near things are more related than distant things. As well, spatial data
often lack independence: “The smaller the spatial units, the greater the probability that
nearby units will be spatially dependent,” (Anselin & Getis, 1992, p. 24). Therefore, the
analytic techniques described below are needed to explore the patterns and correlations
among spatial data (Goodchild, Anselin, Appelbaum, & Harthorn, 2000).

The Distribution of Disamenities and Amenities

As a preliminary step, I calculated the Mean Center for the each of the four types
of disamenities and amenities. The Mean Center examines all of the sites (based on their
XY coordinates) in a particular feature class (e.g., LQGs) and determines the average XY
location, or the where the center of the distribution is located. The average location for
each type of hazard and amenity is useful to identify and compare to reveal any specific
zones or locales of hazards or amenities.

68
One way to describe how disamenities and amenities are distributed is to visually
examine a chloropleth map of Census tracts and a density map that displays the amenity
and hazard sites. However, to describe the sociodemographic characteristics of a Census
tract that contains a site does not account for (1) where that site is located within the tract,
e.g., in the center or on the edge; and (2) populations in surrounding tracts that may be
affected by the sites. To better describe where the sites are located and who is affected by
them, researchers often use buffer containment analysis, which draws a buffer around
each site at a specified distance, resulting in a new feature class. This new class can be
unioned with Census tract or other data to capture the attribute data within the buffer.
Three traditional ways of conducting buffer analysis include: (1) polygon containment,
where the entire population of all the Census (or other) polygons inside and adjacent to
the buffer are summed; (2) centroid containment, where a polygon is only counted if its
centroid is in the buffer; or (3) buffer containment in which the Census units that intersect
the buffer are counted with the values determined via areal interpolation (Chakraborty &
Armstrong, 1997). Centroid and buffer containment are the preferred methods. For this
study, I opted not to use buffer analysis, though I had intended to do so. Initially, I drew
½-mile circular buffers around the disamenities, as per the environmental justice
literature (c.f., Glickman & Hersh, 1995; Glickman, Golding, & Hersh, 1995; Pastor,
Sadd, & Morello-Frosch, 2004b). For the amenity sites, I drew ¼-mile circular buffers,
which are generally considered the distance that people are willing to walk to amenities
(c.f., Wolch, Wilson, & Fehrenbach, 2005). Due to the nearly 3,000 sites in this study,
however, many buffers overlapped; in some cases anywhere from 3-7 buffers overlapped.
Unfortunately, to identify the population within each individual buffer would grossly

69
over-count the populations affected. Likewise, GIS does not allow the user to create a
buffer zone (i.e., merge overlapping buffers into one buffer) without losing the attribute
data contained within those buffers. Therefore, I could not precisely measure the
populations within multiple, overlapping buffers. So, to describe the populations affected
by the hazards, I used several other pattern and cluster analyses, many of them rarely
used in environmental justice studies.

Spatial Patterns and Clusters

Recall that spatial data often exhibit autocorrelation, but in a different sense than
nonspatial data: “Whereas correlation statistics were designed to show relationships
between or among variables, spatial autocorrelation shows the correlation within
variables across a georeferenced space,” (Getis, 2008, p. 298). Spatial statistics are used
to determine if spatial autocorrelation is present in a dataset, but the extent of the spatial
autocorrelation within a particular geographic region. From Tobler’s law discussed
above, similar geographic features tend to cluster, and several studies have shown that
environmental hazards cluster (c.f., Schweitzer, 2006; Szasz & Meuser, 2000). To
determine if the hazard sites or amenity sites in my study cluster, two types of analysis
were required. The first was a global analysis that describes the variance of the
distribution of the entire study area, or describes first-order effects. I used a global spatial
statistic to examine the overall spatial patterns within each city. Several global statistics
may be used (e.g., Moran’s I, General Getis-Ord G, Ripley’s K, Geary’s C), but the most
common global statistic used to test for spatial autocorrelation is Global Moran’s I (Getis,

70
2008, Mitchell, 2005). Moran’s I is a spatial statistic that can confirm if patterns exist by
measuring the similarity of nearby features (based on the XY location of all features) and
identifying the probability that the patterns are not due to a random distribution. An I > 0
indicates the features are clustered, while an I = 0 indicates features are distributed
randomly and an I < 0 indicates that features are dispersed. The Moran’s I tool also
returns Z-scores, confidence levels, and probabilities.
Another way to analyze the global pattern is Ripley’s K (sometimes called a “K
function”). Ripley’s K is a global statistic that measures spatial dependence at different
distances by counting those features within specific geographic areas and comparing
them to a computer-generated random distribution. The statistic can reveal at what
distances the pattern is clustered and/or dispersed. The K function generates values for an
expected K (i.e., the random distribution), the observed K (i.e., the actual distribution),
and lower and upper confidence levels. The confidence or minimum/maximum levels are
generated based on Monte Carlo simulations. I opted for 99 permutations or 100 Monte
Carlo simulations, which translates into a probability level of .01, or the odds that a
results would fall outside the upper and lower confidence levels equal to 1 in 100.

Where Features Cluster

The next step is to determine where amenity and disamenity sites cluster (a
second-order effect), something that a global statistic cannot determine. This is why
global and local statistics should be used in conjunction (Mitchell, 2005). “Local”
statistics calculate values for every feature within a study area and identify where the

71
features may cluster (i.e., where spatial autocorrelation is the strongest) and are
surrounded by similarly high or low attribute values. As such, I conducted a “hot spot
analysis” (which uses the Getis Ord Gi* statistic) to determine if the disamenities or
amenities cluster in minority and/or poor neighborhoods. This spatial statistic enables
researchers “to detect local ‘pockets’ of dependence that may not show up when using
global statistics,” (Getis & Ord, 1992, p. 189). In other words, the hot spot analysis
measures a feature in relation to its neighbors, sums the similarity of those features, and
compares it to the similarity of all the features in the study area. A hot spot occurs where
similarity is high. Getis-Ord Gi* can be calculated separately, or it can be calculated, as I
did, using the “Hot Spot Analysis with Rendering” tool, which identifies hot spots, as
well as returns Z-scores, confidence levels, and probabilities.

Analyzing the Relationship (Regression Analysis)

I initially selected geographically weighted regression (GWR) to analyze the
relationships among the hazards and amenities and Census data. GWR is an optimization
routine, which divides a study area into regions and specifies different regression
parameters for different parts of the study area. This technique is useful because of the
autocorrelation and spatial dependence (i.e., spatial nonstationarity) inherent in spatial
data: “In a hypothetical random spatial distribution, every feature or observation would
have an equal probability of occurring at any given position, and the position of any
given feature or observation would have no influence on any other feature or
observation,” (Mitchell, 2005, p. 66). Spatial dependence and spatial autocorrelation

72
violate the requirements for traditional regression techniques like ordinary least squares
(OLS). Such techniques can be and are used in environmental justice studies, but GWR
may be preferable because it accounts for spatial dependence and autocorrelation, and
assumes “that the regression parameters vary over space,” (Brunsdon, 2008, p. 347).
Comparatively speaking, OLS is considered a global regression technique (Scott & Pratt,
2009). However, “by providing only a ‘global’ measure for the entire space, standard
approaches such as ordinary least squares and (most) spatial econometric models tend to
compromise spatial heterogeneity in favor of average estimates and efficiency,” (Ali,
Partridge, & Olfert, 2007, p. 300). Global models return one regression coefficient for the
entire study area, while GWR returns separate regression parameters within a study area
that “reflect the sample heterogeneity by different marginal responses to an explanatory
variable across space,” (Ali, Partridge, & Olfert, 2007, p. 301).
GWR presents some issues for the researcher. First, because GWR partitions the
landscape or study area, small study areas may compromise the regression. Therefore,
due to the small number of Census tracts in the six smallest cities and six medium-sized
cities in this study, I was only able to conduct GWR on the 6 largest cities because it
requires a minimum of 100 observations to work properly (see Table 3.6). Second,
“GWR output can be overwhelming [because] the number of regression parameters
equals N x (K + 1), where N is the number of observations and K is the number of
explanatory variables,” (Ali, Partridge, & Olfert, 2007, p. 302). Thus, parsimonious
models are recommended, and to accomplish this, I used Excel to construct a correlation
matrix for each city to check for multicollinearity among the independent variables. This
resulted in different models for the six largest cities and the six medium-sized cities (see

73
Tables 3.7 and 3.8). The difference in models for big and medium cities is also justified
because large cities operate on a much larger scale, which may bias regression models
(Johnson & Neimann, 2004).

Table 3.6
Number of Census Tracts by City

Number of Tracts

Dropped from GWR
Albany 26 yes
Austin 184
Boston 157
Boulder 29 yes
Charleston 25 yes
Dayton 69
Decatur 32 yes
Flagstaff 15 yes
Grand Rapids 54 yes
Little Rock 47 yes
Miami 81 yes
Norfolk 85 yes
Philadelphia 381
Salt Lake City 59 yes
San Diego 297
San Jose 204
Santa Fe 25 yes
Seattle 126

Given the different strengths and weaknesses of OLS and GWR, many GIS
researchers suggest combining OLS and GWR (c.f., Mitchell, 2005). As such, I ran OLS
regressions first and then ran GWR regressions using different models for sites and size
categories. This created more work, but allowed for more confidence in the results. Each
model specified was run separately for each dependent variable in each city. I used the
models and variables for each type of regression so that comparisons could be made
across cities and across regression methods.

74
Table 3.7
Hazard Models Used for Six Largest Cities

Basic Model
(Model I)

Adjusted Basic Model
(Model II)

Advanced Model
(Model III)

Advanced Model Controlled
(Model IV)

PCT_MIN PCT_MIN
POP_DENSITY POP_DENSITY
†
POP_DENSITY
†

MED_INC_99 MED_INC_99
†
MED_INC 99
†
MED_INC 99
†

PCT_MFG PCT_MFG PCT_MFG PCT_MFG
OO_HS_VAL OO_HS_VAL OO_HS_VAL OO_HS_VAL
PCT_WHITE PCT_WHITE
PCT_BLACK PCT_BLACK
PCT_HISPAN PCT_HISPAN
PCT_OTHER PCT_OTHER
PCT_ASIAN
††
PCT_ASIAN
††

PCT_LDCV22
PCT_LDCV23

†
Adjusted to the natural log or square root where appropriate.
††
PCT_ASIAN was not included in the Austin models.

I began with OLS regression. The dependent variables were the number of sites
per tract (TRI_NUM and LQG_NUM) in each of their respective models. Due to high
correlations (greater than ±0.7) among some independent variables, I dropped TENURE,
PCT_ED_BA, and PCT_OWNED from both city groups. I also dropped OO_HS_VAL
from the medium-sized cities due to its correlation with two other variables. I began with
a basic model (Model I) for each city group, which was the most parsimonious model
possible. It included PCT_MIN as a proxy for the different racial and ethnic groups, the
independent variables, and POP_DENSITY as a control variable. In the adjusted basic
model (Model II), I adjusted POP_DENSITY and MED_INC_99, using the square or the
natural log where appropriate. In the advanced model (Model III), I dropped PCT_MIN
and added the respective racial and ethnic variables. In the advanced model controlled
(Model IV), I dropped POP_DENSITY and added two land use variables. The first was

75
the percentage of land used for high-intensity residential use (PCT_LDCV22) and the
other was the percentage of land used for industrial, commercial, or transportation uses
(PCT_LDCV23).

Table 3.8
Hazard Models Used for Six Meduim-sized Cities

Basic Model
(Model I)

Adjusted Basic Model
(Model II)

Advanced Model
(Model III)

Advanced Model Controlled
(Model IV)

PCT_MIN PCT_MIN
POP_DENSITY POP_DENSITY
†
POP_DENSITY
†

MED_INC_99 MED_INC_99
†
MED_INC 99
†
MED_INC 99
†

PCT_MFG PCT_MFG PCT_MFG PCT_MFG
PCT_WHITE PCT_WHITE
PCT_BLACK PCT_BLACK
PCT_HISPAN PCT_HISPAN
PCT_OTHER PCT_OTHER
PCT_LDCV22
PCT_LDCV23

†
Adjusted to the natural log or square root where appropriate.

Because of high Variance Inflation Factor numbers (VIF > 7.5) among the race variables
in many of the advanced OLS models, when I ran GWR regressions on the six largest
cities, I dropped the race variables and added PCT_MIN.
I followed the same procedure for parks and community gardens in the six largest
cities, beginning with OLS regressions and adding GWR regression. Unfortunately, too
few dependent variables exist in most medium-sized cities to run regressions. I specified
a different model for parks (see Table 3.9), again aiming for parsimony. I began with
PCT_MIN, MED_INC_99, and OO_HS_VAL. In the advanced models III and IV, I
added the race variables, and also included PCT_LDCV21 (low-intensity residential) and
PCT_LDCV22 (high-intensity residential).

76

Table 3.9
Park Models Used for Six Largest Cities

Basic Model
(Model I)

Adjusted Basic Model
(Model II)

Advanced Model
(Model III)

Advanced Model Controlled
(Model IV)

PCT_MIN PCT_MIN
MED_INC_99 MED_INC_99
†
MED_INC 99
†
MED_INC 99
†

OO_HS_VAL OO_HS_VAL OO_HS_VAL OO_HS_VAL
POP_DENSITY
†
POP_DENSITY
†

PCT_WHITE PCT_WHITE
PCT_BLACK PCT_BLACK
PCT_HISPAN PCT_HISPAN
PCT_OTHER PCT_OTHER
PCT_ASIAN
††
PCT_ASIAN
††

PCT_LDCV21
PCT_LDCV22

†
Adjusted to the natural log or square root where appropriate.
††
PCT_Asian will not be included in the Austin models.

The community garden model was similar to the parks model, but I dropped
OO_HS_VAL because home owners would be more likely to garden in their own yards
(see Table 3.10). I added PCT_TENURE because community gardens have a long time
horizon to start and maintain. I suspect that residents who have been in a neighborhood
for five or more years would be more likely to start or join a community garden.
Unfortunately, due to the small number of community gardens in Austin (7), San Diego
(12), and San Jose (19), I did not run regressions in those three cities. Although
community gardens have a long history in Rust Belt cities, I would argue that fewer
gardens exist in Sun Belt cities, particularly on in California, because land is more
plentiful and density lower so more people may have their own gardens. As well, both
land and water are quite expensive in the West, so community groups may not have the
resources to lease or purchase vacant lots or purchase water for the gardens.

77

Table 3.10
Community Garden Models Used for Six Largest Cities

Basic Model
(Model I)

Adjusted Basic Model
(Model II)

Advanced Model
(Model III)

Advanced Model Controlled
(Model IV)

PCT_MIN PCT_MIN
MED_INC_99 MED_INC_99
†
MED_INC 99
†
MED_INC 99
†

PCT_TENURE PCT_TENURE PCT_TENURE PCT_TENURE
POP_DENSITY
†
POP_DENSITY
†

PCT_WHITE PCT_WHITE
PCT_BLACK PCT_BLACK
PCT_HISPAN PCT_HISPAN
PCT_OTHER PCT_OTHER
PCT_ASIAN
††
PCT_ASIAN
††

PCT_LDCV21
PCT+LDCV22

†
Adjusted to the natural log or square root where appropriate.
††
PCT_Asian will not be included in the Austin models.

Bias and Error in the Study

My study is not immune from potential errors in data collection or processing.
The most vexing errors are those errors inherent in environmental data. Census data also
have limitations, as does GIS mapping.

Problems with Environmental Data

The majority of environmental justice studies rely on EPA data, though the lack
of comprehensive and reliable data vexes researchers. The EPA maintains three major
data sources and all have limitations. Common to all is that they were only mandated

78
within the past few decades and thus are young data sources, making longitudinal studies
very different. The TRI, sanctioned in 1986, is the most popular data source and monitors
650 chemicals manufactured, processed, or otherwise used and released in the United
States and U.S. territories (EPA, 2001). Releases include air and water releases, as well
as the “releases” that occur in the disposal of chemicals underground or above ground or
the transport of chemicals to other facilities. Not all industrial facilities must report.
Those subject to report have a particular Standard Industrial Classification (SIC) code,
have 10 or more full-time employees, and generate chemical waste exceeding 25,000
pounds annually.
11
Thus, most gas stations, dry cleaners, auto repair shops, and similar
businesses are exempt. Furthermore, required to report annually by July 1 or face a per
diem fine of up to $27,500, businesses (and federal facilities) monitor their own chemical
releases and then “self-report” to EPA with federal oversight done only randomly (Krieg,
1998) and with no independent verification (Scott, Cutter, Menzel, Ji, & Wagner, 1997).
The TRI’s design hinders its purpose. The TRI structure: limits the number of facilities
reporting; tracks only 650 out of about 85,000 chemicals in use (Davis & Webster, 2002);
focuses only on large-quantity releases; aggregates releases in yearly totals rather than
tracking daily, weekly, or monthly spikes (Bowen & Wells, 2002); does not provide
sophisticated monitoring equipment or expertise for firms to improve firms’ release
estimates; and has no comprehensive oversight of firms.
The RCRA pre-dated the TRI by 10 years, and requires the reporting of treatment,
storage, and disposal facilities that handled hazardous waste and waste-generating

11
For persistent, bioaccumulative, and toxic (PBT) chemicals, the threshold is 10,000 pounds annually.
Such chemicals include those like mercury, polychlorinated biphenyl, and dioxin compounds that persist in
the air, water, and land and thus, can bioaccumulate in the human food supply.

79
activity.
12
Like the TRI, however, the data are self-reported and are not subject to EPA
oversight or independent verification. And, similar to the TRI, reporting is done on an
aggregate basis; this often includes reporting chemical mixes or “soups” (Scott & Cutter,
2006) that fail to specify particular chemicals and toxicity levels. Furthermore, to reduce
the burden of paperwork on states, changes to the requirements in 2006 now exclude low-
level mixed waste from reporting requirements (EPA, 2007). Like the TRI, critics argue
that the RCRA reporting requirements allow for underreporting by firms, overlook illegal
emissions, and exempt many firms from reporting (c.f., Krieg, 1998). Unfortunately, a
suitable replacement for these problematic data sources does not exist.

Problems with Census Data

The U.S. Decennial Census is one of the most comprehensive, reliable, and
accepted sources of sociodemographic information available to researchers, in spite of its
many problems. First, being administered every 10 years means that data can be
anywhere from a few to nearly 10 years out of date (Liu, 2001; MacDonald, 2006). As
well, Census figures are based on samples of all U.S. households (i.e., those that
complete the long form) and, thus, provide an estimate of actual households, which is
subject to underreporting. Census data errors include missing data, duplicate or
incomplete addresses in the Master Address File, and an undercount of those living in
group quarters (National Research Council, 2004). Another undercounted group is
immigrants. Although the Census attempts to count all citizens, legal residents, and

12
This activity was originally reported through the now-defunct Biennial Reporting System.

80
undocumented residents, obtaining an accurate count of undocumented immigrants is
difficult because the Census does not ask respondents about legal status.
Operationalizing race and ethnicity variables is also problematic. First, these
categories and questions change every 10 years making longitudinal comparisons
difficult because the categories change, merge, disappear, and often do not account for all
possible races or ethnicities (Liu, 2001; Szasz & Meuser, 2000; Peters & MacDonald,
2004). Pulido argues that the Census fails to account for many mixed-race respondents
(2000) and “denies the fluid nature of racial categories (Pulido, Sidawi, & Vos, 1996).
More important, though race is biological, respondents often identify with a particular
racial category that may not be consistent with their biological race (Liu, 2001). And,
respondents self-identify ethnicity (Zimmerman, 1994). The Census data on race and
ethnicity, therefore, are inconsistent, often one-dimensional, and may not accurately
represent the racial make-up of communities.
Another concern is the inability of income data to reflect wealth (Liu, 2001;
Pastor, Morello-Frosh, & Sadd, 2005). Income data reflect before-tax figures, but wealth
is more complex, including home ownership and other financial and real assets. That is,
“It is not just the flow of income but the stock of assets that matters,” (Pastor, Morello-
Frosch, & Sadd, 2005, p. 133). Finally, the Census has governmental (e.g., cities,
counties) and statistical divisions (e.g., tracts, block groups, blocks). These divisions
make defining “community” very difficult. A Census block is an individual neighborhood
block in urban areas. Block groups are delineated groups of blocks, ranging from 1,500 to
3000 people (Sobotta, Campbell, & Owens, 2007). The tract, defined by local Census
committees, is approximately 4,000 people, and is supposed to be relatively homogenous,

81
though the population is often heterogeneous (Wu, Qiu, & Wang, 2005). Researchers
have long debated how well these units represent neighborhoods and their racial make-up
(c.f., Bullard, 1996; Sui, 1999; Rhodes, 2003). Like environmental data, Census data
create estimation and omission problems (Sexton & Adgate, 1999), contributing to
weaker, less generalizable results.

GIS and Uncertainty

GIS is an extremely powerful mapping system, but it is not a perfect
representation of the world. I made every attempt to represent the data in my study as
accurately as possible, but GIS representation is limited by inflexibility and cannot
perfectly discern the geographic complexities in the world. As well, human operators
may make mistakes in data capture and measurement. Users also make decisions about
real-world geography that are based on user values, which can vary from user to user.
I conducted this study according to the standards used in environmental equity
analysis and in GIS mapping of urban problems, given the limits of the environmental
and Census data. These standards are based on both theory and practical applications.

Validity and Reliability

The validity of several early environmental justice studies has been challenged
(c.f., Bowen, 1999; Liu, 2001). Validity “refers to the extent to which a specific
measurement provides data that relate to commonly accepted meanings of a particular

82
concept,” (Babbie, 1992, p. 135). Specifically, “the validity of an environmental equity
analysis depends heavily upon the choice of statistical methods/models and the correct
interpretation of the results,” (Liu, 2001, p. 58). The research design and methods I
employed in this study are based on the review of over 100 environmental justice studies.
Thus, I used the standard statistical analyses and dependent, independent, and control
variables of the most rigorous studies. I also modeled environmental equity based on the
logical relationship among the variables as seen in previous studies. Finally, I have used
multiple sources of data and employed the standard measurements for all variables.
This study can easily be replicated by any user following the methods discussed
here and detailed in the Appendix. The results should be similar, with only slight
variation in the geocoding and geolocating of amenity and disamenity sites, which are
subject to human interpretation of the exact latitude and longitude of each site. This
variation should not affect the results, and these procedures can be used to produce
consistent results over time in these and additional cities. I have used the standard
methods of collection for the hazard, Census, and land cover data. As described above, I
used multiple methods for collecting amenity sites because no standard method exists.
This may be an area where reliability is weak.
In the following chapter, I describe the results of the disamenity mapping and
spatial statistical analyses.

83
CHAPTER FOUR: ENVIRONMENTAL DISAMENITIES

But environmental injustice occurs whenever a community or a people
experiences a greater environmental burden than that of the majority
population.
Edwardo Rhodes, When Environmental Justice in America

In this chapter, I describe the results of the GIS analysis on the distribution of TRI
sites and LQGs in 17 cities. Santa Fe had no disamenities as I defined them. I treat the
hazards separately and have organized the discussion around the four questions described
in Chapter 3: (1) What is the distribution of the disamenities (i.e., features); (2) Do
features cluster; (3) Where do features cluster; and (4) Is a pattern evident in the
distribution of these features vis-à-vis the sociodemographic data (Mitchell, 2005)? As
per Chapter 3, the research question for this portion of the study is: Are minorities and
poor populations over-represented in neighborhoods where environmental disamenities
cluster? The hypothesis driving this portion of the analysis is: The greater the
percentage of poor and/or minority residents in a tract, the greater the number of
disamenities and the more unequal the distribution. Several GIS analyses and regressions
were done separately on TRI facilities and LQGs.
Table 4.1 shows the total number of hazards by type for the 17 cities. LQGs
outnumber TRI sites 471 to 285, for a total of 756 hazard sites. San Diego and Seattle
have the most hazard sites with 125 and 108. San Diego is also the biggest city, covering
372.39 square miles (land only). Flagstaff has only two hazard sites, yet it is far from the
smallest city in the study. Albany is the smallest, covering only 21.87 square miles.

84
Table 4.1
Total TRI and LQG Sites by City

TRI

LQG

Total

Albany 3 12 15
Austin 19 21 40
Boston 6 26 32
Boulder 9 8 17
Charleston 2 6 8
Dayton 25 25 50
Decatur 6 6 12
Flagstaff 1 1 2
Grand Rapids 36 27 63
Little Rock 10 12 22
Miami 13 29 42
Norfolk 7 8 15
Philadelphia 33 39 72
Salt Lake City 34 19 53
San Diego 28 97 125
San Jose 22 58 80
Seattle

31

77

108

Grand Rapids and Seattle had the most TRI facilities; however, Decatur and
Philadelphia led in total on- and off-site releases for reporting year 2006, reporting
5,584,944.74 and 1,146,199.61 pounds, respectively (see Table 4.2). The amount of
hazards in Decatur is not an anomaly or the result of an accidental release. For the
previous two reporting years, releases were actually higher. Charleston and Flagstaff had
the fewest releases, reporting 3,137.56 and 6,350 pounds, respectively. Decatur recorded
the highest mean release with 930,824.12 pounds, but Norfolk had the second-highest
mean release with 80,344.46 pounds. Although San Diego and Seattle had the most
LQGs, Boulder and San Jose recorded the most tons managed for reporting year 2005,
reporting 56,767.2 and 41,939.74 tons respectively. Flagstaff and Charleston reported the

85
fewest tons, with 38.06 and 482.32 respectively.
13
Boulder had the highest mean tonnage
(7,095.9), while Norfolk’s eight LQGs produced the next highest mean tonnage
(2,295.95). Clearly, Charleston and Flagstaff, two of the smaller cities in the study have a
small number of facilities and reported pollution.
Although Salt Lake City, San Diego, San Jose, and Seattle may not have reported
the most amount of releases or tons managed for reporting years 2006 and 2005, they
have the distinction of having tracts with more than 10 disamenities contained within
them. Salt Lake City’s Tract 1003.02 hosts 11 TRI facilities and 3 LQGs; though it is
77% white (20% minority, with 14% Hispanic), its median income is only $13,750 and
only 6% of the residents hold a Bachelor’s degree or better. Seattle’s Tract 109 hosts 7
TRI facilities and 9 LQGs, and is 40% minority (including 15% Hispanic and 14%
Asian), with a median income of $33,654 and only 18% of the residents have a
Bachelor’s degree or better. San Jose’s Tract 5050.06 has 5 TRI sites and 12 LQGs, and
is 47% minority (38% of which are Asian), but has a median income of $97,098 and 67%
of the residents boast a Bachelor’s degree or better. San Diego has the largest amount of
disamenities in one tract (See Figure 4.1). Six TRI sites and 14 LQGs are located in Tract
83.5, which is 60% minority—47% of which are Asian—and a median income of
$60,828 and Bachelor’s degree percentage of 37.

13
Recall that LQGs are defined as those generating 1,000 kilograms or more of hazardous waste monthly,
those generating more than 1 kilogram of acutely hazardous waste monthly, or a per month “acute spill”
greater than 100 kilograms.

86
Figure 4.1
San Diego’s Disamenities
Census Tracts
Percent Minority
5.00- 23.00
23.01 - 41.00
41.01 - 63.00
63.01 - 83.00
83.01 - 97.00
LQG
TRI
0 12,500 25,000 37,500 6,250
Meters
±
Tract 131: 14 LQGs and 6 TRI sites

87

Table 4.2
Total Facilities and Pollution Amounts

TRI

Releases in
Pounds

LQG

Tons Managed

Albany 3 8,610.90 12 7,916.91
Austin 19 131,703.32 21 3,838.94
Boston 6 44,499.80 26 526.05
Boulder 9 305,228.27 8 56,767.20
Charleston 2 3,137.56 6 482.32
Dayton 25 375,889.24 25 13,139.97
Decatur
a
6 5,584,944.74 6 890.20
Flagstaff 1 6,350.00 1 38.06
Grand Rapids 36 298,264.08 27 6,316.34
Little Rock 10 31,424.90 12 545.25
Miami 13 337,906.51 29 1341.291
Norfolk 7 562,411.25 8 2,364.72
Philadelphia 33 1,146,199.61 39 41,575.01
Salt Lake City 34 624,553.55 19 4,692.09
San Diego 28 436,969.75 97 15,526.04
San Jose 22 563,216.84 58 41,939.74
Seattle

31

124,257.80

77

18,356.62

a
Decatur’s TRI releases for 2001 and 2003 were 7,601,758 and 6,173,669, respectively.

As a general way of describing the distribution of the hazards in each city, I
calculated the mean center for each type of hazard. The mean center identifies where the
spatial center of a distribution is, based on the average of all the XY coordinates in the
feature class (e.g., TRI facilities, LQGs). A comparison of the mean centers for the
hazards sites revealed that in five cities (see Table 4.3) the mean centers for both TRI and
LQG hazards were located in the same tract. Little Rock is one such city (see Figure 4.2),
with a minority percentage of 97 and a median income of $16,692 in its mean center
track for TRI sites and LQGs. In an additional five cities (i.e., Boston, Charleston,
Dayton, San Diego, San Jose), the mean centers were in adjacent tracts with distances

88
ranging between less than .5 miles to 1.25 miles between them. This suggests that LQGs
and TRI sites tend to locate in similar locales.

Table 4.3
Cities with Hazard Mean Centers Located in Same Tract

Tract

% White

% Minority

Median Income

Austin

3.01

69

28

32,259
Decatur 9 41 58 18,333
Flagstaff
a
14 86 13 45,625
Little Rock 5 2 97 16,692
Seattle

93 41 53 42,208
a
Flagstaff has only one TRI facility and one LQG.

Further examination of the five cities listed in Table 4.2 reveals that 83.68% of land in
Seattle’s Tract 93 was dedicated to industrial, commercial, or transportation use, which
might be expected for a tract that hosted the mean center for both hazards. However, in
the other four mean center tracts, the percentage of land dedicated to industrial,
commercial, or transportation use was less than 33%.
Next I discuss the individual hazard types, along with the geostatistics and
regression analyses, starting with TRI facilities.

89
Figure 4.2
TRI and LQG Mean Centers in Little Rock
k j k j
Census Tracts
Percent Minority
3.00 - 20.00
20.01 - 38.00
38.01 - 66.00
66.01 - 83.00
83.01 - 97.00
k j
LQG Mean Center
k j
TRI Mean Center
±
0 7,250 14,500 21,750 3,625
Meters
Tract 5 houses both the TRI and the LQG
mean centers. It is 96% African American with
a median income of $16,692.

90
TRI Facilities

This section describes the results of the GIS analysis for TRI facilities. The total
number of TRI facilities with releases > 0 for reporting year 2006 is 285. I did not expect
that so few hazards would meet my selection criteria; I suspect that outsourcing and
deindustrialization, as well as lax company reporting and compliance follow-up by the
EPA, may explain the small number. I expected to see a preponderance of facilities in the
Rust Belt and am not surprised that Grand Rapids and Philadelphia are among the top
three cities (see Figure 4.3). Still, Rust Belt cities account for a little over one third (36%)
of the TRI sites, while Sun Belt cities account for one quarter of the TRI facilities.

Figure 4.3
TRI Sites by City
0
5
10
15
20
25
30
35
40
Grand Rapids
Salt Lake City
Philadelphia
Seattle
San Diego
Dayton
San Jose
Austin
Miami
Little Rock
Boulder
Norfolk
Decatur
Boston
Albany
Charleston
Flagstaff

In terms of sites per person, Salt Lake City and Grand Rapids have ratios of 1 site for
every 5,337 persons and 1 site for every 5,494 persons. Boston has the best ratio at 1 site

91
for every 98,190 persons. (Of course, Santa Fe has no TRI sites, so technically it has the
best ratio of the 18 cities in the study.)
Table 4.4 identifies the tract with the biggest-producing TRI facility in each of the
17 cities, along with racial, income, and education data for the host tract. Albany, Little
Rock, Norfolk, San Diego, San Jose, and Seattle all had predominantly minority
populations in their tract with the single biggest producer. As well, Albany, Dayton,
Decatur, Little Rock, Norfolk, and San Diego’s biggest producers were in tracts with
median incomes of less than $30,000. Finally, Albany, Dayton, Decatur, Little Rock,
Norfolk, San Diego, and Seattle’s biggest producers were in tracts where less than 15%
of the residents had a Bachelor’s degree or better. Those cities fitting all three
qualifications above were Albany, Little Rock, Norfolk, and San Diego. Note that
African Americans were the majority minority population in all but San Diego. Also, San
Diego’s tract with the biggest producer was also host to 3 other TRI sites and 10 LQGs.
A racial and ethnic pattern that emerges from examining the tracts hosting or
adjacent to the TRI sites with the biggest releases is the fact that African American
majorities exist in the host tracts in the eastern and southeastern cities. For western or
northwestern cities, Hispanic and Asian majorities exist in the host tracts.

92
Table 4.4
TRI Facility with Largest Release

Largest
Release in
Pounds

Tract

%
White

%
Minority

Median
Income

% BA
or
Better

Albany

8,572.70 26 39 56.89
a
25,600 10
Austin 47,340.70 18.52
†
59 39 40,033 43
Boston 26,251.00 606 88 10 57,292 41
Boulder 242,927.29 127.05 86 13 51,087 58
Charleston 226.30 21 87 12 55,573 47
Dayton 230,058.00 807 98 2 28,262 3
Decatur 4,811,617.20 21 88 10 25,317 5
Flagstaff 6,350.00 14 86 13 45,625 38
Grand Rapids 48,561.20 22 69 27 30,975 26
Little Rock 23,614.22 40.01 23 75
b
23,021 9
Miami 224.743 14.02
††
3 68
c
15,701 3
Norfolk 354,451.30 46 2 97
d
11,929 3
Philadelphia 503,804.10 58 --- --- --- ---
Salt Lake City 174,215.50 1004 65 33 39,512 10
San Diego 305,740.00 50 4 95
e
22,802 3
San Jose 371,312.30 5043.18 35 62
f
52,065 11
Seattle

19,000.00 112

34

61
g

30,917

5

†
Facility is just outside tract, less than ½ mile away.
††
Facility is outside of tract, five miles away.
a
African Americans make up 48% of minority population.
b
African Americans make up 74% of the minority population.
c
African Americans make up 59% of the minority population.
d
African Americans make up 96% of the minority population.
e
Hispanics make up 91% of the minority population.
f
Hispanics make up 30% and Asians make up 28% of the minority population.
g
Hispanics make up 37% and Asians make up 14% of the minority population.

Do TRI Sites Cluster?

The geostatistic used to determine if sites cluster is Moran’s I. This index is a
global statistic that identifies the spatial autocorrelation of sites, though it does not
determine where sites cluster, by measuring the similarity of nearby features and
identifying the probability that the patterns are not due to a random distribution. Both an
observed index and an expected index are calculated for Moran’s I, the expected index

93
being a hypothetically random distribution. The tool returns Z scores based on the
variance of the expected index score and the observed index. Unfortunately, this spatial
autocorrelation tool needs a minimum of three features (i.e., sites) to run, and the best
results are obtained on feature classes of 30 or more. The results of the Moran’s I analysis
on TRI sites are displayed in Table 4.5. In all, sites were clustered in 11 of 14 cities
where the tool successfully ran (i.e., clustering was observed in 79% of cities). The most
significant clustering was observed in Austin (see Figure 4.4), Philadelphia, Salt Lake
City, and Seattle, followed by strong clustering in Dayton and Grand Rapids. Sites were
randomly distributed in Boston, and only somewhat clustered, though not significantly, in
Decatur and Little Rock.

Table 4.5
Moran’s Index for TRI Sites

Moran’s I

Z Score

Pattern

Albany






Austin 0.93*** 5.61 Clustered
Boston -0.21 -0.03 Random
Boulder 0.85* 2.31 Clustered
Charleston
  
Dayton 0.27** 3.02 Clustered
Decatur 0.25 1.44 Somewhat clustered
Flagstaff
  
Grand Rapids 0.44** 2.96 Clustered
Little Rock 0.27 1.64 Somewhat clustered
Miami 0.52** 2.59 Clustered
Norfolk 0.44* 2.54 Clustered
Philadelphia 1.03*** 5.04 Clustered
Salt Lake City 0.37*** 4.85 Clustered
San Diego 0.68*** 4.25 Clustered
San Jose 0.66* 2.23 Clustered
Seattle 0.50*** 6.18 Clustered

*p < .05, **p < .01, ***p < .001

94
Figure 4.4
Global Clustering of Austin’s TRI Sites

Global Moran's I Summary
Moran's Index: 0.932252
Expected Index: -0.055556
Variance: 0.031047
Z Score: 5.606108
p-value: 0.000000

As discussed in Chapter 3, Ripley’s K is a global statistic that can reveal the
clustered or dispersed pattern of a dataset, i.e., the spatial dependence in the dataset.
ArcGIS generates an expected K, an observed K, a differential K, and lower and higher
confidence levels at each distance. To determine if clustering and/or dispersion are
present and significant, I examined the differential K value to identify the largest positive
and negative values (i.e., the largest difference between the expected K and the observed
K values). At those values, or distance in meters in this case, I compared the observed K

95
value to the higher or lower confidence level to determine if clustering or dispersion was
significant. Where data points cluster, they are significant only when the observed K is
greater than the higher confidence level. Conversely, where data points are dispersed,
they are significant only when the observed K is lower than the lower confidence level.
Recall that I opted for 99 permutations, resulting in a probability level of .01.
The Ripley’s K for TRI facilities was successful for 13 of 17 cities. The statistic
needs a minimum of three data points to run successfully. Of the 13 cities, statistically
significant clustering of TRI sties was found in Dayton, Grand Rapids, Miami, San
Diego, San Jose, and Seattle (see Table 4.6). The six cities trended toward clustering at
shorter distances, as close as 1,178 meters (just under ¾ of a mile), with an average
distance of 4,336 meters (2.7 miles). This is not surprising given that industries tend to
locate near other industries or in industrial zones.

Table 4.6
Ripley’s K Cluster Analysis for TRI Sites
Expected K Observed K Differential K
Lower
Confidence
Higher
Confidence
Albany error
Austin 1830.983 2723.947 892.9642 0 3414.648
Boston 870.551 964.0213 93.47027 0 1472.567
Boulder 468.113 547.3835 79.27047 0 774.1172
Charleston error
Dayton 4613.328 8285.389* 3672.111 3664.864 5922.028
Decatur error
Flagstaff error
Grand Rapids 2584.214 3387.922* 803.7081 2227.024 3216.269
Little Rock 327.9137 0 -327.9137 0 4417.755
Miami 908.7244 1416.308 507.5839 0 2002.962
Norfolk 1275.807 1351.196 75.38816 0 1910.879
Philadelphia 1349.357 1973.947 624.5899 854.744 2151.058
Salt Lake City 2080.533 2530.604 450.0708 1592.137 2708.85
San Diego 3865.56 7018.426* 3152.866 2361.443 4820.275
San Jose 571.5591 1178.097* 606.5374 0 870.082
Seattle 868.6605 1812.987* 944.3267 0 1465.995
*p < .01

96

Statistically significant dispersion of TRI sties was observed only in Little Rock (see
Table A.1 in the Appendix). The results reveal a weakness with the Ripley’s K statistic: it
is extremely sensitive to study area. Thus, results may be skewed in smaller cities with
fewer data points.

Where Do TRI Sites Cluster?

The local statistic Getis-Ord Gi* identifies where clusters exist and measures the
extent or strength of clustering. The statistic returns a Z-score and probability value for
each unit of analysis (here, each Census tract). A high positive Z score (Z ≥ 1.96)
indicates a hot spot or high level of attribute values, which is why this statistic is often
called “hot spot analysis.” A high negative Z score (Z ≥ -1.96) indicates a “cold spot” or
low level of attribute values. Basically, the higher the Z score (either positive or
negative), the stronger the clustering, and generally speaking, high Z scores are often
returned for adjacent Census tracts. Because a Z score is returned for each Census tract,
the tables and discussion in this chapter will only review the Z-scores that are greater than
or equal to ±2.56 and a confidence level of .01 or better. See the Appendix for tables that
include Z-scores of ±1.96 and a confidence level of .05 or better.
Table 4.7 displays the results of the hot spot analysis on TRI locations in the
study, along with the percentage of minority residents and median income for the tracts.
Twelve of the cities exhibited one or more tracts where TRI sites clustered. Albany,
Boston, Charleston, and Flagstaff exhibited no TRI hot spots. The strongest clustering

97
was seen in Philadelphia, San Diego, San Jose (see Figure 4.5), and Seattle. In Austin,
Little Rock, Norfolk, and especially San Jose and Seattle, clustering was evident in tracts
with greater than 50% minority residents. Nine of 10 tracts in San Jose ranged from 51%
minority to 78% minority. In Seattle, 5 of 9 tracts with TRI hot spots had minority
percentages from 60% to 86%. Norfolk only had one tract designated a TRI hot spot, but
that tract was 98% minority with a median income of $12,813. Other cities with hot spots
in low median income tracts included: Little Rock (one tract with median income =
$25,838); Salt Lake City (one tract = $13,750); and Seattle (one tract = $16,285).
Surprisingly, San Jose exhibited very strong clustering in minority tracts, but median
income associated with those tracts ranged from $45,057 to $98,875 in tracts with 50% or
more minority population. The high values for median income could be due to San Jose’s
economy being dependent on high-tech industries, which pay well, but are still
categorized as toxic-producing enterprises. (See Table A.5 for expanded Z-scores.)

Table 4.7
TRI “Hot Spots” or Getis-Ord Gi* Results

City

Tract

Gi* Z Score

Gi* p Value

Percent
Minority

Median
Income

Austin 24.17 2.72 0.006 68 41,369
22.05

3.28

0.001

79

37,111

Boulder 127.08 2.90 0.004 7 87,797
127.05

2.82

0.005

13 51,087
Dayton 1003.02

2.68

0.007

10 44,720
Decatur 22

3.24

0.001

9 45,693
Grand Rapids 133 3.22 0.001 27 41,290
27 2.50 0.012 28 32,607
39

2.81

0.005

80 35,640
Little Rock 40.07

3.82

0.000

66 25,838

98
Table 4.7: Continued

Norfolk 52

2.65

0.008

98 12,813
Philadelphia 43 3.05 0.002 0 0
68 3.07 0.002 0 0
360 5.16 0.000 19 45,539
361 3.32 0.001 19 51,513
324 3.05 0.002 0 41,250
328 3.75 0.000 86 56,250
354 4.02 0.000 38 0
359

3.25

0.001

13

43,582

Salt Lake City 1003.02 3.00 0.003 20 13,750
1139.01 2.79 0.005 21 37,348
1003.03

3.91

0.000

16 69,250
San Diego 83.33 3.19 0.001 27 127,271
83.58 2.98 0.003 57 49,976
95.02 3.53 0.000 24 58,869
83.4 2.48 0.013 26 64,554
85.11 2.69 0.007 32 34,191
83.51 2.68 0.007 58 60,223
94 5.14 0.000 40 38,796
95.06 2.92 0.004 21 63,953
83.6 2.99 0.003 54 52,867
83.46

2.68

0.007

44

99,718

San Jose 5050.05 5.71 0.000 51 71,667
5045.05 3.56 0.000 72 79,259
5043.1 3.02 0.003 78 78,531
5044.1 3.81 0.000 72 78,501
5043.15 2.94 0.003 70 84,764
5050.07 3.43 0.001 65 74,911
5003 2.62 0.009 68 45,057
5044.11 3.43 0.001 65 85,949
5050.06 3.56 0.000 47 97,098
5043.2

3.21

0.001

72

98,875

Seattle 109 3.77 0.000 40 33,654
100 3.28 0.001 71 37,122
112 4.30 0.000 61 30,917
110 3.28 0.001 86 36,754
113 4.02 0.000 43 46,838
104 5.07 0.000 77 48,697
108 5.45 0.000 49 53,198
265 4.55 0.000 60 16,285
264 4.55 0.000 41 40,291

99
Figure 4.5
San Jose’s TRI Hot Spots
City Boundary
TRI_NUM
GiZScore
< -2.0
-2.0 to -1.0
-1.0 to 1.0
1.0 to 2.0
> 2.0
0 8,500 17,000 25,500 34,000 4,250
Meters

100
Only two cities exhibited TRI cold spots; Albany had only one tract with percent
minority equal to 19 and a median income of $36,371 (see Table 4.8). Salt Lake City’s
results were more interesting. Of the city’s 59 tracts, 25 were returned as cold spots with
Z scores ≥ -2.56 (the number of tracts increases to 38 tracts when considering Z ≥ -1.96).
The 24 tracts are clustered in the southeast portion of the city, representing a mix of high-
intensity residential land use and owner-occupied housing values ranging from $85,800-
$691,700. The percentage of white residents in these tracts ranged from 46% to 95%,
with 16 tracts exceeding 75%. (See Table A.7 for expanded Z-scores.)

Table 4.8
TRI “Cold Spots” or Getis-Ord Gi* Results

City

Tract

Gi* Z Score

Gi* p Value

Percent
Minority

Median
Income

Albany 16 -2.79 0.005 19 36371

Salt Lake City 1015 -3.65 0.000 14 26,717
1046 -2.91 0.004 14 38,000
1041 -2.96 0.003 5 70,039
1035 -3.42 0.001 11 49,213
1030 -4.21 0.000 36 32,067
1049 -2.58 0.010 23 33,056
1029 -3.24 0.001 42 24,375
1032 -3.42 0.001 25 27,222
1036 -3.24 0.001 3 68,929
1033 -3.24 0.001 14 38,487
1019 -3.01 0.003 20 25,938
1020 -3.91 0.000 42 24,385
1037 -3.07 0.002 5 61,463
1040 -2.58 0.010 6 57,007
1017 -3.43 0.001 21 27,568
1018 -3.91 0.000 24 33,481
1034 -3.42 0.001 10 42,286
1031 -3.42 0.001 29 37,925
1023 -3.42 0.001 39 16,048
1024 -2.87 0.004 51 23,125
1014 -2.58 0.010 21 22,778
1021 -2.60 0.009 25 16,978
1022 -2.60 0.009 12 21,131
1016 -3.91 0.000 14 31,420

101
Generally speaking, the results suggest that TRI facilities cluster more frequently
than not, particularly when the results are expanded to include Z ≥ -1.96 (p = .05). When
expanded, TRI facilities clustered in 71% of cities in the study, with the percentage of
minority residents as high as 98% in some tracts and a median income level as low as
$12,813. Only four cities showed TRI cold spots when results were expanded.

Is a Pattern Evident?

This section presents the results of the regression models described in Chapter 3.
Table 4.9 displays the results for all models run on the six largest cities, noting model
performance and significance and significant variables. Only the OLS results for the best-
performing models in the six largest cities are included here—all other regression results
are included in the Appendix. Adjusted R
2
is used to measure how much of the model
predicts the variation in TRI sites. However, caution should be used in comparing the
Adjusted R
2
of the OLS model to the Adjusted R
2
of the GWR model. AIC (Akaike’s
Information Criterion) is a relative measure of model performance. It is not a goodness of
fit measure; rather, it is based on the independent variables in the model and can be used
to compare one model’s performance to another’s. Lower AIC numbers indicate a better-
performing model, relatively speaking. The AIC of the OLS model may be compared
with the AIC of the GWR model. Another statistic returned in the OLS model is the
Variance Inflation Factor (VIF). In a properly specified model, this number should be less
than 7.5; a VIF greater than 7.5 indicates serious multicollinearity among the explanatory

102
variables. GIS also returns a map of the residuals. Moran’s I can be used to test
autocorrelation of the residuals; a random distribution of residuals is desired.

103

Table 4.9
TRI Regression Results for Six Largest Cities
Austin Boston Philadelphia San Diego San Jose Seattle
Model I
Performance AIC =95, R
2
=.02 AIC =-73, R
2
=.04* AIC =258, R
2
=.10* AIC =524, R
2
=-.008 AIC =319, R
2
=.04*† AIC =329, R
2
=.05*
Variables PCT_MFG*(+) PCT_MIN*(-) PCT_MIN**(-) OO_HS_VAL**(-) PCT_MFG*(+)
PCT_MFG*(+) MED_INC_99*(-)
OO_HS_VAL***(-)
POP_DENSITY***(-)
Model II
Performance AIC =89, R
2
=.05 Too few sites AIC =246, R
2
=.13* AIC =504, R
2
=.05* AIC =309, R
2
=.09*† AIC =317, R
2
=.14
Variables OO_HS_VAL***(-) LOG_DENSITY***(-) PCT_MFG*(+) LOG_DENSITY*(-)
SQ_POP_DEN***(-) OO_HS_VAL**(-)
SQ_MED_INC*(-) LOG_DENSITY***(-)
Model III††
Performance AIC =86, R
2
=.08* Too few sites AIC =248, R
2
=.14* AIC =506, R
2
=.06* AIC =315, R
2
=.08*† AIC =287, R
2
=.34*†
Variables PCT_MFG***(+) OO_HS_VAL***(-) LOG_DENSITY***(-) OO_HS_VAL**(-) PCT_HISPAN*(+)
LOG_DENSITY*(-) SQ_POP_DEN**(-) LOG_DENSITY***(-) LOG_DENSITY*(-)
Model IV††
Performance AIC =86, R
2
=.09 Too few sites AIC =267, R
2
=.09* AIC =518, R
2
=.02 AIC =316, R
2
=.08* AIC =286, R
2
=.35*†
Variables PCT_WHITE*(+) OO_HS_VAL**(-) PCT_LDCV23***(-) OO_HS_VAL*(-) PCT_HISPAN*(+)
PCT_BLACK*(+) PCT_LDCV23*(-) PCT_LDCV22*(-) MED_INC_99*(+)
PCT_HISPAN**(+) PCT_LDCV23**(-) PCT_LDCV23***(+)
PCT_OTHER*(+)
PCT_MFG*(+)
PCT_LDCV22**(-)
GWR
Performance AIC =86, R
2
=.10 Too few sites AIC =3691, R
2
=.68 AIC =85, R
2
=.10 error AIC =263, R
2
=.50

*p > .05, **p > .01, ***p > .001; †High Clustering of Residuals; ††VIF for Race Variables > 7.5

104

In examining the results of the four OLS models and the GWR model for the six
largest cities, it is evident that in Austin, San Diego, and Seattle, the GWR returned the
lowest AIC numbers. Looking specifically at the OLS models, Model III returned the best
results in all but Seattle, where Model IV worked best (see Tables 4.10-4.15). However,
Models III and IV in all cities have very large VIF numbers and thus, the results cannot
be trusted. The large numbers returned for the race variables indicate that they were
interacting with other the other variables. What Table 4.9 shows is that the Models III
and IV were mis-specified. The breakdown of racial and ethnic categories should be
expunged from the models and a separate regression should be run on just the racial and
ethnic categories. In every city, density (inversely related) was a key factor in explaining
TRI site locations. The percentage of people working in manufacturing was positively
related in only four cities, which was expected. Median income and owner-occupied
housing value were inversely related in three cities, which was surprising in that I
expected that these variables would be significant in most cities. I expect that running
these models without the racial and ethnic variables would return stronger models and
more significant variables. Clustering of the residuals was found in San Jose and Seattle,
which may mean that some other explanatory factors were at work. Running Getis-Ord
Gi* on all variables in those locations may indicate what factor(s) I may be missing.

105
Table 4.10
TRI/OLS Regression Results for Austin Model III
Variable Coefficient StdError t-Statistic Probability Robust_SE Robust_t Robust_Pr VIF [1]
Intercept -3.567836 2.681826 -1.330376 0.185132 1.784085 -1.999813 0.047063* --------
PCT_WHITE 0.038583 0.027236 1.416606 0.158385 0.018342 2.103547 0.036840* 993.092707
PCT_BLACK 0.036008 0.027122 1.327623 0.186038 0.017863 2.015846 0.045341* 243.017481
PCT_HISPAN 0.041527 0.027147 1.529734 0.127899 0.01851 2.243545 0.026105* 602.197802
PCT_OTHER 0.034179 0.028964 1.180041 0.239585 0.01941 1.76088 0.080011 36.929316
PCT_MFG 0.016553 0.006149 2.691816 0.007793* 0.00802 2.06408 0.040479* 2.337109
OO_HS_VAL 0 0 1.029922 0.30446 0.000001 0.798216 0.425818 3.256028
SQ_INCOME -0.001339 0.000792 -1.690663 0.09269 0.000868 -1.541627 0.124982 4.476005
LOG_DENSITY -0.054945 0.026608 -2.065032 0.040388* 0.034231 -1.605125 0.11028 1.234222

Number of Observations: 184 Number of Variables: 9
Degrees of Freedom: 175 Akaike's Information Criterion (AIC) [2]: 86.486661
Multiple R-Squared [2]: 0.120715 Adjusted R-Squared [2]: 0.080519
Joint F-Statistic [3]: 3.003175 Prob(>F), (8,175) degrees of freedom: 0.003504*
Joint Wald Statistic [4]: 14.926828 Prob(>chi-squared), (8) degrees of freedom: 0.060584
Koenker (BP) Statistic [5]: 13.077745 Prob(>chi-squared), (8) degrees of freedom: 0.109202
Jarque-Bera Statistic [6]: 2412.690396 Prob(>chi-squared), (2) degrees of freedom: 0.000000*

Table 4.11
TRI/OLS Regression Results for Boston Model I
Variable Coefficient StdError t-Statistic Probability Robust_SE Robust_t Robust_Pr VIF [1]
Intercept 0.172432 0.081983 2.103255 0.037095* 0.087514 1.970329 0.050627 --------
PCT_MIN -0.001459 0.000634 -2.303456 0.022608* 0.000669 -2.180761 0.030740* 1.681294
MED_INC_99 -0.000003 0.000001 -1.929965 0.055485 0.000002 -1.484234 0.139841 2.046686
PCT_MFG 0.010515 0.004996 2.104796 0.036958* 0.005958 1.764959 0.079598 1.113164
OO_HS_VAL 0 0 0.044853 0.964276 0 0.037771 0.969913 1.589545
POP_DENSITY -0.000424 0.000253 -1.674286 0.096153 0.000213 -1.98706 0.048721* 1.04626

Number of Observations: 157 Number of Variables: 6
Degrees of Freedom: 151 Akaike's Information Criterion (AIC) [2]: -73.411075
Multiple R-Squared [2]: 0.075446 Adjusted R-Squared [2]: 0.044832
Joint F-Statistic [3]: 2.464415 Prob(>F), (5,151) degrees of freedom: 0.035319*
Joint Wald Statistic [4]: 6.214197 Prob(>chi-squared), (5) degrees of freedom: 0.285931
Koenker (BP) Statistic [5]: 13.544323 Prob(>chi-squared), (5) degrees of freedom: 0.018778*
Jarque-Bera Statistic [6]: 2551.383034 Prob(>chi-squared), (2) degrees of freedom: 0.000000*

106
Table 4.12
TRI/OLS Regression Results for Philadelphia Model III
Variable Coefficient StdError t-Statistic Probability Robust_SE Robust_t Robust_Pr VIF [1]
Intercept 0.703534 0.104495 6.732709 0.000000* 0.246125 2.858442 0.004501* --------
PCT_WHITE -0.001793 0.001544 -1.161213 0.2463 0.003289 -0.545236 0.585926 10.989275
PCT_BLACK -0.002587 0.001442 -1.794338 0.073578 0.003077 -0.840654 0.401071 9.988904
PCT_HISPAN -0.002848 0.001937 -1.470379 0.14232 0.003225 -0.8833 0.377633 2.671299
PCT_ASIAN 0.002619 0.002934 0.892783 0.372539 0.004824 0.542969 0.587485 1.422871
PCT_OTHER 0.028159 0.047099 0.597873 0.550293 0.057413 0.490471 0.624102 1.024124
PCT_MFG -0.003676 0.003539 -1.038497 0.299706 0.004204 -0.874336 0.382488 1.50922
OO_HS_VAL -0.000001 0 -2.133676 0.033513* 0 -3.176818 0.001626* 1.417425
SQ_POP_DEN -0.029389 0.006444 -4.56056 0.000009* 0.009674 -3.037896 0.002560* 1.359692
SQ_MED_INC -0.000807 0.000461 -1.751136 0.080756 0.000642 -1.255716 0.210014 1.967586

Number of Observations: 381 Number of Variables: 10
Degrees of Freedom: 371 Akaike's Information Criterion (AIC) [2]: 248.659738
Multiple R-Squared [2]: 0.15555 Adjusted R-Squared [2]: 0.135065
Joint F-Statistic [3]: 7.593274 Prob(>F), (9,371) degrees of freedom: 0.000000*
Joint Wald Statistic [4]: 20.670801 Prob(>chi-squared), (9) degrees of freedom: 0.014195*
Koenker (BP) Statistic [5]: 38.296664 Prob(>chi-squared), (9) degrees of freedom: 0.000015*
Jarque-Bera Statistic [6]: 8061.704524 Prob(>chi-squared), (2) degrees of freedom: 0.000000*

Table 4.13
TRI/OLS Regression Results for San Diego Model III
Variable Coefficient StdError t-Statistic Probability Robust_SE Robust_t Robust_Pr VIF [1]
Intercept -0.931153 2.642283 -0.352405 0.724805 1.12839 -0.825205 0.409929 --------
PCT_WHITE 0.014436 0.027643 0.522223 0.601927 0.013032 1.10775 0.268892 603.85798
PCT_BLACK 0.006894 0.028794 0.239427 0.810947 0.012238 0.56333 0.573656 60.072442
PCT_HISPAN 0.016773 0.026895 0.623645 0.533357 0.01319 1.271644 0.204534 379.406511
PCT_ASIAN 0.018746 0.028889 0.648906 0.516917 0.014888 1.25916 0.209 120.902527
PCT_OTHER 0.055714 0.05329 1.045496 0.296665 0.064188 0.867978 0.386119 1.558558
PCT_MFG -0.001726 0.00849 -0.203268 0.839067 0.007142 -0.241634 0.809239 1.606869
OO_HS_VAL 0 0 -1.446029 0.149272 0 -1.68334 0.093406 2.1966
SQ_INCOME 0.000045 0.000808 0.056261 0.955161 0.000521 0.087215 0.930548 2.289648
LOG_DENSITY -0.100461 0.021289 -4.719038 0.000005* 0.040302 -2.492705 0.013231* 1.098249

Number of Observations: 297 Number of Variables: 10
Degrees of Freedom: 287 Akaike's Information Criterion (AIC) [2]: 506.894076
Multiple R-Squared [2]: 0.09089 Adjusted R-Squared [2]: 0.062381
Joint F-Statistic [3]: 3.188141 Prob(>F), (9,287) degrees of freedom: 0.001080*
Joint Wald Statistic [4]: 11.807521 Prob(>chi-squared), (9) degrees of freedom: 0.224379
Koenker (BP) Statistic [5]: 14.706272 Prob(>chi-squared), (9) degrees of freedom: 0.099327
Jarque-Bera Statistic [6]: 67891.1505 Prob(>chi-squared), (2) degrees of freedom: 0.000000*

107
Table 4.14
TRI/OLS Regression Results for San Jose Model III
Variable Coefficient StdError t-Statistic Probability Robust_SE Robust_t Robust_Pr VIF [1]
Intercept -1.686026 3.793337 -0.44447 0.657209 3.443022 -0.489694 0.624913 --------
PCT_WHITE 0.025814 0.039668 0.65075 0.515977 0.036196 0.71317 0.476593 725.162925
PCT_BLACK 0.049816 0.052188 0.954546 0.340985 0.052079 0.956549 0.339975 6.735111
PCT_HISPAN 0.020326 0.038442 0.52875 0.59759 0.034029 0.597307 0.551002 505.173049
PCT_ASIAN 0.022837 0.039274 0.581472 0.561601 0.03438 0.66426 0.50731 375.195958
PCT_OTHER -0.021784 0.065649 -0.331823 0.740391 0.062155 -0.350479 0.72637 1.247772
PCT_MFG 0.009442 0.01014 0.931235 0.352879 0.019098 0.494427 0.621573 4.39849
OO_HS_VAL -0.000001 0 -2.806244 0.005524* 0.000001 -2.014771 0.045304* 2.21107
SQ_INCOME -0.000079 0.001398 -0.056473 0.955013 0.000889 -0.088809 0.929314 2.504823
LOG_DENSITY -0.091107 0.027262 -3.341931 0.001009* 0.039148 -2.327242 0.020973* 1.030019

Number of Observations: 204 Number of Variables: 10
Degrees of Freedom: 194 Akaike's Information Criterion (AIC) [2]: 315.140298
Multiple R-Squared [2]: 0.119257 Adjusted R-Squared [2]: 0.078398
Joint F-Statistic [3]: 2.918743 Prob(>F), (9,194) degrees of freedom: 0.002891*
Joint Wald Statistic [4]: 10.0411 Prob(>chi-squared), (9) degrees of freedom: 0.347168
Koenker (BP) Statistic [5]: 14.597487 Prob(>chi-squared), (9) degrees of freedom: 0.102602
Jarque-Bera Statistic [6]: 16489.6256 Prob(>chi-squared), (2) degrees of freedom: 0.000000*

Table 4.15
TRI/OLS Regression Results for Seattle Model IV
Variable Coefficient StdError t-Statistic Probability Robust_SE Robust_t Robust_Pr VIF [1]
Intercept -6.478036 6.15522 -1.052446 0.294796 5.253093 -1.233185 0.220023 --------
PCT_WHITE 0.051917 0.064461 0.8054 0.422244 0.053337 0.973374 0.332403 499.862821
PCT_BLACK 0.043978 0.066954 0.656835 0.512598 0.053905 0.815846 0.416267 103.269252
PCT_HISPAN 0.150751 0.070356 2.142683 0.034241* 0.068447 2.202471 0.029620* 20.241021
PCT_ASIAN 0.052017 0.068042 0.764487 0.446137 0.05772 0.901199 0.369357 165.498543
PCT_OTHER 0.214619 0.108686 1.974659 0.050702 0.159899 1.342214 0.182176 2.987323
MED_INC_99 0.00002 0.000007 2.797162 0.006045* 0.000008 2.567038 0.011535* 3.119004
PCT_MFG 0.010158 0.025257 0.402196 0.688295 0.021327 0.47632 0.634758 2.225079
OO_HS_VAL -0.000001 0.000001 -0.886783 0.377038 0.000001 -1.054332 0.293935 2.070704
PCT_LDCV22 0.004874 0.015802 0.308454 0.758301 0.019468 0.250362 0.802757 1.104657
PCT_LDCV23 0.012965 0.003114 4.163022 0.000065* 0.005022 2.581461 0.011090* 1.594464

Number of Observations: 126 Number of Variables: 11
Degrees of Freedom: 115 Akaike's Information Criterion (AIC) [2]: 286.483825
Multiple R-Squared [2]: 0.401911 Adjusted R-Squared [2]: 0.349904
Joint F-Statistic [3]: 7.727923 Prob(>F), (10,115) degrees of freedom: 0.000000*
Joint Wald Statistic [4]: 37.854341 Prob(>chi-squared), (10) degrees of freedom: 0.000040*
Koenker (BP) Statistic [5]: 30.636796 Prob(>chi-squared), (10) degrees of freedom: 0.000673*
Jarque-Bera Statistic [6]: 1973.112729 Prob(>chi-squared), (2) degrees of freedom: 0.000000*

108
Turning to the six medium-sized cities, the results were surprising (see Table
4.16).
14
Due to the small number of dependent variables, Models I and II performed
better in each city, the former performing well in Dayton, Little Rock, and Norfolk.
Again, population density (inversely related) was a key variable in all cities, while
PCT_MFG played a key role in Dayton, Grand Rapids, and Little Rock. Median income
(inversely related) was also a key variable in Dayton and Salt Lake City. The racial and
ethnic variables had high VIF values, indicated that these models were mis-specified.
Examining the other geostatistics for example, revealed that race may indeed explain the
variation of sites in several cities, so separate regressions should be run.

14
Miami is absent from these results because no TRI sites were located within any Census tracts.

109

Table 4.16
TRI Regression Results for Five Medium-sized Cities
Dayton Grand Rapids Little Rock Norfolk Salt Lake City
Model I
Performance AIC =150, R
2
=.10* AIC =107, R
2
=.14* AIC =57, R
2
=.26* AIC =29, R
2
=.14*† AIC =228, R
2
=-.03†
Variables MED_INC_99**(-) POP_DENSITY**(-) PCT_MFG**(+) POP_DENSITY***(-)
PCT_MFG*(+)
Model II
Performance AIC =144, R
2
=.18 AIC =102, R
2
=.22* AIC =58, R
2
=.26* AIC =24, R
2
=.18† AIC =191, R
2
=.45*†
Variables PCT_MFG*(+) PCT_MFG*(+) PCT_MFG***(+) SQ_DENSITY*(-) SQ_INCOME*(-)
LOG_DENSITY*(-) LOG_DENSITY**(-) LOG_DENSITY*(-)
SQ_INCOME*(-)
Model III††
Performance AIC =145, R
2
=.20 AIC =105, R
2
=.20* AIC =60, R
2
=.26* AIC =41, R
2
=.04 AIC =195, R
2
=.43*†
Variables PCT_MFG**(+) LOG_DENSITY**(-) PCT_MFG***(+) PCT_LDCV22*(-) SQ_INCOME*(-)
LOG_DENSITY*(-) LOG_DENSITY*(-)
SQ_INCOME*(-)
Model IV††
Performance AIC =147, R
2
=.18 AIC =108, R
2
=.17* AIC =62, R
2
=.25 AIC =41, R
2
=.05 AIC =217, R
2
=.19*
Variables PCT_HISPAN*(-) PCT_MFG***(+) PCT_LDCV22*(-) PCT_LDCV22*(-)
PCT_MFG**(+) PCT_LDCV23***(-)
PCT_LDCV22*(-) SQ_INCOME**(-)
SQ_INCOME*(-)

*p > .05, **p > .01, ***p > .001; †High Clustering of Residuals; ††VIF for Race Variables > 7.5

110
In spite of the mis-specified models, limited number of TRI sites, and
disappointing results, the pattern evident in most of these cities was that TRI sites tended
to be located in areas with lower density. The results also suggest that the percentage of
manufacturing employees often positively correlated with the location of sites, which is
often the counter-argument to race and income inequity of site location. Median income
and house values were negatively correlated, indicating that poor people may indeed be
over-represented near sites. However, without properly specified models, these results
should be taken only as an indicator of how to re-examine the patterns in each city, and
no statement can be made on racial patterns. As well, explanations of why sites cluster
that move beyond traditional explanations should be developed.

Large-Quantity Generators

This section describes the results of the GIS analysis for LQGs. The total number
of LQGs for reporting year 2005 is 471. Again, I expected to see a preponderance of
LQGs in Rust Belt cities, but was surprised that San Diego, Seattle, and San Jose are
among the top three cities (see Figure 4.6). Rust Belt cities account for only 29% of
LQGs in the study, while Sun Belt cities account for 34%.
In terms of LQGs per person, Dayton has 1 site for every 6,647 persons, followed
by Seattle, which has 1 site for every 7,317 persons. Flagstaff has the best ratio with only
one LQG in the city of 53,137. Philadelphia has the next best ratio with 1 LQG for every
38,912 persons. (Of course, Santa Fe has no LQGs, so technically it has the best ratio of
the 18 cities in the study.)

111
Figure 4.6
LQGs by City
0
20
40
60
80
100
120
San Diego
Seattle
San Jose
Philadelphia
Miami
Grand Rapids
Boston
Dayton
Austin
Salt Lake City
Albany
Little Rock
Norfolk
Boulder
Charleston
Decatur
Flagstaff

Table 4.17 identifies the tract with the largest LQG in each of the 17 cities, along
with racial, income, and education data for the host tract. Dayton, Decatur, Grand Rapids,
Little Rock, Miami, and San Jose’s largest LQGs were in tracts with greater than 51%
minority population. Cities with the largest LQGs in tracts with median income of less
than $30,000 included Albany, Charleston, Dayton, Decatur, Grand Rapids, Little Rock,
and Miami. Finally, those tracts with the largest LQGs and where less than 15% of the
residents had a BA or better were found in Dayton, Decatur, Grand Rapids, Little Rock,
Miami, Norfolk, Philadelphia, Salt Lake City, and San Jose. Those cities matching all
three of the above criteria included: Dayton, Decatur, Grand Rapids, Little Rock, and
Miami. Recall that Little Rock’s host tract to the largest-producing TRI site also matched
all three criteria. As with the TRI sites, African Americans and Hispanics tended to be the
major racial or ethnic categories in many of the tracts hosting or adjacent to the LQGs
managing the most tons.

112
Table 4.17
LQGs with the Most Tons Managed

Most Tons
Managed

Tract

%
White

%
Minority

Median
Income

% BA
or
Better

Albany 7,605.00 3
†
55 42 30,874 18
Austin 1,131.10 18.42 70 28 55,192 21
Boston 63.20 203 72 26 52,160 59
Boulder 47,894.63 122.03 70 27 35,853 48
Charleston 471.68 17 90 8 30,611 18
Dayton 8,562.33 603
†
16 83
a
27,264 10
Decatur 522.76 9 41 58
b
18,333 10
Flagstaff 14 86 13 45,625 38
Grand Rapids 1,241.42 36 6 92 19,866 3
Little Rock 192.48 2 11 88
c
18,099 5
Miami 397.006
††
15.01 0 99
d
8,853 2
Norfolk 868.43 9 51 46 35,521 13
Philadelphia 31,893.48 184 96 33,065 6
Salt Lake City 2,357.45 1004 65 33
e
39,512 10
San Diego 9,273.31 99.02 75 24 --- ---
San Jose 18,093.54 5043.18 35 62
f
52,065 11
Seattle

11,467.36

99

69

27

46,684

34

†
Facility is just outside tract, less than ½ mile away.
††
Facility is just outside tract, less than ½ mile away.
a
African Americans make up 82% of minority population.
b
African Americans make up 56% of the minority population.
c
African Americans make up 86% of the minority population.
d
African Americans make up 96% of the minority population.
e
Hispanics make up 27% of the minority population.
f
Hispanics make up 30% and Asians make up 28% of the minority population.

Do LQGs Cluster?

Using Moran’s I to determine if LQGs cluster, Flagstaff dropped out of the
analysis because it has fewer than three features, the minimum necessary for Moran’s I to
work (see Table 4.18). Of the 16 cities where the tool ran successfully, all cities exhibited
LQG clusters, except for Charleston (somewhat clustered), Little Rock (random), and
Miami (random). In other words, LQGs clustered in 81% of the cities in the study. The
most significant clustering was observed in Austin, Salt Lake City (see Figure 4.7), and

113
San Jose, followed by strong clustering in Seattle. Sites also clustered significantly at the
.05 level in Dayton, Decatur, Grand Rapids, Norfolk, and San Diego.

Table 4.18
Moran’s Index for LQGs

Moran’s I

Z Score

Pattern

Albany

1.61

1.79

Clustered
Austin 1.21*** 4.74 Clustered
Boston 0.28 1.66 Clustered
Boulder 0.08 1.44 Somewhat clustered
Charleston 0.58 1.78 Clustered
Dayton 0.35* 2.30 Clustered
Decatur 0.28* 2.04 Clustered
Flagstaff
  
Grand Rapids 0.46* 2.10 Clustered
Little Rock 0.50 0.68 Random
Miami 0.33 0.73 Random
Norfolk 0.29* 2.35 Clustered
Philadelphia 1.49 1.72 Clustered
Salt Lake City 1.16*** 5.87 Clustered
San Diego 0.95* 2.24 Clustered
San Jose 0.42*** 5.20 Clustered
Seattle

0.45** 11.99 Clustered
*p < .05, **p < .01, ***p < .001

LQGs clustered in fewer cities according to the Ripley’s K analysis (see Table
4.19). Clustering was statistically significant in only Austin, Boston, Dayton, Miami, San
Diego, San Jose, and Seattle. Interestingly, TRI sites also clustered significantly in
Dayton and the three latter cities. LQGs clustered at nearly the same distances as TRI
sites. The smallest distance was 1,221 meters (3/4 mile), with an average distance of
4,103 meters or 2.5 miles (recall the average distance for TRI sites was 4,336 meters or
2.7 miles). This suggests in approximately one third of the cities, LQGs tended to locate
with like businesses.

114

Figure 4.7
Global Clustering of Salt Lake City’s LQGs

Global Moran's I Summary
Moran's Index: 0.373730
Expected Index: -0.030303
Variance: 0.006941
Z Score: 4.849721
p-value: 0.000001

Statistically significant dispersion of LQGs was seen in Albany, Decatur, and Salt
Lake City, though observed K values from Albany and Decatur could be skewed by the
small area of the cities (see Table A.2 in the Appendix).

115
Table 4.19
Ripley’s K Cluster Analysis for LQGs
Expected K Observed K Differential K
Lower
Confidence
Higher
Confidence
Albany 289.85 789.0748 499.2247 0 789.0748
Austin -2144.468 3156.554* 1012.086 789.1384 2733.656
Boston 479.301 1221.911* 742.786 257.6014 793.9808
Boulder 530.7779 877.3909 346.613 0 877.3909
Charleston 303.9051 0 -303.9051 0 646.5822
Dayton 602.4762 1615.579* 1013.103 0 1077.053
Decatur 149.5484 0 -149.5484 0 468.577
Flagstaff error
Grand Rapids 1010.085 1543.36 533.2749 563.5553 1643.032
Little Rock 638.7691 1457.146 818.3773 0 1457.146
Miami 905.9988 3601.282* 2695.283 0 1697.661
Norfolk 1147.586 1976.095 828.509 0 1976.095
Philadelphia 689.446 952.4424 262.9963 0 1010.218
Salt Lake City 2104.148 1885.213 -218.9359 1287.845 2529.278
San Diego 3829.828 6749.832* 2920.003 3377.523 4015.675
San Jose 4478.944 6356.478* 1877.534 3816.475 4825.796
Seattle 3460.984 5523.731* 2062.748 2984.479 3674.617
*p < .01

Where Do Sites Cluster?

Twelve cities exhibited hot spots or clustering of LQGs (Z ≥ 2.56). Austin,
Miami, Philadelphia, San Diego, San Jose, and Seattle exhibited the strongest clustering
(see Table 4.20), while Charleston, Dayton, Decatur, Flagstaff, and Norfolk exhibited
none. Although not identical, San Jose and Seattle’s LQG hot spots and TRI hot spots
were clustered in many of the same tracts, with Seattle’s cluster lying in the southern tip
of the city, near the airport (see Figure 4.8). This is somewhat predictable given Seattle’s
aerospace industry. Although two of Seattle’s tracts had a high percentage of land use
dedicated to industrial, commercial, and transportation use, the other 11 tracts were
primarily low-density residential tracts with minority residents making up 27-77% of the
population. The median income for the 13 tracts ranged from $16,285-$53,198. San

116
Jose’s hot spots were in the northwest section of the city, adjoining the San Francisco
Bay. Median income ranged from $52,065-$100,837 in these tracts, with minority
percentages as high as 80%. Though one tract was 75% Hispanic, the other 13 hot-spot
tracts were majority Asian. Six of Seattle’s 11 hot-spot tracts have Asian populations
over 25%. (See Table A.6 for expanded scores.)

Table 4.20
LQG “Hot Spots” or Getis-Ord Gi* Results

City

Tract

Gi* Z Score

Gi* p Value

Percent
Minority

Median
Income

Albany

4.01

3.16

0.002

13

51,214
Austin 17.22 2.86 0.004 21 46,591
17.06 2.71 0.007 12 63,854
24.17 3.83 0.000 68 41,369
17.45 3.02 0.003 21 70,096
23.14 2.71 0.007 56 23,786
18.29 3.21 0.001 23 45,821
18.28 3.36 0.001 23 70,551
24.13

3.19

0.001

89

34,219

Boston 509 2.96 0.003 47 31,903
501

2.60

0.009

53

28,143

Boulder 127.07 3.38 0.001 15 66,705
122.03

2.85

0.004

27 35,853
Grand Rapids 133 3.22 0.001 27 41,290
39

2.81

0.005

80 35,640
Little Rock 40.07

4.32

0.000

66 25,838
Miami 37.01 2.74 0.006 85 9,628
37.02

5.89

0.000

80

26,196

Philadelphia 50 2.73 0.006 0 0
68 2.88 0.004 41 75,487
181 3.30 0.001 7 26,331
182 3.88 0.000 1 32,820
183 3.30 0.001 7 41,875
185 3.90 0.000 1 28,531
186

3.08

0.002

14

30,950

117
Table 4.20: Continued

Salt Lake City 1139.01 4.34 0.000 21 37,348
1003.03

3.79

0.000

16 69,250
San Diego 83.33 4.95 0.000 27 127,271
83.41 3.71 0.000 31 51,783
83.43 3.20 0.001 43 34,967
83.4 4.04 0.000 26 64,554
83.39 2.95 0.003 37 42,717
83.46

4.19

0.000

44

99,718

San Jose 5050.05 7.27 0.000 51 71,667
5045.05 4.11 0.000 72 79,259
5043.1 3.77 0.000 78 78,531
5043.18 2.95 0.003 62 52,065
5043.11 2.84 0.005 83 100,837
5046.02 3.84 0.000 80 57,589
5044.1 4.39 0.000 72 78,501
5043.15 3.26 0.001 70 84,764
5050.07 4.49 0.000 65 74,911
5043.14 2.68 0.007 73 88,792
5044.11 3.64 0.000 65 85,949
5050.06 4.11 0.000 47 97,098
5051

3.20

0.001

64

59,211

Seattle 109 3.30 0.001 40 33,654
100 3.56 0.000 71 37,122
93 3.15 0.002 53 42,208
112 3.39 0.001 61 30,917
99 2.68 0.007 27 46,684
110 3.11 0.002 86 36,754
113 2.89 0.004 43 46,838
104 4.82 0.000 77 48,697
108 4.58 0.000 49 53198
265 3.69 0.000 60 16,285
264 3.69 0.000 41 40,291

118
Figure 4.8
Seattle’s LQG Hot Spots
City Boundary
LQG_NUM
GiZScore
< -2.0
-2.0 to -1.0
-1.0 to 1.0
1.0 to 2.0
> 2.0
0 7,500 15,000 22,500 3,750
Meters

119
LQG cold spots (Z ≥ -2.56) were found in only two cities, meaning only two of
seventeen cities showed a distinct pattern of non-clustering of LQGs (see Table 4.21). Six
adjacent tracts in the center of Albany registered as strong cold spots; four of the six
tracts had 20% or fewer minorities. Median income ranged from $18,953-$42,981. Salt
Lake City had 19 tracts with significant cold spots, 18 of which were also TRI cold spots.
These coterminous tracts had a minority percentage ranging from 3% to 51%, with 11 of
18 tracts falling below 29% minority. This suggests that LQGs were generally not located
in tracts with low percentages of minorities. Median income ranged from $16,048-
$68,929, with the majority of tracts falling below $33,500. These 18 tracts were located
in the southeastern part of the city, near Wasatch-Cache National Forest. Landcover data
suggests that these tracts were predominantly high-density residential tracts, with a
majority of owner-occupied housing valued on average at $144,415. (See Table A.8 for
expanded scores.)
The Getis-Ord Gi* results for LQGs showed that LQGs clustered more frequently
than not, and when they do, they tended to cluster in minority and poor neighborhoods,
though sometimes with a particular industry located predominantly in those sections of
town. It is interesting that only two cities returned LQG cold spots; even when the results
were expanded to include a Z score ≥ -1.96 (p = .05), Boston was the only other city to
return cold spots, suggesting that LQGs may be more spread across a cityscape, even if in
small numbers in some areas.

120
Table 4.21
LQG “Cold Spots” or Getis-Ord Gi* Results

City

Tract

Gi* Z Score

Gi* p Value

Percent
Minority

Median
Income

Albany 17 -3.71 0.000

15

42,981
16 -3.36 0.001 19 36,371
3 -2.61 0.009 43 30,874
15 -3.14 0.002 20 21,250
5.02 -2.90 0.004 21 36,563
5.01 -2.53 0.011 41 18,953

Salt Lake City 1015 -3.53 0.000 14 26,717
1035 -3.29 0.001 11 49,213
1030 -4.10 0.000 36 32,067
1049 -2.78 0.005 23 33,056
1029 -3.73 0.000 42 24,375
1032 -3.08 0.002 25 27,222
1036 -2.89 0.004 3 68,929
1033 -2.89 0.004 14 38,487
1019 -3.08 0.002 20 25,938
1020 -3.79 0.000 42 24,385
1037 -2.72 0.006 5 61,463
1017 -3.53 0.000 21 27,568
1018 -3.79 0.000 24 33,481
1034 -3.08 0.002 10 42,286
1031 -3.29 0.001 29 37,925
1023 -3.29 0.001 39 16,048
1024 -2.72 0.006 51 23,125
1016 -3.79 0.000 14 31,420
1114 -2.66 0.008 27 30,470

Is a Pattern Evident?

Table 4.22 displays the results of the regression models described in Chapter 3
that were run on the six largest cities. The table shows model performance, significance,
and significant variables for the OLS and GWR regressions. Tables 4.23-4.28 present the
full results for the best performing OLS model in each city. GWR results are in the
Appendix.

1 2 1

Table 4.22
LQG Regression Results for Six Largest Cities
Austin Boston Philadelphia San Diego San Jose Seattle
Model I
Performance AIC =177, R
2
=.02 AIC =376, R
2
=.04* AIC =272, R
2
=.09* AIC =1101, R
2
=-.08† AIC =644, R
2
=.06† AIC =453, R
2
=.04†
Variables MED_INC_99*(-) MED_INC_99***(-) PCT_MIN**(-) PCT_MIN*(-)
PCT_MFG***(+) PCT_MFG*(-) LOG_DENSITY*(-)
POP_DENSITY***(-)
Model II
Performance AIC =177, R
2
=.02 AIC =327, R
2
=.04* AIC =267, R
2
=.10* AIC =1081, R
2
=.06 AIC =629, R
2
=.12† AIC =436, R
2
=.16†
Variables PCT_MFG*(+) MED_INC_99***(-) PCT_MIN*(-) LOG_DENSITY*(-) PCT_MIN*(-) MED_INC_99*(-)
SQ_MED_INC*(-) SQ_POP_DEN**(-) OO_HS_VAL**(-) LOG_DENSITY*(-)
SQ_MED_INC*(-) LOG_DENSITY***(-)
Model III††
Performance AIC =179, R
2
=.02 AIC =323, R
2
=.08* AIC =259, R
2
=.13*†† AIC =1083, R
2
=.06 AIC =635, R
2
=.11*† AIC =410, R
2
=.34*†
Variables PCT_MFG*(+) MED_INC_99*(-) PCT_OTHER*(+) LOG_DENSITY*(-) OO_HS_VAL**(-) PCT_HISPAN***(+)
OO_HS_VAL*(-) LOG_DENSITY*(-) PCT_OTHER**(+)
SQ_POP_DEN**(-) LOG_DENSITY*(-)
SQ_MED_INC*(-)
Model IV††
Performance AIC =182, R
2
=.01 AIC =322, R
2
=.10†† AIC =271, R
2
=.10* AIC =1090, R
2
=.04* AIC =636, R
2
=.11† AIC =401, R
2
=.38*†
Variables PCT_OTHER*(+) PCT_LDCV23***(-) PCT_LDCV22*(-) PCT_WHITE*(+)
PCT_LDCV23*(-) PCT_LDCV23**(-) PCT_BLACK*(+)
PCT_HISPAN*(+)
PCT_ASIAN*(+)
PCT_OTHER**(+)
PCT_LDCV23***(+)
GWR
Performance AIC =178, R
2
=.05 AIC =329, R
2
=.05 AIC =279, R
2
=.08 AIC =178, R
2
=.05 error AIC =396, R
2
=.45

*p > .05, **p > .01, ***p > .001; †High Clustering of Residuals; ††VIF for Race Variables > 7.5

122

In examining the results of the four OLS models and the GWR model for the six
largest cities, it is evident that in Austin, San Diego, and Seattle, the GWR returned the
lowest AIC numbers, as with the TRI regressions. No significant model was returned for
San Jose, and in San Jose, all OLS model residuals showed high clustering, indicating
that some other explanatory factor for LQG location may be at work. Looking
specifically at the OLS models, Model III returned the best results in Boston and
Philadelphia, while Model IV worked best in San Diego and Seattle (though the residuals
were highly clustered in the latter). Similar to the TRI regressions, Models III and IV in
all cases had very large VIF numbers and thus, thus, the models were most likely mis-
specified and the racial and ethnic variables should be regressed separately. Median
income was significant and inversely related in four out of six cities, while owner-
occupied housing was significant and inversely related in only Philadelphia and San Jose,
similar to the TRI regressions. Population density was inversely related in four out of six
cities, but PCT_LDCV23 (industrial, commercial, and transportation land use) was also
inversely related in three of those cities. This was highly unexpected, as PCT_LDCV23
was expected to be positively related as in Seattle. Another unexpected result, which may
be due to model mis-specification, was the positive significance of every racial and ethnic
variable in Seattle. In Seattle’s Model IV results, however, significant clustering of the
residuals was evident. Running Getis-Ord Gi* on all variables in Seattle (and San Jose)
may indicate what factor(s) I may have missed.

123
Table 4.23
LQG/OLS Regression Results for Austin Model I
Variable Coefficient StdError t-Statistic Probability Robust_SE Robust_t Robust_Pr VIF [1]
Intercept 0.011024 0.131851 0.083611 0.933449 0.108203 0.101884 0.918953 --------
PCT_MIN 0.000052 0.001505 0.034638 0.9724 0.001591 0.032766 0.973891 1.827271
MED_INC_99 -0.000005 0.000002 -2.484127 0.013902* 0.000002 -2.244858 0.025998* 4.018365
PCT_MFG 0.017578 0.006467 2.718231 0.007210* 0.009357 1.878622 0.061932 1.553291
OO_HS_VAL 0.000001 0.000001 1.817972 0.070753 0.000001 1.748432 0.08212 3.337452
POP_DENSITY -0.00014 0.000245 -0.570406 0.569126 0.000084 -1.659915 0.098701 1.014616

Number of Observations: 184 Number of Variables: 6
Degrees of Freedom: 178 Akaike's Information Criterion (AIC) [2]: 177.292789
Multiple R-Squared [2]: 0.049327 Adjusted R-Squared [2]: 0.022622
Joint F-Statistic [3]: 1.847147 Prob(>F), (5,178) degrees of freedom: 0.105938
Joint Wald Statistic [4]: 9.009385 Prob(>chi-squared), (5) degrees of freedom: 0.10869
Koenker (BP) Statistic [5]: 7.646943 Prob(>chi-squared), (5) degrees of freedom: 0.176797
Jarque-Bera Statistic [6]: 3403.305716 Prob(>chi-squared), (2) degrees of freedom: 0.000000*

Table 4.24
LQG/OLS Regression Results for Boston Model III
Variable Coefficient StdError t-Statistic Probability Robust_SE Robust_t Robust_Pr VIF [1]
Intercept -2.257759 4.083394 -0.552912 0.581168 2.371247 -0.95214 0.34258 --------
PCT_WHITE 0.029672 0.041327 0.717972 0.473909 0.024827 1.195142 0.233956 629.558545
PCT_BLACK 0.028698 0.042527 0.674809 0.500854 0.025364 1.131408 0.259723 456.186703
PCT_HISPAN 0.025166 0.042082 0.598037 0.550738 0.025041 1.005012 0.316534 115.322246
PCT_ASIAN 0.048375 0.042499 1.138283 0.256851 0.030145 1.604767 0.110701 57.134692
PCT_OTHER 0.039201 0.054154 0.723871 0.47029 0.030789 1.273221 0.204953 15.879092
MED_INC_99 -0.000011 0.000005 -2.172254 0.031431* 0.000007 -1.559868 0.120951 2.442082
PCT_MFG -0.021284 0.021285 -0.999943 0.318973 0.012831 -1.658792 0.099299 1.656422
OO_HS_VAL 0.000001 0.000001 0.842849 0.400673 0.000001 0.953591 0.341847 1.6551
SQ_DENSITY -0.007391 0.017176 -0.430319 0.667604 0.01115 -0.662865 0.508453 1.116597

Number of Observations: 157 Number of Variables: 10
Degrees of Freedom: 147 Akaike's Information Criterion (AIC) [2]: 323.092022
Multiple R-Squared [2]: 0.140181 Adjusted R-Squared [2]: 0.087539
Joint F-Statistic [3]: 2.662922 Prob(>F), (9,147) degrees of freedom: 0.006855*
Joint Wald Statistic [4]: 14.002648 Prob(>chi-squared), (9) degrees of freedom: 0.122231
Koenker (BP) Statistic [5]: 11.031738 Prob(>chi-squared), (9) degrees of freedom: 0.27354
Jarque-Bera Statistic [6]: 6695.202536 Prob(>chi-squared), (2) degrees of freedom: 0.000000*

124
Table 4.25
LQG/OLS Regression Results for Philadelphia Model III
Variable Coefficient StdError t-Statistic Probability Robust_SE Robust_t Robust_Pr VIF [1]
Intercept 0.60538 0.105976 5.712447 0.000000* 0.252128 2.401083 0.016825* --------
PCT_WHITE -0.000531 0.001566 -0.339083 0.734751 0.003022 -0.175706 0.860614 10.989275
PCT_BLACK -0.001732 0.001462 -1.184787 0.23686 0.002858 -0.606083 0.544833 9.988904
PCT_HISPAN -0.001745 0.001965 -0.888175 0.375009 0.002986 -0.584283 0.55939 2.671299
PCT_ASIAN 0.006151 0.002975 2.067401 0.039380* 0.005373 1.144849 0.253007 1.422871
PCT_OTHER 0.113775 0.047767 2.381885 0.017715* 0.053548 2.124734 0.034258* 1.024124
PCT_MFG -0.004332 0.003589 -1.206791 0.228283 0.005191 -0.834523 0.404511 1.50922
OO_HS_VAL -0.000001 0 -1.961202 0.050598 0 -2.068517 0.039274* 1.417425
SQ_POP_DEN -0.02491 0.006535 -3.811523 0.000171* 0.008032 -3.101299 0.002085* 1.359692
SQ_MED_INC -0.001022 0.000467 -2.188446 0.029247* 0.000519 -1.970871 0.049476* 1.967586

Number of Observations: 381 Number of Variables: 10
Degrees of Freedom: 371 Akaike's Information Criterion (AIC) [2]: 259.381096
Multiple R-Squared [2]: 0.1468 Adjusted R-Squared [2]: 0.126103
Joint F-Statistic [3]: 7.092645 Prob(>F), (9,371) degrees of freedom: 0.000000*
Joint Wald Statistic [4]: 22.014199 Prob(>chi-squared), (9) degrees of freedom: 0.008834*
Koenker (BP) Statistic [5]: 43.772339 Prob(>chi-squared), (9) degrees of freedom: 0.000002*

Table 4.26
LQG/OLS Regression Results for San Diego Model IV
Variable Coefficient StdError t-Statistic Probability Robust_SE Robust_t Robust_Pr VIF [1]
Intercept -2.689632 7.077947 -0.380002 0.704239 3.071004 -0.875815 0.381853 --------
PCT_WHITE 0.048764 0.073953 0.659395 0.510171 0.035774 1.363118 0.173927 606.806919
PCT_BLACK 0.038988 0.076896 0.507016 0.612544 0.034664 1.124712 0.26165 60.152606
PCT_HISPAN 0.046158 0.071956 0.641477 0.521726 0.034683 1.330861 0.184303 381.307773
PCT_ASIAN 0.070284 0.077189 0.910545 0.36329 0.043009 1.634182 0.103332 121.191229
PCT_OTHER -0.032124 0.1427 -0.225118 0.822048 0.144019 -0.223057 0.82365 1.569165
PCT_MFG -0.016057 0.022665 -0.708431 0.479249 0.017918 -0.896145 0.370917 1.608108
OO_HS_VAL -0.000001 0.000001 -1.128684 0.259973 0.000001 -1.615201 0.107381 2.195136
PCT_LDCV22 -0.009645 0.00834 -1.156443 0.248463 0.004573 -2.109379 0.035771* 1.254385
PCT_LDCV23 -0.014688 0.004099 -3.583702 0.000410* 0.005111 -2.873594 0.004366* 1.159423
SQ_INCOME -0.002348 0.002256 -1.040896 0.298794 0.00165 -1.423177 0.155786 2.508181

Number of Observations: 297 Number of Variables: 11
Degrees of Freedom: 286 Akaike's Information Criterion (AIC) [2]: 1090.935229
Multiple R-Squared [2]: 0.073931 Adjusted R-Squared [2]: 0.041551
Joint F-Statistic [3]: 2.28324 Prob(>F), (10,286) degrees of freedom: 0.013782*
Joint Wald Statistic [4]: 11.373411 Prob(>chi-squared), (10) degrees of freedom: 0.329175
Koenker (BP) Statistic [5]: 16.710325 Prob(>chi-squared), (10) degrees of freedom: 0.081025
Jarque-Bera Statistic [6]: 35402.8643 Prob(>chi-squared), (2) degrees of freedom: 0.000000*

125
Table 4.27
LQG/OLS Regression Results for San Jose Model III
Variable Coefficient StdError t-Statistic Probability Robust_SE Robust_t Robust_Pr VIF [1]
Intercept -3.107072 8.32444 -0.373247 0.709383 7.848435 -0.395884 0.692637 --------
PCT_WHITE 0.051604 0.087052 0.5928 0.554009 0.082564 0.625026 0.532689 725.162925
PCT_BLACK 0.100956 0.114527 0.88151 0.379122 0.111262 0.90738 0.365321 6.735111
PCT_HISPAN 0.035957 0.08436 0.426234 0.670421 0.077686 0.462851 0.644001 505.173049
PCT_ASIAN 0.035108 0.086187 0.40735 0.684212 0.0765 0.458935 0.646806 375.195958
PCT_OTHER 0.054145 0.144066 0.375833 0.707463 0.132151 0.409721 0.682474 1.247772
PCT_MFG 0.045544 0.022251 2.046785 0.042021* 0.038364 1.18714 0.236624 4.39849
OO_HS_VAL -0.000002 0.000001 -2.609989 0.009755* 0.000001 -2.224809 0.027235* 2.21107
SQ_INCOME -0.001677 0.003069 -0.546522 0.585342 0.001889 -0.888066 0.375594 2.504823
LOG_DENSITY -0.22536 0.059826 -3.766946 0.000228* 0.089181 -2.52701 0.012294* 1.030019

Number of Observations: 204 Number of Variables: 10
Degrees of Freedom: 194 Akaike's Information Criterion (AIC) [2]: 635.807788
Multiple R-Squared [2]: 0.153147 Adjusted R-Squared [2]: 0.11386
Joint F-Statistic [3]: 3.898161 Prob(>F), (9,194) degrees of freedom: 0.000144*
Joint Wald Statistic [4]: 10.111514 Prob(>chi-squared), (9) degrees of freedom: 0.341533
Koenker (BP) Statistic [5]: 19.792743 Prob(>chi-squared), (9) degrees of freedom: 0.019235*
Jarque-Bera Statistic [6]: 16933.10478 Prob(>chi-squared), (2) degrees of freedom: 0.000000*

Table 4.28
LQG/OLS Regression Results for Seattle Model IV
Variable Coefficient StdError t-Statistic Probability Robust_SE Robust_t Robust_Pr VIF [1]
Intercept -20.56295 9.731174 -2.113101 0.036749* 7.869479 -2.613 0.010169* --------
PCT_WHITE 0.191006 0.101911 1.874252 0.063435 0.081483 2.344119 0.020780* 499.862821
PCT_BLACK 0.184653 0.105852 1.744443 0.083759 0.081883 2.255082 0.026013* 103.269252
PCT_HISPAN 0.323597 0.111231 2.909235 0.004352* 0.099187 3.262501 0.001461* 20.241021
PCT_ASIAN 0.211465 0.107572 1.965797 0.051731 0.086559 2.443011 0.016078* 165.498543
PCT_OTHER 0.540663 0.171829 3.146513 0.002110* 0.204392 2.645221 0.009301* 2.987323
MED_INC_99 0.000036 0.000011 3.276164 0.001399* 0.000011 3.211125 0.001721* 3.119004
PCT_MFG -0.023222 0.039931 -0.581547 0.562013 0.032906 -0.705711 0.481791 2.225079
OO_HS_VAL -0.000002 0.000002 -1.268809 0.207074 0.000001 -1.72966 0.08638 2.070704
PCT_LDCV22 -0.018221 0.024982 -0.72938 0.467247 0.025514 -0.714153 0.476575 1.104657
PCT_LDCV23 0.026629 0.004924 5.40846 0.000001* 0.008388 3.174515 0.001933* 1.594464

Number of Observations: 126 Number of Variables: 11
Degrees of Freedom: 115 Akaike's Information Criterion (AIC) [2]: 401.90843
Multiple R-Squared [2]: 0.433226 Adjusted R-Squared [2]: 0.383941
Joint F-Statistic [3]: 8.790275 Prob(>F), (10,115) degrees of freedom: 0.000000*
Joint Wald Statistic [4]: 34.647434 Prob(>chi-squared), (10) degrees of freedom: 0.000143*
Koenker (BP) Statistic [5]: 41.312573 Prob(>chi-squared), (10) degrees of freedom: 0.000010*
Jarque-Bera Statistic [6]: 429.6244 Prob(>chi-squared), (2) degrees of freedom: 0.000000*

126
As for the six medium-sized cities (see Table 4.29), Model III worked best in
Miami, Norfolk, and Salt Lake City, while Model II worked best in Dayton and Grand
Rapids. Large VIF numbers for the race variables were again present in all cities in
Models III and IV. Interestingly, Model I worked best in Little Rock, where the model
explained 66% of the variation in LQG locations. In that model, density was inversely
related and employment in manufacturing was positively related. PCT_MFG was
positively related in Dayton, Little Rock, and Norfolk, while density was inversely
related in all but Dayton. In every model in Dayton, median income was inversely
related, suggesting that poor people may be more exposed to LQG locations in Dayton
than in the other medium-sized cities, though median income was also significant in three
of four models in Salt Lake City.
In spite of the mis-specified models, limited number of LQG locations, and
disappointing results, one evident pattern was that in most of these cities, LQGs tended to
locate, like TRI sites, in areas with lower population density. It is also evident that the
percentage of manufacturing employees often positively correlated with the location of
sites, providing a counter-argument to race and income inequity. Lower income was
negatively correlated, indicating that poor people may indeed be over-represented near
sites, particularly in the middle of the country. However, without properly specified
models, these results should be taken only as an indicator of how to re-examine the
patterns in each city, and no statement can be made on racial patterns.

127

Table 4.29
LQG Regression Results for Six Medium-sized Cities
Dayton Grand Rapids Little Rock Miami Norfolk Salt Lake City
Model I
Performance AIC =156, R
2
=.07 AIC =111, R
2
=.06 AIC =91, R
2
=.66* AIC =205, R
2
=-.03* AIC =110, R
2
=.12* AIC =140, R
2
=-.02†
Variables MED_INC_99**(-) PCT_MFG*(+) POP_DENSITY**(-)
PCT_MFG*(+) POP_DENSITY*(-)
Model II
Performance AIC =151, R
2
=.13* AIC =106, R
2
=.15* AIC =90, R
2
=.37 AIC =197, R
2
=.06 AIC =105, R
2
=.16 AIC =102, R
2
=.50*
Variables PCT_MFG*(+) LOG_DENSITY*(-) PCT_MFG*(+) LOG_DENSITY**(-) PCT_MIN*(+) SQ_INCOME**(-)
SQ_INCOME*(-) POP_DENSITY*(-) PCT_MFG*(+) LOG_DENSITY***(-)
SQ_DENSITY***(-)
Model III††
Performance AIC =156, R
2
=.11* AIC =111, R
2
=.10 AIC =60, R
2
=.26* AIC =187, R
2
=.19* AIC =109, R
2
=.16* AIC =104, R
2
=.47*
Variables PCT_MFG*(+) LOG_DENSITY*(-) PCT_MFG*(+) PCT_OTHER***(+) SQ_DENSITY***(-) SQ_INCOME**(-)
SQ_INCOME*(-) POP_DENSITY*(-) LOG_DENSITY*(-) LOG_DENSITY***(-)
Model IV††
Performance AIC =159, R
2
=.08 AIC =115, R
2
=.06 AIC =99, R
2
=.29 AIC =190, R
2
=.17* AIC =119, R
2
=.06 AIC =123, R
2
=.27*
Variables PCT_MFG*(+) PCT_MFG*(+) PCT_OTHER***(+) PCT_MFG*(+) PCT_LDCV22**(-)
SQ_INCOME*(-) PCT_LDCV22**(-) PCT_LDCV23**(-)
SQ_INCOME**(-)

*p > .05, **p > .01, ***p > .001; †High Clustering of Residuals; ††VIF for Race Variables > 7.5

128
Conclusion

As the results discussed here suggest, TRI facilities and LQGs are not evenly
distributed across a city’s landscape. Both TRI facilities and LQGs are clustered in 79%
and 81% of the cities studied, often in poor and minority areas. Cold spot analysis
revealed that in some cities, like Salt Lake City and Seattle, no facilities were found in
several predominantly white and affluent areas. How do the cities compare in terms of
region and size? Table 4.30 arrays the cities by region, and displays the results of several
spatial statistics for TRI sites.

129

Table 4.30
TRI/City Comparison by Region
Sites Total Release Persons/Site Biggest Producer Tract Mean Center Moran's I Mean Getis-Ord Gi* (Hot)
RUST BELT
Philadelphia 33 1,146,199.61 45,986 no data 81% Minority, $15,687 clustered 35% Minority, $36,512
Boston* 6 44,499.80 98,190 10% Minority, $57,292 20% Minority, $61,458 random none
Grand Rapids 36 298,264.08 5,494 27% Minority, $30,975 92% Minority, $19,866 clustered 45% Minority, $69,442
Dayton 22 375,889.24 6,647 2% Minority, $28,262 7% Minority, $25,238 clustered 10% Minority, $44,720
Decatur IL 6 5,584,944.74 13,686 10% Minority, $25,317 58% Minority, $18,333† mild cluster 9% Minority, $45,693
Albany* 3 8,610.90 31,866 57% Minority, $25,600 52% Minority, $10,897 error none
SUN BELT
San Diego 28 436,969.75 43,691 95% Minority, $22,802 48% Minority, $35,162 clustered 38% Minority, $65,041
Austin* 19 131,703.32 34,542 39% Minority, $40,033 28% Minority, $32,259† clustered 73.5% Minority, $39,240
Miami 13 337906.51 27,889 68% Minority, $15,701 unincorporated clustered none
Little Rock* 10 31,424.90 18,356 75% Minority, $23,021 97% Minority, $16,692 mild cluster 66% Minority, $25,838
Flagstaff 1 6,350.00 53,137 13% Minority, $45,625 none†† error none
OTHER
San Jose 22 563,216.84 40,631 62% Minority, $52,065 76% Minority, $51,646 clustered 66% Minority, $79,461
Seattle 31 124,257.80 18,173 61% Minority, $30,917 53% Minority, $42,208† clustered 59% Minority, $38,195
Norfolk 7 562,411.25 33,486 97% Minority, $11,929 97% Minority, $22,679 clustered 98% Minority, $12,813
Salt Lake City* 34 624,553.55 5,337 33% Minority, $39, 512 40% Minority, $13,750 clustered 19% Minority, $40,116
Boulder 9 305,228.27 10,501 13% Minority, $51,087 unincorporated area clustered 10% Minority, $69,442
Charleston* WV 2 3,137.56 26,598 12% Minority, $55,573 38% Minority, $9,397 error none

*Capital
†Mean Center for TRI is same as Mean Center for LQG
††Single TRI site and single LQG located in same tract

130

Several observations about these regions can be made. First, TRI facilities are
spread across all regions, though the largest amount of TRI releases come from the Rust
Belt region. Even so, in only one Rust Belt city (i.e., Albany), the largest-producing
facility was located in a predominantly minority tract. In three of five Sun Belt cities and
three of five of the Other cities, the biggest producers were in minority tracts. Income
tells a different story, however. In the Rust Belt and Sun Belt cities, the biggest producers
were in low- to middle-income tracts. In the Other cities, though Norfolk is an exception,
the biggest producers were in middle- to upper-middle income tracts. In terms of the
mean center for TRI sites in each city, the Rust Belt stands out as having an inequitable
distribution. In four of six cities, the mean center was in a minority tract, and in five of
six cities, the mean center was in a very low to low-middle income tract. The Other cities
were evenly split among minority and low income tracts where the mean center was
located. Sun Belt cites shone in this respect—only one mean center (i.e., Little Rock) was
located in a minority and very low-income tract. As far as global clustering (using
Moran’s I), the Other cities displayed the most consistent clustering. Table 4.18 also
displays the mean percent minority and mean median income for all the hot spots found
(using Getis-Ord Gi*) in each city. Again, the Other cities had hot spots in five of six
cities (consistent with global clustering), and means for three of the five cities exceeded
59% minority, though income varied between $12,813 and $79,461. The latter, present in
San Jose, is most likely due to the fact that San Jose is a leader in high and nano
technology. In terms of region, neither the Sun Belt nor the Rust Belt emerged as the

131
most consistent region of environmental inequity. In fact, the Other cities appeared to be
more consistently inequitable, something that I was not expecting.
Table 4.31 arrays the same data in terms of city size, comparing large, medium,
and small cities. The pattern that emerges is that size matters—in some respects. The
smaller cities, with the exception of Decatur, had few sites with relatively small releases.
Collectively, the largest number of sites was located in big cities, where three of five
cities had their biggest producing facility in a minority tract. Of the big cities, Boston had
the fewest number of TRI facilities, but also the most equitable in terms of the
demographics of the tracts with the biggest producer and mean center. The medium-sized
cities operate more like big cities than they do small cities, having only 17 fewer facilities
collectively than the big cities. The medium-sized cities also have some of the worst
persons-to-site ratios. In three of the six medium-sized cities, the tract with the biggest
producer had minority percentages ranging from 68% to 97%, with similarly high
number for the mean center tracts. Norfolk scores particularly poorly on both race and
income in several categories.

132

Table 4.31
TRI/City Comparison by Population
Population Sites Total Release Persons/Site Biggest Producer Tract Mean Center Moran's I Mean Getis-Ord Gi* (Hot)
Large Cities
Philadelphia 1,517,550 33 1,146,199.61 45,986 no data 81% Minority, $15,687 clustered 35% Minority, $36,512
San Diego 1,223,341 28 436,969.75 43,691 95% Minority, $22,802 48% Minority, $35,162 clustered 38% Minority, $65,041
San Jose 893,889 22 563,216.84 40,631 62% Minority, $52,065 76% Minority, $51,646 clustered 66% Minority, $79,461
Austin* 656,302 19 131,703.32 34,542 39% Minority, $40,033 28% Minority, $32,259† clustered 73.5% Minority, $39,240
Boston* 589,141 6 44,499.80 98,190 10% Minority, $57,292 20% Minority, $61,458 random none
Seattle 563,375 31 124,257.80 18,173 61% Minority, $30, 917 53% Minority, $42,208† clustered 59% Minority, $38,195
Medium Cities
Miami 362,563 13 337906.51 27,889 68% Minority, $15,701 unincorporated clustered none
Norfolk† 234,403 7 562,411.25 33,486 97% Minority, $11,929 97% Minority, $22,679 clustered 98% Minority, $12,813
Grand Rapids 197,800 36 298,264.08 5,494 27% Minority, $30,975 92% Minority, $19,866 clustered 45% Minority, $69,442
Little Rock* 183,558 10 31,424.90 18,356 75% Minority, $23,021 97% Minority, $16,692 mild custer 66% Minority, $25,838
Salt Lake City* 181,456 34 624,553.55 5,337 33% Minority, $39,512 40% Minority, $13,750 clustered 19% Minority, $40,116
Dayton 166,179 22 375,889.24 6,647 2% Minority, $28, 262 7% Minority, $25,238 clustered 10% Minority, $44,720
Small Cities
Albany* 95,658 3 8,610.90 31,866 57% Minority, $25,600 52% Minority, $10,897 error none
Boulder 94,510 9 305,228.27 10,501 13% Minority, $51,087 unincorporated area clustered 10% Minority, $69,442
Decatur IL 82,113 6 5,584,944.74 13,686 10% Minority, $25,317 58% Minority, $18,333† mild cluster 9% Minority, $45,693
Charleston* WV 53,196 2 3,137.56 26,598 12% Minority, $55,573 38% Minority, $9,397 error none
Flagstaff 53,137 1 6,350.00 53,137 13% Minority, $45,625 none†† error
*Capital
†Mean Center for TRI is same as Mean Center for LQG
††Single TRI site and single LQG located in same tract

133

Turning to LQGs, Table 4.32 shows the cities grouped according to region.
Surprisingly, the Other cities had more LQGs that managed more tons than the Sun Belt
or the Rust Belt cities. However, some of the worst ratios of persons-to-site occurred in
the Rust Belt, whose 135 LQGs managed more than twice as much as the 160 LQGs in
the Sun Belt. Three of five Rust Belt cities had their biggest-producing LQGs in minority
tracts with low median income. Though the Other cities had the most LQGs, only San
Jose’s biggest producer was located in a minority tract. In four of six Other cities,
however, the mean center was located in a minority tract. Global clustering, though not
always significant, was evident in all Rust Belt cities and Other cities. In each region,
only two of four cities registering hot spots had means of over 50% minority, though
twice that amount in each region showed that poor people were more likely to be in LQG
hot spots.

134

Table 4.32
LQG/City Comparison by Region
LQG Total Tons People/Site Biggest Producer Tract Mean Center Moran's I Mean Getis-Ord Gi* (Hot)
RUST BELT
Philadelphia 39 41,575.01 38,911 n/a, $33,065 83% Minority, $18,155 clustered (n/s) 12% Minority, $39,322
Boston* 26 526.05 22,659 26% Minority, $52,160 39% Minority, $45,486 clustered (n/s) 50% Minority, $30,023
Grand Rapids 27 6,316.34 7,325 92% Minority, $19,866 66% Minority, $30,479 clustered 54% Minority, $38,495
Dayton 25 13,139.97 6,647 83% Minority, $27,264 8.5% Minority, $23,018 clustered none
Decatur IL 6 890.20 13,685 58% Minority, $18,333 58% Minority, $18,333† clustered none
Albany* 12 7,916.91 7,971 42% Minority, $30,874 17% Minority, $38,301 clustered (n/s) 13% Minority, $51,214
SUN BELT
San Diego 97 15,526.04 12,611 24% Minority, n/a 34% Minority, $22,340 clustered 35% Minority, $70,168
Austin* 21 3,838.94 31,252 28% Minority, $55,192 28% Minority, $32,259† clustered 39% Minority, $49,535
Miami 29 1,341.291 12,502 99% Minority, $8,853 unincorporated area random 83% Minority, $17,912
Little Rock* 12 545.25 15,293 88% Minority, $18,099 97% Minority, $16,692 random 66% Minority, $25,838
Flagstaff 1 38.06 53,137 13% Minority, $45,625 n/a†† error error
OTHER
San Jose 58 41,939.74 15,411 62% Minority, $52,065 76% Minority, $51,646 clustered 62% Minority, $77,628
Seattle 77 18,356.62 7,316 27% Minority, $46,065 53% Minority, $42,208† clustered 55% Minority, $39,311
Norfolk 8 2,364.72 26,300 27% Minority, $46,684 98% Minority, $27,208 clustered none
Salt Lake City* 19 4,692.09 9,550 46% Minority, $35,521 58% Minority, $29,455 clustered 19% Minority, $53,299
Boulder 8 56,767.20 11,813 27% Minority, $35,853 27% Minority, $35,853 mild cluster 21% Minority, $51,279
Charleston* WV 6 482.32 8,866 8% Minority, $30,611 47% Minority, $19,688 clustered (n/s) none

*Capital
†Mean Center for TRI is same as Mean Center for LQG
††Single TRI site and single LQG located in same tract

135

When grouped according to population, it is apparent that size matters (see Table
4.33). The number of LQGs, as well as the tons managed, for the six largest cities far
exceeded the numbers in the other 12 cities. And global clustering occurred in each large
city. Interestingly, the biggest producers were not necessarily in minority or poor tracts,
though San Jose’s biggest producer was in a minority tract. In terms of the biggest
producers, four of six medium-sized cities were in minority tracts with median income
ranging from $8,853 to $46,684. The mean centers in the medium-sized cities were
located in tracts with similar minority percentages and median income. Though poor
people were also exposed in small cities (according to both indicators), those affected in
larger cities tended to have higher median incomes. Clearly, medium-sized cities show an
over-exposure of minority and poor populations compared to large and small cities.

136

Table 4.33
LQG/City Comparison by Population
Population LQG Total Tons People/Site Biggest Producer Tract Mean Center Moran's I Mean Getis-Ord Gi* (Hot)
Large Cities
Philadelphia 1,517,550 39 41,575.01 38,911 n/a, $33,065 83% Minority, $18,155 clustered (n/s) 12% Minority, $39,322
San Diego 1,223,341 97 15,526.04 12,611 24% Minority, n/a 34% Minority, $22,340 clustered 35% Minority, $70,168
San Jose 893,889 58 41,939.74 15,411 62% Minority, $52,065 76% Minority, $51,646 clustered 62% Minority, $77,628
Austin* 656,302 21 3,838.94 31,252 28% Minority, $55,192 28% Minority, $32,259† clustered 39% Minority, $49,535
Boston* 589,141 26 526.05 22,659 26% Minority, $52,160 39% Minority, $45,486 clustered (n/s) 50% Minority, $30,023
Seattle 563,375 77 18,356.62 7,316 27% Minority, $46,065 53% Minority, $42,208† clustered 55% Minority, $39,311
Medium Cities
Miami 362,563 29 1,341.291 12,502 99% Minority, $8,853 unincorporated area random 83% Minority, $17,912
Norfolk 234,403 8 2,364.72 26,300 27% Minority, $46,684 98% Minority, $27,208 clustered none
Grand Rapids 197,800 27 6,316.34 7,325 92% Minority, $19,866 66% Minority, $30,479 clustered 54% Minority, $38,495
Little Rock* 183,558 12 545.25 15,293 88% Minority, $18,099 97% Minority, $16,692 random 66% Minority, $25,838
Salt Lake City* 181,456 19 4,692.09 9,550 46% Minority, $35,521 58% Minority, $29,455 clustered 19% Minority, $53,299
Dayton 166,179 25 13,139.97 6,647 83% Minority, $27,264 8.5% Minority, $23,018 clustered none
Small Cities
Albany* 95,658 12 7,916.91 7,971 42% Minority, $30,874 17% Minority, $38,301 clustered (n/s) 13% Minority, $51,214
Boulder 94,510 8 56,767.20 11,813 27% Minority, $35,853 27% Minority, $35,853 mild cluster 21% Minority, $51,279
Decatur IL 82,113 6 890.20 13,685 58% Minority, $18,333 58% Minority, $18,333† clustered none
Charleston* WV 53,196 6 482.32 8,866 8% Minority, $30,611 47% Minority, $19,688 clustered (n/s) none
Flagstaff 53,137 1 38.06 53,137 13% Minority, $45,625 n/a†† error error

*Capital
†Mean Center for TRI is same as Mean Center for LQG
††Single TRI site and single LQG located in same tract

137

What these and the other results discussed in this chapter suggest is that across the
17 cities, minority and poor people are over-represented near environmental hazards,
though explaining the patterns through traditional regression analysis led to some
ambiguity. The R
2
for most of the regression models was disappointing. This does not
mean that a pattern is not evident; it means that the traditional variables used to explain
environmental inequity may not accurately or completely describe the existing patterns.
Further specification of models is needed, but other explanatory variables are also
needed, like those institutional variables I examine in Chapter 7. However, this chapter
also shows that in some cases white people and affluent people may also be exposed to
environmental hazards, particularly in cities whose economies depend on high and nano
technology. As Pellow (2002) argues:

To one degree or another, solid waste and pollution are in everyone’s
backyard. There is no denying that politically marginal, low-income, and
people of color populations bear the heaviest burden. But sooner or later,
the affluent, the elites, and middle-class white communities will have to
confront environmental contamination in their homes, neighborhoods, and
workplaces. (p. 160)

Perhaps this is the inescapable result of a nation built on industrialization, as well as one
that has modernized its economy continually. The hazard problem, as intractable as it
may be, creates a vexing challenge for policy makers who struggle with incomplete data
on sites and pollution, poor measures of risk, and a dearth of epidemiological studies that
demonstrate what and how great the deleterious health effects are. The United States may
be a first-world country, but our hazards management is woefully reminiscent of a
burgeoning economy that grows without considering long-term consequences.

138
Schnaiberg and Gould (2000) wonder if this is “the price that must be paid for the
unquestionable advantages of industrial society,” (p. 30). Those advantages are not in
dispute, but the distribution of the disadvantages in the forms of pollution and hazardous
waste are very much in dispute. The lack of an environmental justice strategy, complete
with legislation and standards at any level of government ensures that vulnerable
minority and/or poor populations will remain inadequately protected from hazards
exposure. This may be an inescapable problem, particularly when, as the explanatory
models show here, many factors contribute to the distribution of hazards, while some
factors have yet to be tested. For this reason, we must continue to examine the issue and
begin to examine alternative explanations, as I do in Chapter 7.
In the next chapter, I discuss the distribution of parks and community gardens.

139
CHAPTER FIVE: ENVIRONMENTAL AMENITIES

There is also a need in the environmental justice movement to
move from a vision of rejecting things to a more positive vision of
what kind of world we want.
Manuel Pastor, Reframing Sustainability

In this chapter, I describe the results of the GIS analysis on the distribution of city
parks and community gardens in 18 cities. I treat the amenities separately and have
organized the discussion around the four questions described in Chapter 3: (1) What is
the distribution of the disamenities (i.e., features); (2) Do features cluster; (3) Where do
features cluster; and (4) Is a pattern evident in the distribution of these features vis-à-vis
the sociodemographic data (Mitchell, 2005)? Recall that the overall research question for
this portion of the study is: Are minorities and poor populations under-represented in
neighborhoods where environmental amenities cluster? The hypothesis driving this
portion of the analysis is: The greater the percentage of poor and/or minority residents in
a tract, the fewer the number of amenities and the more unequal the distribution. As
detailed in Chapter 3, several GIS analyses and regressions were done separately on parks
and community gardens.

Some Preliminary Observations

Table 5.1 shows the total number of amenities by type for the 18 cities. Parks
outnumber community gardens 1,535 to 456, for a total of 1,991 amenities. One might

140
expect to see cities with larger land areas to have a greater number of amenities. This is
not always the case, however, as Boston exemplifies. Boston has a land mass area of
89.65 square miles, yet it boasts 249 parks, second only to Seattle’s 328 parks (Seattle’s
land mass area is 142.58 square miles). As well, Boston has 148 community gardens
compared to Seattle’s 56 community gardens. As will be explained in Chapter 7, the
amount of city parks and gardens in Boston is very much a part of the city’s historical
development and social fabric. Note that these calculations are done solely on the count
of amenities and not the size of those amenities.

Table 5.1
Total Parks and Community Gardens by City

Parks

Gardens

Total

Albany 35 20 55
Austin 144 7 151
Boston 249 148 397
Boulder 78 8 86
Charleston 17 2 19
Dayton 33 18 51
Decatur 40 6 46
Flagstaff 23 4 27
Grand Rapids 76 7 83
Little Rock 52 9 61
Miami 39 22 61
Norfolk 32 6 38
Philadelphia 79 100 179
Salt Lake City 65 7 72
San Diego 73 12 85
San Jose 146 19 165
Santa Fe 26 5 31
Seattle

328

56

384

The number of total amenities outnumbers total disamenities in this study by
roughly 38%. One reason may be the fact that hazards are separated into many categories,

141
such as TRI facilities, large- and small-quantity waste generators, and waste-to-energy
facilities, and I only included two hazards in this study. “City parks” often include many
types of municipal parks, including playgrounds, traditional parks, and recreation
facilities (e.g., softball fields), so the potential for more sites is greater. Another reason
for the difference in total amenities versus disamenities may be that federal reporting
requirements for hazards limit the number of facilities that must report and the EPA does
not strictly enforce reporting requirements. Thus, more hazards may exist than are
reported each year. In contrast, reporting amenities yields many benefits for cities, so
they are likely motivated to keep lists current and available for residents, visitors, and
prospective businesses.
As a general way of describing the distribution of the amenities in each city, I
calculated the mean center for each amenity type. The mean center identifies where the
spatial center of a distribution is, based on the average of all the XY coordinates in the
feature class (e.g., parks, community gardens). Unlike the hazards, a comparison of the
amenities revealed that only in Santa Fe (see Figure 5.1) were the mean centers for parks
and community gardens located in the same tract (which was 74% minority, with a
median income of $26,522). However, in six cities (i.e., Albany, Boulder, Charleston,
Little Rock, Salt Lake City, Seattle), the mean centers were in adjacent tracts with
distances between them ranging from less than .5 miles to 1.25 miles between them. This
suggests that parks and community gardens may locate in similar locales. I now turn to
discussing the results of the analyses done on individual amenities, starting with parks.

142
Figure 5.1
Park and Community Garden Mean Centers in Santa Fe
k j
k j
Census Tracts
Percent Minority
14.00 - 20.00
20.01 - 35.00
35.01 - 47.00
47.01 - 64.00
64.01 - 81.00
k j
Gardens_MeanCenter
k j
Parks_MeanCenter
0 4,250 8,500 12,750 2,125
Meters
±
Tract 10.02: 74% minority with a median
income of $26,522

143
Parks

This section describes the results of the GIS analysis for parks. The total number
of parks in this study was 1,535. I expected to see more parks in Sun Belt cities as these
cities are larger in square miles and were built with more green space (Nicolaides, 2003).
Sun Belt cities only account for 23% of the parks in the study, while Rust Belt cities
account for 33% (see Figure 5.2). This could be because several Sun Belt cities exhibit
natural features that are designated as amenities other than city parks. For example,
Miami and San Diego have beaches that are not counted as city parks. As well, San
Diego, Flagstaff, and Santa Fe have federally governed amenities like National Parks and
Forest Recreation Areas. Interestingly, Boston and Philadelphia boast a historic tradition
of parks, including Boston’s famous “pocket parks.” In terms of parks per person,
Boulder and Seattle have ratios of 1 park for every 1,212 persons and 1 park for every
1,718 persons, respectively. The two cities with the worst ratio of parks to people are
Philadelphia and San Diego with ratios of 1 park for every 19,209 persons and 1 park for
every 16,758 persons, respectively.

144
Figure 5.2
Number of Parks by City
0
50
100
150
200
250
300
350
Seattle
Boston
San Jose
Austin
Philadelphia
Boulder
Grand Rapids
San Diego
Salt Lake City
Little Rock
Decatur
Miami
Albany
Dayton
Norfolk
Santa Fe
Flagstaff
Charleston

Do Parks Cluster?

The spatial statistic used to determine if features (i.e., parks) cluster is Moran’s I.
As explained previously, this index is a global statistic that identifies the spatial
autocorrelation of parks, though it does not determine where parks cluster, by measuring
the similarity of nearby features and identifying the probability that the patterns are not
due to a random distribution. Moran’s I calculates an expected and observed index. The
tool also returns Z scores based on the variance of the expected index score and the
observed index. This spatial autocorrelation tool obtains the best results when used on
feature classes of 30 or more; only one city (Charleston) had less than 30 parks. The
results of the Moran’s I analysis on parks are displayed in Table 5.2. In all, parks were
clustered in 15 of 18 cities (i.e., clustering was observed in 83% of cities). The most
significant clustering was observed in Austin, Boston, Dayton, Decatur, Grand Rapids,

145
Little Rock, Miami, Philadelphia (see Figure 5.3), San Diego, San Jose, and Seattle.
Parks were randomly distributed Charleston and Salt Lake City, and only somewhat
clustered, though not significantly, in Boulder.

Table 5.2
Moran’s Index for Parks

Moran’s I

Z Score

Pattern

Albany

0.78*

2.28

Clustered
Austin 0.37*** 16.36 Clustered
Boston 0.21*** 12.29 Clustered
Boulder 0.34 1.77 Somewhat clustered
Charleston 0.58 1.78 Random
Dayton 0.64*** 6.77 Clustered
Decatur 0.51*** 5.89 Clustered
Flagstaff 0.34** 2.64 Clustered
Grand Rapids 0.50*** 7.56 Clustered
Little Rock 0.40*** 6.94 Clustered
Miami 0.76*** 6.33 Clustered
Norfolk 0.04* 2.33 Clustered
Philadelphia 0.88*** 13.05 Clustered
Salt Lake City 0.17 0.70 Random
San Diego 0.75*** 6.07 Clustered
San Jose 0.36*** 10.87 Clustered
Santa Fe 0.86* 2.56 Clustered
Seattle 0.49*** 10.57 Clustered

*p < .05, **p < .01, ***p < .001

146
Figure 5.3
Global Clustering of Parks in Philadelphia

Global Moran's I Summary
Moran's Index: 0.881248
Expected Index: -0.012821
Variance: 0.004693
Z Score: 13.051591
p-value: 0.000000

As explained in Chapter 4, another way to identify global clustering is with the
Ripley’s K statistic. The value of interest in the Ripley’s K analysis is the differential K
(which when positive, indicates clustering and when negative, indicates dispersion).
Clustering (or dispersion) is statistically significant when the Observed K is greater than
the higher confidence level (or when the Observed K is less than the lower confidence
level for dispersion). One hundred Monte Carlo simulations were done, resulting in a
probability of .01 for these results. Table 5.3 shows clustering of parks in 72% of the

147
cities studied, including: Albany, Austin, Boston, Boulder, Charleston, Grand Rapids,
Little Rock, Norfolk, Philadelphia, Salt Lake City, San Diego, San Jose, and Seattle. No
statistically significant dispersion patterns were found (see Table A.3 in the Appendix).

Table 5.3
Ripley’s K Cluster Results for Parks
Expected K Observed K Differential K
Lower
Confidence
Higher
Confidence
Albany 126.2783 285.5254* 159.2741 0 233.1305
Austin 6647.756 10354.92* 3707.16 5874.026 6690.914
Boston 7833.167 21905.6* 14072.43 7126.485 7800.494
Boulder 2371.451 2962.822* 591.3682 2090.728 2491.679
Charleston 3015.696 3946.031* 930.3349 2030.216 3450.704
Dayton 1844.523 1858.761 14.23735 1488.761 2010.395
Decatur 863.1208 891.5443 28.42345 594.3629 1050.695
Flagstaff 275.4597 292.2348 16.77459 0 462.063
Grand Rapids 3402.97 4364.538* 961.5675 3043.551 3555.981
Little Rock 3724.84 4611.092* 866.2519 3085.295 3923.086
Miami 1969.51 1951.76 -17.75034 1640.717 2208.164
Norfolk 6211.846 14255.99* 8044.147 4546.185 6944.412
Philadelphia 1606.336 2546.826* 940.4907 1470.411 2051.391
Salt Lake City 5977.578 11640.34* 5662.759 5924.532 7305.779
San Diego 7265.384 10716.68* 3451.297 6224.25 7460.142
San Jose 10471.43 14793.4* 4321.967 8061.045 8941.268
Santa Fe 243.0866 275.1873 32.10064 0 435.1093
Seattle 2668.33 3525.827* 584.4979 2480.663 2651.358
*p < .01

Where Do Parks Cluster?

Hot spots (Z ≥ 2.56) for parks were identified in 9 of 18 cities: Austin, Charleston,
Dayton, Grand Rapids, Philadelphia, Salt Lake City, San Diego, San Jose, and Seattle
(see Table 5.4). The results for Austin were the strongest, with 28 of 184 tracts, all
adjacent in the south central portion of the city. Clustering was very strong, with 25 of 28
tracts significant at the .001 level or better. The mean percentage of minorities was 51%,
with a range from 5% to 96%. Median income ranged from $7,423 to $84,338, averaging

148
$31,487. Home ownership for this area was 67% on average, with a mean owner-
occupied housing value of $131,471. Interestingly, Austin is considered a college town,
yet residents holding a bachelor’s degree or better for these hots spots was only 31%.
Charleston and Dayton (see Figure 5.4) registered hot spots in 4 and 5 tracts,
respectively, but it is worth noting that the mean percentage of white residents in those
tracts were 71% and 72%, respectively, while average median income was $33,431 and
$25,626 respectively. Philadelphia’s hot spots clustered in the south central part of the
city in 24 tracts. Average minority percentage for those tracts was just 39%. On average,
45% of residents lived in owner-occupied housing, with an average value of $104,400.
San Diego displayed better minority and poor representation in park hot spots, which
were clustered in the southern part of the city. Average minority percentage was 77%
(with the majority being Hispanic residents), and mean home ownership was only 17%
with an average value of $123,845. Median income in the 11 tracts averaged $26,751.
Even when the results are expanded to include Z ≥1.96 (p = .05), minorities averaged
60% of the population of hot spots, making San Diego a standout (see Table A.9).

Table 5.4
Park “Hot Spots” or Getis-Ord Gi* Results

City

Tract

Gi* Z Score

Gi* p Value

Percent
Minority

Median
Income

Austin 8.01 3.70 0.000 92 35,478
9.02 3.76 0.000 94 23,700
11 4.43 0.000 39 45,063
12 4.47 0.000 26 41,366
16.05 4.18 0.000 11 50,375
14.02 4.68 0.000 25 47,550
13.03 3.73 0.000 24 45,515
8.02 3.58 0.000 96 12,427
4.01 3.34 0.001 29 32,719
8.03 4.47 0.000 88 25,703

149
Table 5.4: Continued

10 4.64 0.000 85 23,597
23.04 3.96 0.000 74 27,551
14.03 4.18 0.000 37 40,491
16.03 2.58 0.010 5 84,338
7 4.98 0.000 28 27,768
23.14 2.82 0.005 56 23,786
6.03 2.59 0.010 21 9,580
21.11 2.62 0.009 92 26,463
8.04 4.98 0.000 93 17,725
14.01 4.64 0.000 17 40,095
9.01 4.80 0.000 94 31,538
13.05 4.47 0.000 52 31,700
23.15 4.15 0.000 62 22,363
13.04 3.39 0.001 22 40,871
23.16 4.45 0.000 65 21,411
6.01 4.11 0.000 38 13,750
16.02 3.25 0.001 38 31,299
6.04

3.55

0.000

27

7,423

Charleston 7 2.64 0.008 48 20,868
8 2.65 0.008 30 22,206
21 2.64 0.008 12 55,573
6

2.95

0.003

15 35,077
Dayton 22 2.76 0.006 10 20,799
32 2.73 0.006 10 32,358
33 2.69 0.007 7 34,537
34 2.90 0.004 7 20,972
35 2.58 0.010 95 19,467
Grand Rapids 19

2.79

0.005

34 35,489
Miami 37.01 2.97 0.003 9,628 85

Philadelphia 42.01 3.06 0.002 21 32,564
63 2.58 0.010 82 21,320
1 3.35 0.001 19 48,886
6 2.79 0.005 30 41,563
9 3.03 0.002 30 20,725
10 4.25 0.000 9 72,625
11 3.67 0.000 20 36,564
14 3.54 0.000 67 26,897
15 3.67 0.000 35 38,026
16 4.07 0.000 17 50,598
17 4.51 0.000 21 48,889
18 3.81 0.000 31 36,458
19 2.67 0.008 94 21,766
22 3.18 0.001 86 30,156
23 4.27 0.000 31 29,806
24 3.54 0.000 41 34,247
25 4.58 0.000 50 26,250
26 4.22 0.000 0 0
27 5.01 0.000 45 23,750

150
Table 5.4: Continued

28 4.06 0.000 41 22,759
30 3.42 0.001 83 20,294
29 3.79 0.000 11 26,744
41.01 2.62 0.009 63 20,759
366

4.31

0.000

13

87,027

Salt Lake City 1019 2.58 0.010 20 25,938
1007 2.71 0.007 28 31,265
1002

2.92

0.003

8

67,841

San Diego 50 4.29 0.000 95 22,802
49 3.42 0.001 97 23,728
52 2.69 0.007 39 18,057
53 2.58 0.010 46 19,522
63 3.43 0.001 37 90,957
36.01 3.14 0.002 96 23,750
39.02 2.75 0.006 96 21,477
47 3.05 0.002 90 18,850
51 3.83 0.000 63 11,535
48 2.59 0.010 95 19,925
35.02

3.13

0.002

96

23,667

San Jose 5120.29 2.84 0.005 26 102,652
5120.26

2.89

0.004

89

63,578

Seattle 56 2.76 0.006 7 87,578
100 2.62 0.009 71 37,122
95 3.30 0.001 44 53,447
89 3.46 0.001 55 47,431

Only two cities registered park cold spots at the .01 level or better (Z ≤ -2.56).
Salt Lake City had one tract, while Seattle registered 10 of 126 tracts, all grouped
together in the northeast section of the city. Seattle has 328 parks geographically
scattered across the city. In this cluster of cold spots, the mean percentage of white
residents was 80, while median income averaged $53,776. The mean population density
for the 10 tracts was 29 and the mean percentage of land devoted to low-density
residential land use was 83%, while the mean of industrial, commercial, or transportation

151
use was 9%. These cold spot tracts boasted a mean owner-occupied housing value of
$252,650. See Table 5.5 for minority percentages. (See Table A.11 for expanded scores.)
Figure 5.4
Park Hot Spots in Dayton
PARKS_NUM
GiZScore
< -2.0
-2.0 to -1.0
-1.0 to 1.0
1.0 to 2.0
> 2.0
40°
0 7,500 15,000 22,500 3,750
Meters
.

152
Table 5.5
Park “Cold Spots” or Getis-Ord Gi* Results

City

Tract

Gi* Z Score

Gi* p Value

Percent
Minority

Median
Income

Salt Lake City 1045

-2.65

0.008

4 51,615
Seattle 21 -2.65 0.008 16 50,284
12 -2.64 0.008 30 32,463
36 -2.96 0.003 13 47,547
20 -3.20 0.001 14 50,351
26 -2.75 0.006 11 66,066
11 -2.76 0.006 21 54,776
24 -2.86 0.004 13 62,784
27 -2.94 0.003 10 63,952
25 -2.89 0.004 13 57,778
19 -2.60 0.009 17 51,760

When the Getis-Ord Gi* results are expanded to include tracts with Z ≤ -1.96 (p =
.05), eight cities register park cold spots, but most had only one tract that qualified as a
cold spot. The exception is San Jose. See the Appendix for more information.

Is a Pattern Evident?

This section discusses the results for the regression models specified in Chapter 3.
Table 5.6 displays the model performance, significance, and the significant variables for
the regressions done on the six biggest cities. The medium-sized cities did not have
enough observations to support regression analysis. Only results for the best-performing
OLS models are presented here (see Tables 5.7-5.12). Additional results are located in the
Appendix. As discussed in Chapter 4, adjusted R
2
is used to measure how much of the
model predicts the variation in the dependent variable; however, it should not be
compared between the OLS and GWR regressions. Instead, the AIC, a relative measure of

153
model performance based on the independent variables, can be used to compare one
model’s performance to another’s. Lower AIC numbers indicate a better-performing
model, relatively speaking. The OLS results return a VIF number for the explanatory
variables, which should not exceed 7.5. A VIF greater than 7.5 indicates serious
multicollinearity among the explanatory variables. GIS also returns a map of the
residuals, the distribution of which can be tested with Moran’s I; if the residuals are
clustered, some explanatory factor is missing from the model.
In examining the results of the four OLS models and the GWR model for the six
largest cities, it is evident Model II worked best in four of the six cities, but none of the
results were particularly strong. This could be due to the intense global clustering of
parks in each of these cities. Recall that clustering was significant at the .001 level in
each of the six cities. None of the models worked well in Austin, while in Seattle, Model
III worked best even though the VIF numbers for the racial and ethnic categories far
exceeded 7.5. The large numbers returned for the race variables indicate that they were
interacting with the other variables. Table 5.6 shows that Models III and IV were mis-
specified. The breakdown of racial and ethnic categories should be expunged from the
models and a separate regression should be run on just the racial and ethnic categories.
Seattle was the only city where the GWR worked better than the OLS regressions, and
clustering of the residuals was evident in several cities.

154

Table 5.6
Park Regression Results for Six Largest Cities
Austin Boston Philadelphia San Diego San Jose Seattle
Model I
Performance AIC =561, R
2
=.03† AIC =548, R
2
=.02 AIC =542, R
2
=-.007† AIC =500, R
2
=-.02* AIC =515, R
2
=-.01 AIC =556, R
2
=.10*†
Variables MED_INC_99*(-) MED_INC_99**(-) PCT_MIN**(+)
OO_HS_VAL*(+) OO_HS_VAL*(+)
Model II
Performance AIC =562, R
2
=.02† AIC =548, R
2
=.02 AIC =530, R
2
=.03*† AIC =448, R
2
=.06* AIC =504, R
2
=.04* AIC =546, R
2
=.17*†
Variables SQ_POP_DEN***(+) SQ_INCOME**(-) LOG_DENSITY***(-) PCT_MIN*(+)
LOG_DENSITY**(-) OO_HS_VAL**(+)
LOG_DENSITY**(-)
Model III††
Performance AIC =564, R
2
=.03† AIC =555, R
2
=.006 AIC =537, R
2
=.02† AIC =494, R
2
=.05* AIC =521, R
2
=-.02† AIC =545, R
2
=.20*
Variables OO_HS_VAL*(+) SQ_POP_DEN***(+) SQ_INCOME***(-) LOG_DENSITY***(-) PCT_BLACK*(+)
LOG_DENSITY**(-) OO_HS_VAL**(+)
LOG_DENSITY**(-)
Model IV††
Performance AIC =567, R
2
=.02† AIC =559, R
2
=.01 AIC =548, R
2
=-.008† AIC =504, R
2
=.02 AIC =636, R
2
=.11† AIC =548, R
2
=.19*
Variables SQ_INCOME*(-) PCT_BLACK*(+)
PCT_LDCV21***(-)
GWR
Performance AIC =566, R
2
=.02† AIC =553, R
2
=.02 AIC =539, R
2
=.01† AIC =498, R
2
=.04 AIC =519, R
2
=-.01 AIC =542, R
2
=.27

*p > .05, **p > .01, ***p > .001; †High Clustering of Residuals; ††VIF for Race Variables > 7.5

155
Table 5.7
Park/OLS Regression Results for Austin Model I
Variable Coefficient StdError t-Statistic Probability Robust_SE Robust_t Robust_Pr VIF [1]
Intercept 0.518355 0.349219 1.484326 0.139482 0.306586 1.690734 0.092627 --------
PCT_MIN 0.005845 0.004247 1.376228 0.170469 0.004073 1.434882 0.153066 1.788971
MED_INC_99 -0.00001 0.000005 -2.031024 0.043718* 0.000005 -1.921025 0.056308 2.611878
OO_HS_VAL 0.000004 0.000002 2.359061 0.019382* 0.000002 1.85009 0.065942 2.796687

Number of Observations: 184 Number of Variables: 4
Degrees of Freedom: 180 Akaike's Information Criterion (AIC) [2]: 561.030395
Multiple R-Squared [2]: 0.041387 Adjusted R-Squared [2]: 0.02541
Joint F-Statistic [3]: 2.590404 Prob(>F), (3,180) degrees of freedom: 0.054331
Joint Wald Statistic [4]: 6.604731 Prob(>chi-squared), (3) degrees of freedom: 0.085622
Koenker (BP) Statistic [5]: 5.082791 Prob(>chi-squared), (3) degrees of freedom: 0.165834
Jarque-Bera Statistic [6]: 534.319365 Prob(>chi-squared), (2) degrees of freedom: 0.000000*

Table 5.8
Park/OLS Regression Results for Boston Model II
Variable Coefficient StdError t-Statistic Probability Robust_SE Robust_t Robust_Pr VIF [1]
Intercept 0.523828 0.531278 0.985976 0.325693 0.528695 0.990794 0.323342 --------
PCT_MIN 0.005529 0.004514 1.224766 0.222547 0.004489 1.231641 0.219975 1.602411
MED_INC_99 0.000004 0.000009 0.470384 0.63876 0.000009 0.496384 0.620344 1.978641
OO_HS_VAL 0.000003 0.000001 1.904573 0.058712 0.000001 1.801896 0.073535 1.53256

Number of Observations: 157 Number of Variables: 4
Degrees of Freedom: 153 Akaike's Information Criterion (AIC) [2]: 548.79134
Multiple R-Squared [2]: 0.036786 Adjusted R-Squared [2]: 0.0179
Joint F-Statistic [3]: 1.947758 Prob(>F), (3,153) degrees of freedom: 0.124261
Joint Wald Statistic [4]: 5.043546 Prob(>chi-squared), (3) degrees of freedom: 0.168636
Koenker (BP) Statistic [5]: 4.268139 Prob(>chi-squared), (3) degrees of freedom: 0.233928
Jarque-Bera Statistic [6]: 19.05683 Prob(>chi-squared), (2) degrees of freedom: 0.000073*

Table 5.9
Park/OLS Regression Results for Philadelphia Model II
Variable Coefficient StdError t-Statistic Probability Robust_SE Robust_t Robust_Pr VIF [1]
Intercept 0.031877 0.108146 0.294759 0.768349 0.060894 0.523482 0.600957 --------
PCT_MIN -0.00101 0.000784 -1.288474 0.198379 0.000681 -1.484001 0.13866 1.35098
OO_HS_VAL 0 0 0.204398 0.83815 0 0.265851 0.790505 1.325196
SQ_POP_DEN 0.032661 0.008906 3.667406 0.000292* 0.008236 3.965676 0.000095* 1.22401

Number of Observations: 381 Number of Variables: 5
Degrees of Freedom: 376 Akaike's Information Criterion (AIC) [2]: 530.35382
Multiple R-Squared [2]: 0.035348 Adjusted R-Squared [2]: 0.025086
Joint F-Statistic [3]: 3.4445 Prob(>F), (4,376) degrees of freedom: 0.008812*
Joint Wald Statistic [4]: 21.283136 Prob(>chi-squared), (4) degrees of freedom: 0.000278*
Koenker (BP) Statistic [5]: 12.900059 Prob(>chi-squared), (4) degrees of freedom: 0.011775*
Jarque-Bera Statistic [6]: 1150.677598 Prob(>chi-squared), (2) degrees of freedom: 0.000000*

156

Table 5.10
Park/OLS Regression Results for San Diego Model II
Variable Coefficient StdError t-Statistic Probability Robust_SE Robust_t Robust_Pr VIF [1]
Intercept 0.994156 0.188482 5.274534 0.000000* 0.291075 3.415468 0.000740* --------
PCT_MIN -0.000561 0.001379 -0.407138 0.684218 0.001334 -0.420593 0.674375 1.566215
OO_HS_VAL 0 0 1.036068 0.301019 0 1.347685 0.178813 1.968628
SQ_INCOME -0.002647 0.000702 -3.773288 0.000205* 0.000971 -2.725444 0.006807* 1.808773
LOG_DENSITY -0.06534 0.020603 -3.171397 0.001690* 0.022431 -2.912886 0.003863* 1.076965

Number of Observations: 297 Number of Variables: 5
Degrees of Freedom: 292 Akaike's Information Criterion (AIC) [2]: 488.393362
Multiple R-Squared [2]: 0.074086 Adjusted R-Squared [2]: 0.061402
Joint F-Statistic [3]: 5.841034 Prob(>F), (4,292) degrees of freedom: 0.000156*
Joint Wald Statistic [4]: 10.819741 Prob(>chi-squared), (4) degrees of freedom: 0.028666*
Koenker (BP) Statistic [5]: 11.112889 Prob(>chi-squared), (4) degrees of freedom: 0.025324*
Jarque-Bera Statistic [6]: 3579.082767 Prob(>chi-squared), (2) degrees of freedom: 0.000000*

Table 5.11
Park/OLS Regression Results for San Jose Model II
Variable Coefficient StdError t-Statistic Probability Robust_SE Robust_t Robust_Pr VIF [1]
Intercept 1.408894 0.54435 2.588216 0.010352* 0.538473 2.616462 0.009561* --------
PCT_MIN 0.000603 0.002803 0.215003 0.829985 0.002385 0.252641 0.800809 1.448699
OO_HS_VAL 0 0.000001 0.128961 0.897509 0.000001 0.144562 0.885195 1.834384
SQ_INCOME -0.000792 0.001801 -0.439601 0.660715 0.0018 -0.439827 0.660551 1.60426
LOG_DENSITY -0.157103 0.043481 -3.613158 0.000393* 0.039501 -3.977141 0.000104* 1.011673

Number of Observations: 204 Number of Variables: 5
Degrees of Freedom: 199 Akaike's Information Criterion (AIC) [2]: 504.465236
Multiple R-Squared [2]: 0.063098 Adjusted R-Squared [2]: 0.044266
Joint F-Statistic [3]: 3.350526 Prob(>F), (4,199) degrees of freedom: 0.011107*
Joint Wald Statistic [4]: 19.041694 Prob(>chi-squared), (4) degrees of freedom: 0.000771*
Koenker (BP) Statistic [5]: 11.151325 Prob(>chi-squared), (4) degrees of freedom: 0.024915*
Jarque-Bera Statistic [6]: 56.133983 Prob(>chi-squared), (2) degrees of freedom: 0.000000*

157
Table 5.12
Park/OLS Regression Results for Seattle Model III
Variable Coefficient StdError t-Statistic Probability Robust_SE Robust_t Robust_Pr VIF [1]
Intercept -23.891961 17.690585 -1.350547 0.179453 15.534075 -1.538036 0.126749 --------
PCT_WHITE 0.270163 0.182261 1.482286 0.140961 0.163216 1.655254 0.100561 505.039489
PCT_BLACK 0.350749 0.18972 1.848769 0.067018 0.169882 2.064663 0.041160* 104.791245
PCT_HISPAN 0.342431 0.197101 1.737334 0.084966 0.173781 1.970475 0.051144 20.07643
PCT_ASIAN 0.279602 0.193482 1.445104 0.15111 0.177497 1.575247 0.117908 169.122315
PCT_OTHER 0.268386 0.299575 0.895889 0.372143 0.214388 1.25187 0.213117 2.868297
MED_INC_99 -0.000003 0.000017 -0.172888 0.863035 0.000021 -0.137108 0.891175 2.270698
OO_HS_VAL 0.000007 0.000003 2.587637 0.010882* 0.000003 2.701906 0.007917* 1.917845
LOG_DENSITY -0.687867 0.201154 -3.419594 0.000872* 0.222608 -3.09004 0.002506* 1.303689

Number of Observations: 126 Number of Variables: 9
Degrees of Freedom: 117 Akaike's Information Criterion (AIC) [2]: 545.28139
Multiple R-Squared [2]: 0.252892 Adjusted R-Squared [2]: 0.201808
Joint F-Statistic [3]: 4.950481 Prob(>F), (8,117) degrees of freedom: 0.000027*
Joint Wald Statistic [4]: 36.272913 Prob(>chi-squared), (8) degrees of freedom: 0.000016*
Koenker (BP) Statistic [5]: 20.627942 Prob(>chi-squared), (8) degrees of freedom: 0.008204*
Jarque-Bera Statistic [6]: 67.272415 Prob(>chi-squared), (2) degrees of freedom: 0.000000*

Recall that the explanatory variables included median income, owner-occupied
housing value, population density, percent minority (or the separate racial and ethnic
percentages), and the landcover variables representing both low-intensity residential and
high-intensity residential use. In several cities, (San Diego, San Jose, and Seattle), density
was inversely related to parks. However, in Philadelphia, density was positively related.
This is not surprising given the square miles of the cities involved. Western cities cover
more space, whereas Philadelphia packs the most population (of those cities in the study)
into 143 square miles, compared to San Diego’s 372 square miles. Philadelphia’s density
is more than double San Diego’s. Median income performed as expected in every city in
which it was significant; that is, parks increased as median income decreased. In the two
cities where owner-occupied housing value was significant, it was positively related to
parks, another indication that poor people may be under-represented near parks.
PCT_MIN and the separate racial categories were only significant in Seattle.

158
Surprisingly, PCT_MIN was positively related to parks, as was PCT_BLACK. Given the
high VIF value and the mis-specified models, these results tell us very little about how
race interacts with parks. Future studies should run separate regressions on the racial and
ethnic variables and to examine the Getis-Ord Gi* results for all variables in the six cities
to determine what explanatory factor(s) I am missing. As well, additional explanatory
variables, as I suggest in Chapter 7, should be modeled.

Community Gardens

This section describes the results of the GIS analysis for community gardens. The
total number of gardens is 456. My suspicion that more community gardens would be
found in the Rust Belt cities because of the historical function of community and urban
gardens was confirmed. The two cities with the most community gardens were Boston
and Philadelphia with 148 and 100 gardens, respectively (see Figure 5.5). The Rust Belt
cities accounted for 64% of the community gardens in the study, while the Sun Belt cities
only accounted for 13%. I expected this because the Sun Belt cities tend to have better
weather that allows for home gardens, cheaper and more available produce for sale,
larger lot sizes, and more restrictive land and water use ordinances. The cities with the
fewest community gardens are Charleston with two gardens and Flagstaff with four
gardens. This may be due to city size as both cities had the fewest parks as well. Also,
Flagstaff is located in the Coconino National Forest and is fueled by recreation and
tourism.

159
Figure 5.5
Number of Community Gardens by City
0
20
40
60
80
100
120
140
160
Boston
Philadelphia
Seattle
Miami
Albany
San Jose
Dayton
San Diego
Little Rock
Boulder
Salt Lake City
Grand Rapids
Austin
Norfolk
Decatur
Santa Fe
Flagstaff
Charleston

In terms of community gardens per person, Boston has 1 community garden for
every 3,981 persons and Albany has 1 community garden for every 4,783 persons. The
cities with the poorest ratios are San Diego with 1 community garden for every 101,945
persons and Austin with 1 community garden for every 93,757 persons.

Do Community Gardens Cluster?

Using Moran’s I to determine if community gardens cluster, Charleston dropped
out of the analysis because it has fewer than three features, the minimum necessary for
Moran’s I to work (see Table 5.13). Of the 17 cities where the tool ran successfully,
community gardens were clustered in 13 cities (76%). The strongest and most significant
clustering was seen in Albany (see Figure 5.6), Austin, Dayton, Philadelphia and Seattle,
followed by Boulder and Santa Fe. Gardens were clustered at the .05 level in Decatur,

160
Grand Rapids, Norfolk, and San Diego. Only Little Rock and San Jose displayed random
patterns.

Table 5.13
Moran’s Index for Community Gardens

Moran’s I

Z Score

Pattern

Albany

0.26***

4.24

Clustered
Austin 1.00*** 3.09 Clustered
Boston 0.31 1.83 Clustered
Boulder 0.32** 3.06 Clustered
Charleston
  
Dayton 0.27*** 3.49 Clustered
Decatur 0.52* 2.43 Clustered
Flagstaff 0.37 1.67 Clustered
Grand Rapids 0.24* 2.44 Clustered
Little Rock 0.05 0.91 Random
Miami 0.18 1.34 Somewhat clustered
Norfolk 0.20* 2.47 Clustered
Philadelphia 0.93*** 3.15 Clustered
Salt Lake City 0.37 1.29 Somewhat clustered
San Diego 0.10* 2.30 Clustered
San Jose 0.46 0.69 Random
Santa Fe 0.65** 2.84 Clustered
Seattle

0.03*** 5.67 Clustered
*p < .05, **p < .01, ***p < .001

According to the Ripley’s K analysis, gardens did not cluster as often as parks.
Clustering at the .01 level occurred in only five cities (Austin, Boston, Miami,
Philadelphia, and San Jose), while dispersion occurred in just three cities (Dayton,
Norfolk, and Salt Lake City). Table 5.14 displays the clustered results, while Table A.4 in
the Appendix displays the dispersion results. As mentioned in Chapter 4, Ripley’s K is
extremely sensitive to study area and to the number of points in the data set. Because

161
some cities have very few gardens or small square mileage, the results are skewed. In this
respect, these results should be viewed with caution.

Figure 5.6
Global Clustering of Community Gardens in Albany

Global Moran's I Summary
Moran's Index: 0.264050
Expected Index: -0.058824
Variance: 0.005789
Z Score: 4.243680
p-value: 0.000022

162
Table 5.14
Ripley’s K Cluster Results for Gardens
Expected K Observed K Differential K
Lower
Confidence
Higher
Confidence
Albany 660.7289 816.5939 215.8648 378.0095 872.9755
Austin 11226.982 18226.561* 6999.5776 5336.1904 14118.233
Boston 2532.563 3598.938* 1066.375 2180.265 2409.186
Boulder 337.7414 0 -377.7414 0 911.0277
Charleston error
Dayton 528.8012 655.5621 126.7519 0 1001.388
Decatur 147.3735 0 -174.3735 0 840.1489
Flagstaff 56.68562 0 -56.68562 0 484.4682
Grand Rapids 253.0253 0 -253.0253 0 867.6493
Little Rock 1993.036 2958.447 965.4111 935.5431 3102.845
Miami 1741.987 3342.389* 1600.402 863.0012 2729.049
Norfolk 256.4808 0 -256.4808 0 627.3322
Philadelphia 1396.188 1980.762* 584.5742 1239.052 1480.318
Salt Lake City 235.1691 0 -235.1691 0 811.5362
San Diego 2856.472 294.6379 89.90674 1317.661 5743.549
San Jose 4121.274 6090.101* 1968.826 2528.777 5540.272
Santa Fe 323.6992 0 -323.6992 0 471.9305
Seattle 1669.886 1670.979 1.093052 1344.135 1985.419
*p < .01

Where Do Gardens Cluster?

Garden hot spots (Z ≥ 2.56, p ≥ .01) were found in eight cities, with significant
clustering in Austin, Boston, Philadelphia, and San Diego (see Table 5.15). Other cities
with clustering in one or a few tracts were Grand Rapids, Norfolk, and San Jose
(clustering in one tract only) and Seattle (clustering in four tracts). (See Table A.10 for
expanded results, where Z ≥ 1.96, p ≥ .05; once expanded, 15 cities had hot spots.) San
Diego’s clustering only appeared in 7 of 297 tracts, and Austin’s clustering appeared in
18 of 184 tracts. The mean for San Diego’s minority population in those tracts was 53%,
while Austin’s was 73%. Austin’s median income for garden hot spots ranged from
$12,427-$45,063, with a mean of $25,658.

163
Boston (see Figure 5.7) and Philadelphia with their 148 and 100 gardens tell the
most interesting story of clustering. Boston registered hot spots in 33 tracts, in three
groups on the north, northeast, and southern edges of the city. Minority percentage in
those tracts ranged from 0-92%; the mean was 36%. Average median income for those
hot spots was $41,285, with a range of $12,132-$81,804. Though the average minority
percentage and the median income were somewhat low, the results indicate that gardens
do cluster in minority and poor Census tracts. Of Philadelphia’s 381 tracts, 50 tracts
registered hot spots, with a mean minority percentage of 71%. Median income for those
tracts averaged $21,574. Philadelphia’s tracts clustered in two groups in the central part
of the city, with an average industrial, commercial, or transportation land use of 40%.

Table 5.15
Garden “Hot Spots” or Getis-Ord Gi* Results

City

Tract

Gi* Z
Score

Gi* p
Value

Percent
Minority

Median
Income

Austin 8.01 4.10 0.000 92 35,478
9.02 3.41 0.001 94 23,700
21.09 2.79 0.005 93 30,234
11 3.97 0.000 39 45,063
23.11 2.79 0.005 67 17,321
8.02 3.51 0.000 96 12,427
4.01 3.97 0.000 29 32,719
8.03 3.15 0.002 88 25,703
10 3.07 0.002 85 23,597
7 4.95 0.000 28 27,768
23.14 3.01 0.003 56 23,786
21.11 4.58 0.000 92 26,463
8.04 2.99 0.003 93 17,725
9.01 2.99 0.003 94 31,538
21.1 3.77 0.000 97 27,344
23.16 2.60 0.009 65 21,411
4.02 4.16 0.000 65 25,829
6.01

5.28

0.000

38

13,750

Boston 6.02 2.83 0.005 44 20,175
612 2.71 0.007 7 37,188

164
Table 5.15: Continued

203 2.91 0.004 26 52,160
1105.02 3.04 0.002 33 51,964
1401.04 3.12 0.002 56 27,030
607 2.74 0.006 57 16,500
5.02 3.31 0.001 23 38,958
603 2.69 0.007 2 49,825
1404 3.31 0.001 83 42,986
605 2.61 0.009 3 48,475
1104.02 3.02 0.003 38 50,410
911 2.62 0.009 46 39,222
8.01 3.00 0.003 32 30,968
1010.01 3.05 0.002 92 37,635
1008 2.80 0.005 36 42,392
611 2.71 0.007 59 12,132
608 2.82 0.005 5 41,976
909 2.78 0.005 65 32,731
7.02 2.60 0.009 35 27,675
602 2.78 0.005 0 55,952
1010.02 2.96 0.003 89 36,742
7.01 2.74 0.006 31 36,875
201 2.64 0.008 6 81,804
4.02 3.13 0.002 21 54,833
610 2.74 0.006 43 13,973
1007 2.89 0.004 6 45,566
1401.03 3.09 0.002 46 51,071
1403 3.02 0.003 65 41,209
8.02 2.79 0.005 42 29,555
303 3.22 0.001 12 70,854
1009 3.07 0.002 71 47,109
202 2.82 0.005 13 53,470
604

2.83

0.005

3

43,015

Grand Rapids 9

2.84

0.005

15 36,783
Norfolk 39

2.71

0.007

14

50,398

Miami 52.02 2.56 0.010 94 19,271
37.02 3.20 0.001 85 9,628
29 2.77 0.006 93 24183
30.04 2.58 0.010 89 12,967
27.02 3.65 0.000 88 22,917
36.01 2.58 0.010 96 7,595
28

2.72

0.007

95

9,671

Philadelphia 80 3.46 0.001 97 28,366
106 2.59 0.010 96 15,952
107 2.61 0.009 97 15,417
174 2.92 0.003 97 15,050
86 4.31 0.000 83 19,612
85 3.34 0.001 97 22,077
73 2.59 0.010 97 25,778
74 2.93 0.003 96 17,604

165
Table 5.15: Continued

75 2.97 0.003 0 0
76 5.13 0.000 41 75,487
77 3.98 0.000 66 14,628
78 3.52 0.000 67 28,885
79 4.31 0.000 70 32,380
176.01 3.04 0.002 95 11,909
87 4.37 0.000 51 21,131
88 4.74 0.000 41 9,320
89 3.38 0.001 50 6,311
91 3.22 0.001 68 17,500
92 4.37 0.000 95 20,083
8 2.92 0.003 16 42,346
15 2.73 0.006 35 38,026
18 2.83 0.005 31 36,458
19 2.75 0.006 94 21,766
22 2.58 0.010 86 30,156
13 2.92 0.003 76 24,508
34 3.08 0.002 77 20,500
108 2.61 0.009 98 17,310
176.02 2.95 0.003 92 12,076
141 4.03 0.000 92 12,165
143 2.84 0.004 11 40,293
144 4.68 0.000 69 23,720
145 4.11 0.000 99 12,092
146 4.43 0.000 78 26,295
153 2.82 0.005 90 14,826
154 3.53 0.000 64 31,923
155 4.26 0.000 98 12,333
156 5.09 0.000 93 14,524
157 5.83 0.000 69 22,478
158 5.74 0.000 8 31,862
159 2.93 0.003 5 28,871
160 5.45 0.000 6 29,621
161 4.44 0.000 44 19,098
162 4.59 0.000 96 13,833
163 4.29 0.000 94 15,865
164 3.88 0.000 98 12,314
165 3.85 0.000 98 16,964
166 2.82 0.005 96 14,250
175 2.82 0.005 97 14,294
177 2.97 0.003 81 15,687
178

3.27

0.001

56

14,793

San Diego 19 2.62 0.009 22 47,866
75.02 2.80 0.005 10 34,934
25.01 2.65 0.008 84 25,963
16 2.89 0.004 72 26,216
24.01 2.98 0.003 81 24,274
15 2.73 0.006 44 31,543
17

3.11

0.002

56

26,761

166
Table 5.15: Continued

San Jose 5006

2.88

0.004

33

73,918

Seattle 95 3.91 0.000 44 53,447
103 2.97 0.003 70 39,544
89 2.79 0.005 55 47,431
101 3.21 0.001 60 47,926

167
Figure 5.7
Garden Hot Spots in Boston
GARD_NUM
GiZScore
< -2.0
-2.0 to -1.0
-1.0 to 1.0
1.0 to 2.0
> 2.0
0 10,000 20,000 30,000 5,000
Meters
Ü

168
No cities registered community garden cold spots. When results were expanded to
include tracts with Z ≤ -1.96, only Albany and Norfolk registered cold spots in one and
three tracts, respectively (see Table A.12).

Is a Pattern Evident?

This section presents the regression results for the four community garden models
specified in Chapter 3. Only Boston, Philadelphia, and Seattle had enough gardens to run
regressions. Table 5.16 displays the model performance and significance and the
significant variables. Complete results for the best OLS model for each city are included
here, while GWR results may be found in the Appendix. In all cities, the VIF numbers for
the racial and ethnic variables in Models III and IV exceeded 7.5, indicating that those
variables should be regressed on community gardens separately.

169
Table 5.16
Community Garden Regression Results for Six Largest Cities
Boston Philadelphia Seattle
Model I
Performance AIC =557, R
2
=.12*† AIC =1034, R
2
=.03† AIC =238, R
2
=.04
Variables PCT_MIN***(+) PCT_MIN*(+) PCT_MIN*(+)
MED_INC_99*(+) PCT_TENURE**(-)
Model II
Performance AIC =559, R
2
=.12*† AIC =1026, R
2
=.06*† AIC =240, R
2
=.03
Variables PCT_MIN***(+) PCT_TENURE***(-) PCT_MIN*(+)
MED_INC_99*(+) SQ_POP_DEN***(+)
Model III††
Performance AIC =562, R
2
=.12*† AIC =1025, R
2
=.07*† AIC =263, R
2
=.09*
Variables MED_INC_99*(+) PCT_TENURE***(-) PCT_WHITE*(+)
SQ_POP_DEN***(+) PCT_BLACK**(+)
PCT_HISPAN*(+)
PCT_AISIAN*(+)
Model IV††
Performance AIC =563, R
2
=.11*† AIC =1032, R
2
=-.05*† AIC =233, R
2
=.12*
Variables MED_INC_99*(+) PCT_WHITE*(+) PCT_WHITE**(+)
PCT_BLACK***(+) PCT_BLACK**(+)
PCT_TENURE**(-) PCT_HISPAN**(+)
PCT_LDCV21*(-) PCT_ASIAN**(+)
PCT_OTHER*(+)
PCT_LDCV21*(+)
GWR
Performance AIC =563, R
2
=.21† AIC =1034, R
2
=.04† AIC =242, R
2
=.05

*p > .05, **p > .01, ***p > .001; †High Clustering of Residuals; ††VIF for Race Variables > 7.5

In each city, a different model performed best, and in all cities, the best OLS
regressions performed better than the geographically weighted regressions. (See Tables
5.17-5.19 for OLS results; see the Appendix for GWR results.) In Boston, Model I
returned the best results, with PCT_MIN being positively related to community gardens
(at the .001 level) and median income also being positively related. This is the ideal—to
see both race and income positively correlated with community gardens. Model III
worked best in Philadelphia, where density was again positively related, and as expected,
PCT_TENURE was inversely related. Interestingly, in Models I and IV in Philadelphia,
PCT_MIN and two race variables were positively correlated. Median income was not

170
significant. In both Boston and Philadelphia, however, the residuals were highly
clustered, which calls into question the results of the models. In Seattle, Model IV
worked best and while no clustering of the residuals was evident, the VIF numbers for the
six positively related racial and ethnic variables far exceeded 7.5. What we see in the
results is what must be confirmed with further testing: community gardens may be
positively correlated with minorities in all three cities. Unfortunately, only in Boston was
median income significant. Perhaps with better models that separate the racial and
economic factors, we can better understand the patterns of community gardens.

Table 5.17
Community Garden /OLS Regression Results for Boston Model I
Variable Coefficient StdError t-Statistic Probability Robust_SE Robust_t Robust_Pr VIF [1]
Intercept -0.782964 0.525784 -1.489135 0.13852 0.313312 -2.49899 0.013503* --------
PCT_MIN 0.022391 0.004729 4.73481 0.000006* 0.004078 5.490566 0.000000* 1.662822
MED_INC_99 0.000023 0.000011 2.218399 0.027991* 0.000007 3.456102 0.000719* 2.348189
PCT_TENURE -0.013896 0.010693 -1.299607 0.195695 0.007811 -1.779066 0.077219 1.545286

Number of Observations: 157 Number of Variables: 4
Degrees of Freedom: 153 Akaike's Information Criterion (AIC) [2]: 557.57057
Multiple R-Squared [2]: 0.135379 Adjusted R-Squared [2]: 0.118426
Joint F-Statistic [3]: 7.98539 Prob(>F), (3,153) degrees of freedom: 0.000056*
Joint Wald Statistic [4]: 38.129483 Prob(>chi-squared), (3) degrees of freedom: 0.000000*
Koenker (BP) Statistic [5]: 3.634814 Prob(>chi-squared), (3) degrees of freedom: 0.303694
Jarque-Bera Statistic [6]: 904.567214 Prob(>chi-squared), (2) degrees of freedom: 0.000000*

171
Table 5.18
Community Garden /OLS Regression Results for Philadelphia Model III
Variable Coefficient StdError t-Statistic Probability Robust_SE Robust_t Robust_Pr VIF [1]
Intercept 0.023363 0.289957 0.080573 0.935811 0.027672 0.844261 0.399054 --------
PCT_WHITE 0.001444 0.004179 0.345604 0.729848 0.001403 1.029453 0.303926 10.456121
PCT_BLACK 0.002192 0.003971 0.55189 0.581362 0.001589 1.379542 0.168567 9.844124
PCT_HISPAN 0.010931 0.004794 2.279959 0.023162* 0.010197 1.071957 0.284427 2.125054
PCT_ASIAN -0.005559 0.007951 -0.699166 0.48488 0.005417 -1.026138 0.305484 1.357106
PCT_OTHER 0.107931 0.130564 0.826652 0.408952 0.130672 0.825971 0.409338 1.022161
PCT_TENURE -0.010292 0.002924 -3.519502 0.000499* 0.003199 -3.217201 0.001421* 1.486032
SQ_POP_DEN 0.054917 0.018111 3.032266 0.002607* 0.015981 3.436473 0.000669* 1.394887
SQ_MED_INC 0.000374 0.001325 0.281986 0.778118 0.000771 0.484897 0.628047 2.115325

Number of Observations: 381 Number of Variables: 9
Degrees of Freedom: 372 Akaike's Information Criterion (AIC) [2]: 1025.359169
Multiple R-Squared [2]: 0.086996 Adjusted R-Squared [2]: 0.067361
Joint F-Statistic [3]: 4.43075 Prob(>F), (8,372) degrees of freedom: 0.000039*
Joint Wald Statistic [4]: 40.186985 Prob(>chi-squared), (8) degrees of freedom: 0.000003*
Koenker (BP) Statistic [5]: 23.340159 Prob(>chi-squared), (8) degrees of freedom: 0.002954*
Jarque-Bera Statistic [6]: 71262.75367 Prob(>chi-squared), (2) degrees of freedom: 0.000000*

Table 5.19
Community Garden /OLS Regression Results for Seattle Model IV
Variable Coefficient StdError t-Statistic Probability Robust_SE Robust_t Robust_Pr VIF [1]
Intercept -13.831908 5.037671 -2.745695 0.006999* 5.617188 -2.462426 0.015262* --------
PCT_WHITE 0.144656 0.052283 2.766812 0.006588* 0.05828 2.482087 0.014486* 497.509073
PCT_BLACK 0.167149 0.054197 3.084073 0.002558* 0.061633 2.712011 0.007703* 102.377671
PCT_HISPAN 0.154472 0.056998 2.710146 0.007743* 0.06255 2.469585 0.014976* 20.098886
PCT_ASIAN 0.149078 0.055378 2.692008 0.008150* 0.061758 2.413919 0.017338* 165.860573
PCT_OTHER 0.200436 0.086593 2.314694 0.022381* 0.09165 2.186971 0.030745* 2.868984
MED_INC_99 -0.000003 0.000007 -0.45205 0.652085 0.000006 -0.554231 0.580492 5.293706
PCT_TENURE 0.001588 0.00585 0.271502 0.786492 0.005215 0.304531 0.761276 3.872055
PCT_LDCV21 0.00382 0.001796 2.126497 0.035575* 0.001671 2.285518 0.024091* 1.174996
PCT_LDCV22 -0.011378 0.012565 -0.905492 0.367073 0.011553 -0.984808 0.326761 1.056778

Number of Observations: 126 Number of Variables: 10
Degrees of Freedom: 116 Akaike's Information Criterion (AIC) [2]: 233.400427
Multiple R-Squared [2]: 0.180941 Adjusted R-Squared [2]: 0.117393
Joint F-Statistic [3]: 2.847329 Prob(>F), (9,116) degrees of freedom: 0.004570*
Joint Wald Statistic [4]: 21.982537 Prob(>chi-squared), (9) degrees of freedom: 0.008934*
Koenker (BP) Statistic [5]: 15.296736 Prob(>chi-squared), (9) degrees of freedom: 0.083101
Jarque-Bera Statistic [6]: 11.023661 Prob(>chi-squared), (2) degrees of freedom: 0.004039*

172
Conclusion

As this chapter has shown, city parks and community gardens are not evenly
distributed across a city’s landscape, but not in the way that one might expect. Parks are
highly clustered in 78% of cities, while community gardens cluster in 65% of cities. How
do cities compare in terms of region and population? Table 5.20 arrays the results of
several spatial statistical analyses for parks, organized by region. I expected the Sun Belt
cities to have the most parks, given the climate and the history of development that
focused on incorporating green spaces (Nicolaides, 2003). However, the six cities that
represent various other regions far outnumber the Sun Belt cities with 666 parks. Really,
the cities in this study in the Western half of the United States host 60% of the parks,
which might be expected given that land is more expansive and density often lower in the
West. The mean centers for parks in the Rust Belt cities fell in minority tracts in only
three cities, but all six mean center tracts were low income. In this respect, poor people
may not be under-represented in terms of where city parks are located. In the Sun Belt
cities, the mean center fell in minority tract in three cities, as well, and fell in low- to
lower-middle income tracts in only four cities. Minority tracts were not well-represented
in the mean centers for the Other cities, but in five of six Other cities, the mean centers
fell in low-income tracts. Again, racial and ethnic minorities may be under-represented,
but low-income populations may not be. Looking at the Moran’s I results and the
Ripley’s K results, it is evident that parks tended to cluster globally. What’s surprising,
however, is that park hot spots registered in only half of the cities in the study. The most
startling pattern is that of the nine cities with park hot spots, the mean minority

173
percentage only exceeded 50% in three cities. This would confirm the mean center
results, where only seven cities registered mean centers in minority tracts. According to
the hot spot results, the average median income ranged from $25,626 to $83,115. Recall
that cold spots registered in Salt Lake City and Seattle, in white and upper-middle income
areas.

174

Table 5.20
Park/City Comparison by Region
Parks Persons/Park Mean Center Moran's I Ripley's K Getis-Ord Gi* (Hot)
RUST BELT
Philadelphia PA 79 19,209 90% Minority, $14,826 clustered clustered 41% Minority, $35,594
Boston* MA 249 2,366 79% Minority, $36,010 clustered clustered none
Grand Rapids MI 76 2,603 27% Minority, $26,250 clustered clustered 34% Minority, $35,489
Dayton, OH 33 5,036 53% Minority, $8,253 clustered random 27% Minority, $25,626
Decatur IL 40 2,053 36% Minority, $27,109 clustered random none
Albany* NY 35 2,733 23% Minority, $26,419 clustered clustered none
SUN BELT
San Diego CA 73 16,758 27% Minority, $49,842 clustered clustered 77% Minority, $25,751
Austin* TX 144 4,558 22% Minority, $28,842 clustered clustered 51% Minority, $31,130
Miami FL 39 9,296 89% Minority, $12,967 clustered random none
Little Rock* AK 52 3,530 82% Minority, $20,938 clustered random none
Santa Fe* NM† 26 2,377 74% Minority, $26,552 clustered clustered none
Flagstaff AZ 23 2,310 19% Minority, $35,592 clustered random none
OTHER
San Jose CA 146 6,123 69% Minority, $43,636 clustered clustered 56% Minority, $83,115
Seattle WA 328 1,718 37% Minority, $21,605 clustered clustered 44% Minority, $56,394
Norfolk VA 32 7,325 97% Minority, $16,698 clustered clustered none
Salt Lake City* UT 65 2,792 39% Minority, $16,048 random clustered 19% Minority, $41,681
Boulder CO 78 1,212 9% Minority, $28,326 mild cluster clustered none
Charleston* WV 17 3,129 48% Minority, $22,206 random clustered 26% Minority, $33,431

*Capital
†Mean center in same tract as community garden mean center.

175

Table 5.21 compares parks by population. The six largest cities boast 67% of the
parks in the study. However, the best persons-to-parks ratio belongs to Boulder, where
there is one park for every 1,212 people. This may be due to Boulder being nestled
against the Front Range of the Rocky Mountains, which drives a strong interest in
recreation and environmental amenities. Philadelphia and San Diego have two of the
worst persons-to-parks ratios, with their populations well over 1 million and only 79 and
73 parks respectively.
15
Only three of the six largest cities have their mean centers in
minority tracts, though the lower-middle income range is represented. Four of six
medium-sized cities boasted mean centers in minority tracts, while all mean centers were
in low-income tracts. Of the small cities, only Santa Fe registered a mean center in a
minority tract, though the small cities represented lower-middle income residents well.
Though global clustering was evident in most of the cities, hot spots registered in five of
six big cities, with three being in minority areas. Hot spots did not register in five of the
six small cities, even though the global measures indicated clustering. While the
regression results were disappointing, particularly in terms of the low R
2
, the pattern of
park location vis-à-vis minority and poor populations may be better explained by
variables other than those included in traditional environmental equity analyses. I explore
additional explanatory variables in Chapter 7.
In all, the results presented here indicate that in most of the cities in the study,
racial and ethnic minorities are under-represented near parks; poor people are as well, but
to a lesser degree. Boston stands out as a city where minority and poor people are well-

15
One caveat about these ratios is that I only included city parks in my study; San Diego and Philadelphia
could have additional county, state, and national parks available to residents.

176
represented near parks, while Seattle—even with its great persons-to-parks ratio—under-
represents minority and poor people near parks.

177

Table 5.21
Park/City Comparison by Population
Population Parks Persons/Park Mean Center Moran's I Ripley's K Getis-Ord Gi* (Hot)
Large Cities
Philadelphia PA 1,517,550 79 19,209 90% Minority, $14,826 clustered clustered 41% Minority, $35,594
San Diego CA 1,223,341 73 16,758 27% Minority, $49,842 clustered clustered 77% Minority, $25,751
San Jose CA 893,889 146 6,123 69% Minority, $43,636 clustered clustered 56% Minority, $83,115
Austin* TX 656,302 144 4,558 22% Minority, $28,842 clustered clustered 51% Minority, $31,130
Boston* MA 589,141 249 2,366 79% Minority, $36,010 clustered clustered none
Seattle WA 563,375 328 1,718 37% Minority, $21,605 clustered clustered 44% Minority, $56,394
Medium Cities
Miami FL 362,563 39 9,296 89% Minority, $12,967 clustered random none
Norfolk VA 234,403 32 7,325 97% Minority, $16,698 clustered clustered none
Grand Rapids MI 197,800 76 2,603 27% Minority, $26,250 clustered clustered 34% Minority, $35,489
Little Rock* AK 183,558 52 3,530 82% Minority, $20,938 clustered random none
Salt Lake City* UT 181,456 65 2,792 39% Minority, $16,048 random clustered 19% Minority, $41,681
Dayton, OH 166,179 33 5,036 53% Minority, $8,253 clustered random 27% Minority, $25,626
Small Cities
Albany* NY 95,658 35 2,733 23% Minority, $26,419 clustered clustered none
Boulder CO 94,510 78 1,212 9% Minority, $28,326 mild cluster clustered none
Decatur IL 82,113 40 2,053 36% Minority, $27,109 clustered random none
Santa Fe* NM† 61,805 26 2,377 74% Minority, $26,552 clustered clustered none
Charleston* WV 53,196 17 3,129 48% Minority, $22,206 random clustered 26% Minority, $33,431
Flagstaff AZ 53,137 23 2,310 19% Minority, $35,592 clustered random none

*Capital
†Mean center in same tract as community garden mean center.

178

Turning to community gardens, Table 5.22 compares the cities by region. The
Rust Belt boasts the most gardens, which is not surprising given the historical importance
of community gardens and also the higher densities of the Rust Belt cities, particularly in
the Northeast. Boston had the best persons-to-garden ratio, with 1 garden for every 3,981
people. While Albany had a similar ratio, the worst ratio occurred in San Diego where
only 1 garden exists for every 101,945 people. This is not surprising given the fact that
people are more land consumptive in the West and in Sun Belt cities, and in the
temperate climates, people may be more inclined to tend personal gardens. However, the
mean centers in the Sun Belt cities fell in minority tracts half of the time, versus only one
third of the time in the Rust Belt region and Other cities. Overall, with the exception of
San Jose, most mean centers fell in lower-middle to middle income tracts. Due to the
small number of gardens in most cities, the global and local clustering tools did not work
particularly well.

179

Table 5.22
Community Garden/City Comparison by Region
Gardens Persons/Garden Mean Center Moran's I Ripley's K Getis-Ord Gi* (Hot)
RUST BELT
Philadelphia PA 100 15,176 28% Minority, $41,536 clustered clustered 73% Minority, $22,055
Boston* MA 148 3,981 96% Minority, $27,165 clustered (N/S) clustered 37% Minority, $41,285
Grand Rapids MI 7 28,257 37% Minority, $25,496 clustered 15% Minority, $36,783
Dayton, OH 18 9,232 89% Minority, $19,144 clustered dispersed none
Decatur IL 6 13,686 48% Minority, $20,218 clustered none
Albany* NY 20 4,783 41% Minority, $23,198 clustered none
SUN BELT
San Diego CA 12 101,945 34% Minority, $27,938 clustered 53% Minority, $30,651
Austin* TX 7 93,757 94% Minority, $27,619 clustered clustered 73% Minority, $25,658
Miami FL 22 16,480 87% Minority, $29,519 mild cluster clustered none
Little Rock* AK 9 20,395 80% Minority, $27,468 random none
Santa Fe* NM† 5 12,361 74% Minority, $26,552 clustered none
Flagstaff AZ 4 13,284 17% Minority, $35,592 clustered (N/S) none
OTHER
San Jose CA 19 47,047 82% Minority, $81,916 random clustered 33% Minority, $73,918
Seattle WA 56 10,060 19% Minority, $32,389 clustered 57% Minority, $47,087
Norfolk VA 6 39,067 35% Minority, $44,716 clustered dispersed 14% Minority, $50,398
Salt Lake City* UT 7 25,922 51% Minority, $23,125 mild cluster dispersed none
Boulder CO 8 11,814 27% Minority, $35,853 clustered none
Charleston* WV 2 26,598 30% Minority, $22,206 error error

*Capital
†Mean center in same tract as park mean center.

180

Examining the cities in terms of population (see Table 5.23), the hot spot analysis
worked in all six of the largest cities and of those six, average minority percentage
exceeded 50% only in three cities. Income ranged from $22,055 to $73,918. The mean
centers of the six medium-sized cities fell in minority tracts in four of six cities, and low
to lower-middle income tracts in all five tracts. According to the mean centers, minorities
were under-represented in the smaller cities, though poor people may be better-
represented in small cites. These results suggest that minorities and poor people are
under-represented where community gardens are concerned overall, though the weak
regression results suggest that additional explanatory variables need to be modeled.

181

Table 5.22
Community Garden/City Comparison by Population
Population Gardens Persons/Garden Mean Center Moran's I Ripley's K Getis-Ord Gi* (Hot)
Big Cities
Philadelphia PA 1,517,550 100 15,176 28% Minority, $41,536 clustered clustered 73% Minority, $22,055
San Diego CA 1,223,341 12 101,945 34% Minority, $27,938 clustered 53% Minority, $30,651
San Jose CA 893,889 19 47,047 82% Minority, $81,916 random clustered 33% Minority, $73,918
Austin* TX 656,302 7 93,757 94% Minority, $27,619 clustered clustered 73% Minority, $25,658
Boston* MA 589,141 148 3,981 96% Minority, $27,165 clustered (N/S) clustered 37% Minority, $41,285
Seattle WA 563,375 56 10,060 19% Minority, $32,389 clustered 57% Minority, $47,087
Medium Cities
Miami FL 362,563 22 16,480 87% Minority, $29,519 mild cluster clustered none
Norfolk VA 234,403 6 39,067 35% Minority, $44,716 clustered dispersed 14% Minority, $50,398
Grand Rapids MI 197,800 7 28,257 37% Minority, $25,496 clustered 15% Minority, $36,783
Little Rock* AK 183,558 9 20,395 80% Minority, $27,468 random none
Salt Lake City* UT 181,456 7 25,922 51% Minority, $23,125 mild cluster dispersed none
Dayton, OH 166,179 18 9,232 89% Minority, $19,144 clustered dispersed none
Small Cities
Albany* NY 95,658 20 4,783 41% Minority, $23,198 clustered none
Boulder CO 94,510 8 11,814 27% Minority, $35,853 clustered none
Decatur IL 82,113 6 13,686 48% Minority, $20,218 clustered none
Santa Fe* NM† 61,805 5 12,361 74% Minority, $26,552 clustered none
Charleston* WV 53,196 2 26,598 30% Minority, $22,206 error error
Flagstaff AZ 53,137 4 13,284 17% Minority, $35,592 clustered (N/S) none

*Capital
†Mean center in same tract as park mean center.

182

The results of the amenity analysis belie two weaknesses. First, city planners
generally aim for 2.5 acres of parkland per 1,000 people, but measuring park acreage for
the 1,535 parks in this study was beyond the scope of this dissertation. This should be
addressed in further studies. Second, the record-keeping for community gardens is woeful
at best in several cities. I triangulated methods to locate all community gardens, but field
research in 18 cities was beyond the scope of this dissertation. Further study of
community gardens should include site visits to cities for a more complete listing of
community gardens. Further studies should also include a layer of cadastral data to
determine who owns the community gardens. Identifying ownership and long- and short-
term leases would provide a better understanding of how community gardens are
managed, and if they are community-driven or fostered by nonprofit organizations
located outside of the community. This information might provide important clues as to
how well minority and/or poor people are represented.

183
CHAPTER SIX: INTRODUCTION TO POLITICAL ANALYSIS

The dominant beliefs—those of political and economic
entrepreneurs in a position to make policies—over time result in the
accretion of an elaborate structure of institutions that determine
economic and political performance.
Douglass North, Understanding the Process of Economic Change

This chapter provides an introduction to Phase II of my study: the analysis of the
political institutional structures at work in the study’s cities. In Phase I, I mapped the
distribution of environmental disamenities and amenities in relation to minority and poor
populations to explore the manifestation and pattern of hazards and amenities in each city
and to understand the nature of environmental equity problems. In Phase II, I investigated
the political institutional variables that may help explain the equity patterns in each city.
The Phase II analysis was designed to answer the question: What are the city-level
political processes and institutional variables that are more likely to be associated with
better disamenities and amenities distributions? This question is important in pushing
beyond the traditional racial and socioeconomic explanations of environmental inequity.
If only racism or classism (institutional or intentional) were the root causes of
environmental inequity, then hypothetically, the distributions might not vary much from
city to city. Further, today’s urban centers are facing similar challenges from
globalization, migration of jobs, deindustrialization, immigration, white flight,
decentralization of metropolitan areas, decreased housing affordability, and budget
constraints. That said, if all cities face those issues to some degree, why does one city

184
have a better equity distribution than the next? To answer this question, I used the
political process approach of the Social Movements Framework, which is useful for
studying meso-level structures to understand the impact of those structures on public
policy (Kriesi, 2007).
I begin with a brief review of the usual explanations for environmental equity
outcomes and a brief synopsis of the major schools of thought on urban politics. Then, I
briefly review the Social Movements Framework and detail the political processes
approach of the framework, describing the city-level processes and structures that can
affect governance and policy outcomes. Next, I describe the methods used in Phase II,
specifying variables, propositions, and data collection.

Environmental Justice and Policymaking in Cities: The Usual Suspects

As I discussed in Chapter 2, several explanations for environmental equity
outcomes have been offered and tested in the literature. Early on, researchers opted for
the “environmental racism” explanation; that is, environmental equity outcomes were due
to intentional or institutional racism (c.f., Bullard, 1983; UCC, 1987; Pinderhughes,
1996; Pellow, 2007). Some claimed intentional racism in siting decisions (c.f., Novotny,
2000) or the intentional exclusion of racial minorities in decision-making (c.f.,
Rechtschaffen & Gauna, 2002). Intentional racism proved very difficult to confirm, but
the institutional racism claim gained steam. Some researchers elaborated this idea by
examining housing policies and racial segregation (c.f., Liu, 1997; Hite, 2000; Boone,
2002) and normal socioeconomic processes like unregulated development (c.f., Szazs &

185
Meuser, 2000) as more nuanced explanations of institutionalized racism. Other
researchers were not convinced by these arguments or research supporting them and
sought different explanations. In a landmark study, Been (1994, p. 1383) examined
“minority move-in,” finding that minorities may have moved to hazardous areas for
cheaper housing, the job market, or because of transportation limitations. Other
researchers focused on postwar social arrangements (c.f., Hurley, 1995) and the historical
development of industry and transportation (c.f., Bolin, Grineski, & Collins, 2005). Still
others examined neighborhood change (c.f., Boone & Modarres, 1999). No single
explanation has dominated the discourse or can fully explain environmental inequity.
Building on the urban politics and city policymaking literature, I attempt to derive
additional explanations. First, however, I briefly review the major schools of thought in
urban politics, which began to take shape around the pluralist ideas of Dahl (1961) and
Banfield (1961). The pluralist explanation of urban politics argued that myriad groups
and coalitions may command resources and drive the policy agenda at any given time,
rather than a single group of powerful economic and political elites. This view was
challenged by various structuralist explanations of urban politics (Mollenkopf, 1994).
While some structuralists focused on power (c.f., Bachrach & Baratz, 1962) or a neo-
Marxist critique of capital (c.f., Harvey, 1985), the public choice school of urban politics
came to dominate. Tiebout’s (1956) consumer-voter theory argued that voters will seek
the cities that best satisfy the goods and services they desire, creating variance among
cities as they work to retain and attract voters. Lindblom (1977) added that businesses
had major influence over politics and even participated “as functionaries performing
functions that government officials regard as indispensable,” (p. 175). These ideas helped

186
fueled the idea that rather than politics, economics drove city policy decisions. Peterson
(1981) proclaimed the death of the theory of federalism and argued that city government
sought to maintain and develop its economic base, leaving redistributive or allocation
policies to the federal government. Calling city politics “mostly limited” (p. 4), he argued
that cities will follow whatever policies “maintain or enhance the economic position,
social prestige, or political power of the city, taken as a whole,” (p. 20). Peterson’s views,
while dominant, have been criticized. Though most would agree with Lindblom (1977)
and Elkin (1987) that cities are driven by economic interests, many argued that Peterson
dismissed the social and political structure of the city (Mollenkopf, 1994) and portrayed
the stranglehold of economics on cities as a prison (Swanstrom, 1988). Calling Peterson’s
theory “incomplete,” Swanstrom argued that “economic pressures are always mediated
by political structures and political competitions” (1988, p. 107) and noted that the
political leadership and vision of a city matter as well. Stone (1989) also questioned “the
privileged position of capital” (p. 6) in urban politics, noting that regimes, or the
coalitions between public and private parties, matter in urban politics along with
resources and agenda. Adding to the idea that economics and politics matter, Sharp
(1997) noted that Peterson failed to account for some social policies and claims that “are
not as readily understood in terms of conflict over shares of divisible benefit” (p. 278).
One thing that is commonly understood is that in addition to succumbing to
economic pressures, most cities derive their power and political structure from the state
because the U.S. Constitution speaks to either a national government or a state

187
government.
16
As such, many have noted that cities are “very weak institutions” (Rae,
2006, p. 280); that city policymaking is “exceedingly fragile” (Waste, 1989, p. vii); that
local policies are subordinate to state preferences (McCabe & Feiock, 2005); and yet,
“state governments have rarely stepped in to develop more extensive urban policies of
their own,” (Sellers, 2002, p. 622). Thus, as these local “islands” struggle to enact
policies, raise revenues, and attract economic development, they might pursue economic
and zoning policies that harm or exclude residents who require reallocation policies and
services or forego more costly policies like environmental equity (c.f., Dreier,
Mollenkopf, & Swanstrom, 2004; Sellers, 2002; Stein, 1990; Elster, 1992).
How might these dominant theories of urban politics and power inform
environmental equity? Peterson might argue that environmental equity is a policy issue,
like immigration or import control, that is not within the purview of city governments. He
might also argue that cities would not trade economic development and stability for more
rigid zoning and pollution control laws, or for reallocation of housing for people in
hazard-prone areas. Stone might argue that environmental equity would need to be a goal
of the governing regime and on the agenda of the leading coalitions to see any progress
made. Lindblom might say that if polluting industries fueled the local economy, then
those industries might be favored over any political claims of justice. Tiebout might
argue that those who favor environmental equity might seek out cities with environmental
justice programs.

16
Home-rule cities (e.g., Boulder, Grand Rapids, Philadelphia) are different in that their charters offer them
broad powers of determination for their structure, elections, and balloting (c.f., Morgan, England, &
Pelissero, 2007).

188
While many of the above environmental equity and city politics explanations have
merit, researchers have yet to achieve consensus or offer a definitive explanation of
environmental inequity in cities; yet cities are on the environmental justice frontlines.
Environmental equity is a complex policy problem that may have multiple explanations
borne by studies involving intricate concepts and measurements. Yet, environmental
inequity—an intractable policy problem—may require broader explanations that capture
residual phenomena at work in cities. The Social Movements literature, particularly the
political process approach, offers several theoretical concepts that may get at those
residual phenomena and alternative explanatory factors. I do not suggest that the Social
Movements literature should replace the existing explanations for environmental
inequity. Rather, the Social Movements literature and its policy process approach (often
called the political opportunity structure) may add to our understanding of environmental
injustice—and options for governability—because the underlying structural variables
might interact with other process variables. That is, the Social Movements literature may
augment the “usual suspects” in the environmental justice literature, and focus the
discussion on the governability of a seemingly intractable problem.

Theoretical Framework: Social Movements

As I discussed in Chapter 2, the environmental equity literature is not a concept
literature per se, with an overarching, self-identified theory. Studies are driven by
multiple theories from various disciplines and fields. The extant corpus is replete with
methodologically complex studies that map the location of hazards and the affected

189
populations with precision. Yet, many studies stop short of focusing on the attendant
governance mechanisms that may affect hazard distributions. Several critics have noted
this, calling for a broader framework: “simply demonstrating that there is environmental
inequality…turns out not to be enough. We need to have a broader narrative, a broader
story, a broader way of thinking about these things in a way that captures people’s
imaginations,” (Pastor, 2001, p. 9; see also Bowen, 2002; Pellow & Brulle, 2005). To
broaden the narrative of environmental equity and to take my distribution study further, I
used the Social Movements Framework’s political process approach in Phase II.
The study of social movements is essentially the study of power in society and the
organized, collective attempts to change the balance of power. Rather than a civil war or
coup, a social movement is “a sustained challenge to power holders” by claims-making
groups who recruit “third parties such as other power holders, repressive forces, allies,
competitors and the citizenry as a whole,” (Tilly, 1999, p. 257, emphasis original; see
also Tarrow, 1998). Social movements unite independent groups and actors who have
“common goals, targets, and ideology” to carry out sustained actions for social change
(Jenkins & Form, 2005, p. 332). These actors or groups may be marginalized and
formerly voiceless actors who join together in some organized effort. Social change
ranges from short-term decisions, changes in political structure or decision making, to
“long-term changes in the distribution of socially valued goods,” (Jenkins, 1983, p. 544).
The actors reject “conventional political means” in their pursuit of either value-oriented
goals that represent fundamental shifts in the values of society’s institutions or norm-
oriented goals that represent narrower and more concrete changes in social structures
(McAdam & Snow, 1997). The state-related effects of social movements may include the

190
access to and “continuing leverage over political processes” (Amenta & Caren, 2007, p.
464) by formerly disadvantaged groups; and specific and/or overarching changes in
policy and implementation that benefit the group (Amenta & Caren, 2007). Social
movements can change the public discourse about an issue and may act as an agenda-
setting mechanism (Dryzek, Downes, Hunold, & Schlosberg, 2003) as well.
Movement actors view social problems as rooted in the inherent power
asymmetries of the social structure. Environmental justice is one such problem. Beck
(1992) describes environmental problems as social problems:

Environmental problems are not problems of our surroundings, but – in
their origins and through their consequences – are thoroughly social
problems, problems of people, their history, their living conditions, their
relation to the world and reality, their social, cultural and political
situations. (p. 81, emphasis original)

That is, environmental problems originate in the social structure of society (Dobson,
1993). Power—or lack thereof—in determining who ultimately bears the burden of
environmental disamenities and who reaps the rewards of environmental amenities is a
root cause of environmental inequity. At play are governments, policy entrepreneurs,
corporate power brokers, nonprofit groups, activists, and residents in neighborhoods that
house environmental bads and goods. Usually the costs are borne by society’s racial and
ethnic minorities and poor populations. To improve environmental equity means
changing the balance of power in society and thereby resolving the attendant
social/environmental ills. Several authors have acknowledged the Environmental Justice
Movement’s (EJM’s) importance as a social movement (c.f., Faber, 2008; McCarthy &

191
King, 2005) and critiqued the EJM (c.f., Dryzek, Downes, Hunold, & Schlosberg, 2003;
Cable, Mix, & Hastings, 2005; Pellow & Brulle, 2005), yet few have incorporated GIS
analysis in their work.

The History and Major Theories of Social Movements

The study of social movements emerged from an attempt by sociologists in the
classical tradition to understand the differences between institutional forms of collective
action versus non-institutional forms, with a focus on the latter (de Souza Medeiros,
2008). Similarly, the Chicago School focused on meaning and communication among
movement participants as a way to study collective behavior. These early treatments are
sometimes referred to as the “symbolic interactionist theories of collective behavior”
(Jenkins & Form, 2005, p. 335). From there, the functionalists or structural functionalists
focused on the social and cultural systems driving collective action and the abilities and
vulnerabilities of social structures to norm-oriented or value-oriented change (Jenkins &
Form, 2005; McAdam & Snow, 1997; de Souza Medeiros, 2008). Interestingly, Tarrow
argues that the study of power and social change emanated long before, in the works of
Marx and Engels, Lenin, and Gramsci: “though they are seldom made explicit, these
three elements of Marxist theory have strong parallels in recent theorizing about
collective action and social movements” (1998, pp. 13-14).
The current theories driving the Social Movements Framework are the only ones
that “actually discuss social movement change” (Jenkins & Form, 2005, p. 335). They
are: the political process approach (often called the political opportunity structure),

192
resource mobilization theory, and framing/cultural processes (Klandermans &
Staggenborg, 2002; McAdam & Snow, 1997; McAdam, McCarthy, & Zald, 1996;
Jenkins & Form, 2005; McAdam, 1999). Generally speaking, political opportunity
structures are defined as the political organization (or environment) that promotes or
inhibits collective action. The political opportunity structure affects “the degree to which
groups are likely to be able to gain access to power and to manipulate the political
system,” (Eisinger, 1973, p. 25). Social movements are shaped by the environmental
structure in which they emerge and are path-dependent in that their outcomes are often
“the product of specific sequences of political interactions between social movements,
their allies, opponents, and authorities,” (Jenkins & Form, 2005, p. 333). Social
movement success is also a function of the resources mobilized (Klandermans &
Staggenborg, 2002). Resource mobilization theory concerns: acquiring resources,
managing and controlling resources, and mobilizing resources into collective-action
outputs (Jenkins, 1983). Such resources include financial ones, but also include
volunteers, publicity, leadership, issue recognition, and expertise (Jenkins & Form,
2005). Though resources come from many sources, foundations are widely acknowledged
as being key funders of social movements (Ostrander, 2005; Faber & McCarthy, 2005).
Challenging the more rationalistic foundations of resource mobilization theory and the
deterministic foundations of political process theory is the more sociocultural frame
analysis or framing process perspective of Social Movements (McAdam & Snow, 1997).
It is here that researchers acknowledge the importance of cultural discourse and frames in
movement formation and success, studying “the relationship among framing, collective
identity, and ideology” (Jenkins & Form, 2005, p. 341). Whereas political structure and

193
resource mobilization theories exclude key elements of culture and discourse, this branch
of social movements theory debates processes like “creativity in movements” and “the
interplay between culture and structure,” but has serious methodological issues that have
yet to be resolved (Klandermans & Staggenborg, 2002, p. xii). Frame and cultural
constructionism are hard to analyze because of the difficulties in identifying cultural and
ideological changes empirically without “sustained interactions” in longitudinal research
designs (Klandermans, Staggenborg, & Tarrow, 2002, p. 330). The importance of frame
is well-established (c.f., Brulle & Jenkins, 2006; McAdam & Snow, 1997; Faber, 2008),
but frame is difficult to analyze: “it is impossible to disentangle the effects of identity
construction from collective action because at this [collective] level they are inseparably
and simultaneously constructed” (Jenkins & Form, 2005, p. 341).
The Social Movements Framework is theoretically rich and complex though is not
without its critics, many from within the body of theorists that developed it. Much of the
attention of movement researchers has been on the emergence of social movements and
the participation of individuals and organizations in collective action. Jenkins and Form
note that “whether and how [social movements] actually cause social change has received
little attention,” (2005, p. 331, emphasis original). Others have noted this as well (c.f.,
Cable, Mix, & Hastings, 2005; Armenta & Young, 1999; Giugni, 1999). One of the
impediments to studying movement outcomes is the difficulty in separating movement
actors and actions from non-movement actors and actions, and non-institutionalized
efforts from institutionalized efforts—the boundaries between which are “hard to draw
with precision,” (Klandermans, Staggenborg, & Tarrow, 2002, p. 336). Really, this is a
problem of causality or “how to establish a causal link between a given movement and an

194
observed change,” (Giugni, 1999, p. xxiv). The difficulty in showing a direct impact of a
social movement is hampered by spuriousness (McAdam & Snow, 1997; Amenta &
Caren, 2007); the lack of “analytical clarity about the term impact” (Armenta & Young,
1999, p. 22, emphasis original); and the lack of research on countermovements and non-
movement actors (Giugni, 1999). Critics note that although the study of social
movements is essentially the study of power, much social movement research fails to
document changes in society’s power dynamics (Jenkins & Form, 2005; Giugni, 1999).
Finally, the inward focus on policy outcomes often misses a movement’s “possible
impact on social change in the broader society,” (Jenkins & Form, 2005, p. 331). One
such impact or social phenomenon that the Social Movements Framework may miss is
social capital. Others have touched on this (c.f., de Souza Medeiros, 2008), but this is not
a concept often seen in literature. Three competing strains of social capital exist:
Coleman’s rational strain, Putnam’s democratic strain, and Bourdieu’s Marxist strain
(c.f., Lewandowski, 2005; Ramos-Pinto, 2005). Most definitions agree that social capital
consists of networks and social norms in and among communities that promote social
cohesion and collective action. Woolcock (2001) distinguishes between bonding and
bridging social capital, the former being the social capital that exists among family
members and close friends (see also Putnam, 2000; Wagner & Fernandez-Gimenez,
2008). Bridging social capital exists among broader and more distant friends or associates
and exists among communities and community groups. Woolcock also highlights linking
social capital, an important form of social capital that vertically links groups in society
and offers “the capacity to leverage resources, ideas and information from formal
institutions beyond the community,” (2001, p. 13). It is in these community networks of

195
associations, or the meso-level, where most action in social movements activities occur
(McAdam, McCarthy, & Zald, 1988). Social movements rely on networks that can
facilitate communication, provide access to a large number of people, and perhaps even
pressure potential participants (Diani, 2003), something that bridging and linking social
capital can provide. As well, social movements require nonprofit organizations or social
movement organizations for sustained activities. Social capital can have a positive impact
on “a community’s voluntary sector and, in particular, on the foundings of new nonprofit
organizations,” (Saxton & Benson, 2005, p. 19). On an individual level, social capital
provides the trust and norms of reciprocity needed for individuals to join in collective
action. The way social movements mobilize actors “is to generate solidarity and moral
commitments to the broad collectives in whose name movements act,” (Jenkins, 1983, p.
538). Social capital allows formerly independent and marginalized actors and groups to
unite because it provides the “solidarities and moral commitment” (Jenkins, 1983, p. 537)
necessary for collective action (see also McAdam, 2003). This is clearly an area for
future development in the Social Movements Framework.

Political Process Approach

As rich as social movement analysis can be, a thorough analysis of the EJM is beyond
the scope of this dissertation. Still, the political process approach is very useful in
identifying political institutional variables to study because it provides “a general
conceptual toolkit” (Kriesi, 2007, p. 69). The elements of the political process
approach—which mostly relate to political structure—can be selected and combined for

196
the appropriate level of analysis, and “the approach is well-suited for the integration of
the study of social movements and public policy analysis (Kriesi, 2007, p. 85).
Furthermore, the political process approach allows researchers to do “an institutional
analysis of the political system framed in terms of interorganizational networks and the
interaction of collective action and political opportunities,” (Jenkins & Form, 2005, p.
332). Thus, this approach helps identify the institutional structures and political variables
that constrain the opportunities of actors seeking environmental justice.
As noted above, the political process model emerged from the structural functionalist
strain of Social Movement theories, somewhat in response to resource mobilization
theory. According to critics, the latter failed to capture the political context, including the
governance structure, political alliances, and the changing political opportunities, within
which mobilization occurred and challenges were sustained (c.f., Jenkins, 1983;
Klandermans & Staggenborg, 2002). Where resource mobilization theory worried that
actors would lack resources to mount challenges, the political process approach argued
that what was more important to challengers was a changing or favorable political
environment in which political actors had their own sets of incentives and interests
(Jenkins & Perrow, 1977; Giugni, 1999). These realizations followed Eisinger’s 1973
study of political protest in cities, in which he argued that protests are facilitated or
constrained by political factors such as if the political system was open or closed, what
type of chief executive was in place, and the nature of election of city representatives.
The two theorists often credited with elaborating the political process approach are
Tilly (1978) and McAdam (1982, cited as 1999 in this study). Tilly’s contribution was to
focus on the probability of success based on the opportunities or threats the political

197
structure presented to challengers (often referred to as the political opportunity structure).
McAdam added to this a focus on political processes, or the changing alignments of
political alliances, pressure from political elites, or events (e.g., wars, economic crises)
that create the political opportunities needed for challenges to succeed (McAdam, 1983,
1999; McAdam & Snow, 1997). He then synthesized the ideas of process and opportunity
(Tarrow, 1998). McAdam’s crucial ingredients of the political process model were: the
organizational strength of challengers (i.e., “the resources of the minority community that
enable insurgent groups to exploit these political opportunities,” 1999, p. 42); the level of
consciousness among challengers (i.e., “cognitive liberation,” 1999, p. 48); the political
opportunities facing challengers; and the broad socioeconomic processes at work.
The political opportunity structure includes opportunities and constraints, though
it should not be viewed “as an invariant model producing social movements, but as a set
of clues for when contentious politics will emerge,” (Tarrow, 1998, p. 20). Constraints
are potential threats to challengers, meaning how repressive or aggressive a government
might be toward challengers. Eisinger (1973) viewed political systems as curvilinear in
terms of being open to challenges versus being closed and repressive, arguing that most
challenges take place in intermediate systems, i.e., in the bottom of the U-shaped curve.
Constraints also include the stability of the polity, elite alliances within the polity, and the
polity’s “capacity and propensity for repression,” (McAdam, 1996, p. 27; see also
McAdam, McCarthy, & Zald, 1996; Tarrow, 1998). Four factors can add to the openness
of the polity (Kitschelt, 1986): (1) the greater the number of political parties and groups
making demands on a government, the more open a system is likely to be; (2) when the
legislative and executive functions of a government are separate, a government is more

198
open; (3) open systems allow interest groups access to the executive/administrative
branch of government; and (4) open systems allows for policy coalitions that can
aggregate challengers’ demands.
Opportunities include the institutional structure of the government. Some
structural components, like the degree of federalism in a system and the degree of
separation of powers in government, can have different impacts on the access of
challengers to government, as well as the government’s capacity to act on challengers’
demands (Amenta & Young, 1999; Kitschelt, 1986; Kriesi, 2007; Van Der Heijden,
1999). These structural aspects are invariant in the United States, and thus, in my study.
Other structural dimensions of political opportunity may vary. The first includes the
configuration of political actors. Eisinger (1973) found that challenges occurred less
frequently in council-manager governments than in mayor-council governments.
Generally, the council-manager form of government “insulates” government from extra-
governmental or private demands (McCabe & Feiock, 2005). As well, if a mayor had
veto power over a city council, that could translate into less access for challengers.
However, if municipal law imposed term limits on the mayoral post, it could increase
access to challengers by preventing mayors and their political biases to become
entrenched. In some cases, though, if a mayor were favorable to certain challengers, then
term limits might decrease those challengers’ access once that mayor had served his/her
term(s).
The electoral system also presents an important political opportunity for
challengers (Tarrow, 1998). Regarding legislative elections, proportional electoral
systems afford more access to challengers than majoritarian or pluralist systems (Kriesi,

199
2007; Amenta & Caren, 2007; Rootes, 2007). As well, proportional systems tend to have
more parties vying for votes, thus increasing access to challengers and policy coalitions
(Kriesi, 2007). Similarly, one might argue that whether a ballot is partisan or non-partisan
could influence the access challengers have, with a non-partisan ballot being more open
(Morgan, England, & Pelissero, 2007). The number of parties in an electoral system is
also important to consider. If the system is dominated by one or two parties, the access of
challengers may be limited as elite coalitions become entrenched (Rootes, 2007;
McAdam, 1999). Finally, the several fiscal factors can have impacts on the political
opportunity structure. The resources available to the administrative branch of government
can affect its openness; that is, the better funded the administrative branch through tax
revenues and intergovernmental aid, the less open it is to challengers (Kriesi, 2007).
Criticisms specific to the political process approach include a common criticism
of many social theories. That is, in an effort to make the theory comprehensive, the
concepts are too loosely defined, and the terms are widely interpreted and used in myriad
ways (Kriesi, 2007). Indeed, some worry that the approach is “in danger of becoming a
sponge that soaks up virtually every aspect of the social movement environment…[and]
threatens to become an all-encompassing fudge factor for all the conditions and
circumstances that form the context for collective action” (Gamson & Meyer, 1996, p.
275). Others argue that the approach omits the very important concept of exogenous
processes like immigration and deindustrialization (McAdam, 1999) or the “determining
environment” (Minkoff, 2002, p. 262) that also gives way to change. On a more micro-
level, McAdam argues that instead of being structuralist, researchers should concentrate
on “dynamic mechanisms (and concatenated processes) that shape ‘contentious politics’”

200
(2003, p. 284, emphasis original). Researchers might also focus on multiple institutional
levels, not simply national institutions like many researchers have done since Eisinger’s
1973 look at municipalities (McAdam, 1999). Finally, although the political process
approach was founded on the idea of political opportunities and threats, researchers have
focused primarily on opportunities (McAdam, 1999). This focus is “excessively narrow”
(McAdam, 1999, p. xi) because “opportunities alone do not make a movement,”
(McAdam & Snow, 1997, p. 80). Finally, researchers in this tradition need to be
cognizant of the “culturally constructed nature” of political threats and opportunities,
(McAdam, 1999, p. xi). Perhaps in acknowledgement of the sociocultural wing of social
movement research, McAdam relinquishes his earlier structurally deterministic approach:
“the ongoing interpretation of events by various collectivities shapes the likelihood of
movement emergence, as it shapes all of social life. Indeed, these continuous processes of
sense making and collective attribution are arguably more important in movements,”
(McAdam, 1999, p. xxi).

Research Design for Phase II

As stated above, the research question driving this phase of the study is: What are
the city-level political processes and institutional variables that are more likely to be
associated with better disamenities and amenities distributions? To answer this question,
I employ a comparative case-study approach that is largely descriptive and analyzes eight
propositions based on political institutional factors derived from the Social Movements

201
literature reviewed above. The propositions are would-be hypotheses; due to the small
number of cities in my study, I cannot test those hypotheses in a traditional manner. Still,
I have an empirical strategy for making sense of the propositions to see which ones might
explain environment justice in cities and thus, merit further testing. My sample of cities
remains the same as in Phase I, and the unit of analysis in Phase II is the city, ranked in
terms of high, medium, or low levels of equity.

Specification of Variables and Propositions

The following explanatory variables and propositions as derived from the
literature discussed above. The first set of hypotheses relate to the idea of how open
versus closed and repressive a polity might be (c.f., McAdam, 1996; Eisinger, 1973;
Tarrow, 1998; Kitschelt, 1986). The form of government in each city is directly related to
its potential openness or repression. Several types of municipal government exist in the
United States, including mayor-council, council-manager, commission, and town hall
forms. The two predominant forms are the mayor-council and the council-manager
forms, which are prevalent in 92% of American cities (Frederickson, Johnson, & Wood,
2004), though some cities add a ceremonial mayor with limited powers to the council-
manager form. Relative to challenger access and equity, the mayor-council form of
government may be more receptive than the council manager (Eisinger, 1973; Morgan,
England, & Pelissero, 2007; Feiock, Jeong, & Kim, 2003; Clingermayer & Feiock, 2001).
Generally speaking, the council-manager form of government grew out of the Progressive
Era, which sought to limit political influence and create a rule of administrative expertise.

202
As such, council-manager forms of government are less open to challenges, seeking to
serve the public interest, rather than particular constituencies or narrow interests. Mayor-
council forms of government may be more open to those constituencies or interests in
their efforts to seek re-election, and strong mayors may lead the city according to their
own vision. Thus, mayor-council cities may show a more equitable distribution of
environmental hazards and amenities. Mayoral term-limits, however, may impede a
mayor’s ability to push a specific policy agenda, particularly in the case where issues
have long time horizons (Pierson, 2000a). As well, mayoral veto power may be a critical
component of a mayor’s ability to promote or squelch environmental equity. The effect of
this may be mixed, however, as in the case of a mayor strongly opposed to environmental
equity. However, if a mayor were pro-environmental equity, then term limits may
actually constrain access once the mayor’s term nears its limit. Finally, neighborhood
councils may play an important part in a city’s openness to and access for challengers.
Representation of diverse interests might be better received through formalized
neighborhood councils that work with the city leadership. If such councils were
incorporated in city governance structures, then challengers would have increased access
that could result in better equity outcomes. Therefore, the propositions related to form of
government are:

P
1
: Cities with the mayor-council form of government will have a better equity
distribution than cities with a council-manager form of government.
P
2
: Cities with mayoral term limits will have better equity distributions than cities
without mayoral term limits.

203
P
3
: Cities without mayoral veto power will have better equity distributions than
cities with veto power.
P
4
: Cities with formal neighborhood councils will have better equity distributions
than cities without formal neighborhood councils.

The form of elections is the next institutional factor I examined. As discussed
above, citywide or at-large elections (versus district- or ward-based elections) would tend
to disadvantage vulnerable populations like racial minorities and poor people, as well as
heavily concentrated groups (Kriesi, 2007; Amenta & Caren, 2007; Rootes, 2007). Also,
district-based councils may ensure broad geographic representation, while at-large
elections may force council members to think in terms of “aggregate welfare,” (Lubell,
Feiock, & Ramirez, 2005, p. 714). Some critics argue that in ward-based elections,
stronger wards may dominate the policy agenda and may be captured by ward elites
(Morgan, England, & Pelissero, 2007). In some cities, city council members are elected
through both ward-based balloting and at-large balloting, though this combination may
disadvantage minority communities (Morgan, England, & Pelissero, 2007). Another
consideration is whether ballots are partisan or nonpartisan. The latter tends to favor the
higher socioeconomic segments of the population and disadvantage vulnerable
populations (Morgan, England, & Pelissero, 2007). Two party-related factors are
important to consider. First, cities with multiple parties vying in elections will be more
open to challengers (Rootes, 2007; McAdam, 1999) and may have better equity
distributions. Second, whereas a city votes Democrat versus Republican, the former may

204
give policy challengers more access and may have better equity distributions. Thus, the
electoral propositions are:

P
5
: Cities with district- or ward-based elections will have better equity
distributions.
P
6
: Cities using partisan ballots will have better equity distributions than cities
with nonpartisan ballots.
P
7
: Cities with multiple parties vying for election will have better equity
distributions than cities with two dominant parties.
P
8
: Majority Democrat cities will have better equity distributions than majority
Republican cities.

Data Collection and Operationalization

Data for Phase II were collected as followed. For form of government, term
limits, mayor veto power, and neighborhood council information, data were collected
from the individual city websites, particularly from each city’s charter, along with follow-
up calls to the appropriate city department when website information was lacking. Form
of government (P
1
) was operationalized as mayor-council, council-manager or mixed
(i.e., a council-manager form with a ceremonial mayor). Term limit data (P
2
) were
operationalized as existent or non-existent, while mayoral veto power (P
3
) was
operationalized as veto or no veto. The presence of formal neighborhood councils (P
4
)
was operationalized as yes or no.

205
Data on city elections, ballots, and number of parties on the ballot were gathered
from city, county, and federal websites, along with follow-up calls to the appropriate city
elections department (usually the city clerk). Type of election (P
5
) was operationalized as
district-based, ward-based, or mixed. Type of ballot (P
6
) was operationalized as partisan
or non-partisan. Number of parties (P
7
) was operationalized as two-party versus multi-
party. Majority party (P
8
) was operationalized as Democrat or Republican for both the
2004 and 2008 presidential elections (determined by the party that the majority of city
residents voted for in the last two presidential elections). For this, data were obtained
from each city’s respective Secretary of State.
In addition, I also examined each city’s organizational chart to determine the
institutional context of the city. First, I examined if each city had a dedicated
environmental department or the environmental services were subsumed under another
department. If a city has a stand-alone environmental department, it could signify that the
city makes environmental issues a priority by dedicating staff and resources to work
primarily on environmental issues. I also examined if environmental justice merited its
own department, was assumed under another department, or was evident in any of the
city’s major programs. Next, I determined if the city had a separate department for parks
and/or community gardens. Stand-alone parks departments might signify the importance
of parks to the city because of the time and resources dedicated to a stand-alone
department. Similarly, if a city had a program devoted to community gardens, it might
signify the importance of such amenities to the city. I also examined each city’s website
to determine if the city had environmental laws, policies, or executive orders pertaining
to environmental equity or if any state laws on environmental equity were in place. I

206
determined if each city had master plans for parks and/or community gardens. Using the
EPA website, I also determined if any of the cities were awarded environmental justice
grants from EPA. Finally, for each of the issues of environmental hazards, parks, and
community gardens, I tried to identify any major nonprofit organizations that work with
the city government to limit hazards and promote parks and community gardens. The
existence of strong nonprofits or a network of nonprofits may indicate the presence of
social capital, which may supplement a city’s policymaking efforts.
Once I obtained the information needed for the eight propositions, as well as the
additional descriptive data, I grouped the 18 cities according to high, medium, or low
equity, and then created a matrix of the descriptive data to make comparisons and draw
conclusions about my propositions. I am confident that the eight political institutional
variables I have chosen, as well as the additional descriptive data, capture the political
opportunity structure ad context in each city as predicted by the literature.
In the next chapter, I describe the results of the political institutional analysis and
case comparisons.

207
CHAPTER SEVEN: EQUITY AND THE CITY

The present and future problems of our cities are as complex as they are
manifold...We will neglect our cities at our peril, for in neglecting them
we neglect the Nation.
John F. Kennedy, Special Message to the Congress

This chapter presents the results of the institutional analysis of the political
process and opportunity structures in each city and how they may relate to environmental
equity. The research question driving this analysis is: What are the city-level political
processes and institutional variables that are more likely to be associated with better
disamenities and amenities distributions? Before analyzing that question, I briefly review
the status of environmental justice legislation. As I said in Chapter 1, the federal
government is guided by EO 12898, which encourages federal agencies to make
environmental justice a core goal. However, that order, over 15 years old, has not evolved
into any specific mandates, nor has Congress passed laws relating to environmental
justice—more than 20 environmental justice bills have failed to make it out of committee.
Even the EPA’s own Local Government Advisory Committee failed to mention
environmental justice in its guiding documents and has no environmental justice
subcommittee (EPA, 2006). This lack of federal leadership has left states and cities to
enact legislation and/or policies to reduce environmental inequities. Many states,
however, have no environmental justice programs or regulations (e.g., Michigan, Ohio,
Arizona, Virginia, Utah, West Virginia). Some have programs or advisory groups, but no
formal regulations (e.g., Pennsylvania, Illinois, New York, Texas, Arkansas, New

208
Mexico, Washington, Colorado). Massachusetts has several policies promoting
environmental justice, but California is the only state in the study with formal
environmental justice laws (American Bar Association, 2004). Recall that any
environmental justice efforts take place under the auspices of the Supreme Court’s 2001
ruling in Alexander v. Sandoval, 99 U.S. 1908, which limits an individual’s right of
action for race discrimination unless the complainant can prove intentional racism. This
lack of guidance or mandates from higher levels of government hampers environmental
justice efforts in cities, particularly under the specter of having to prove intentional
racism in the courts. Given that, what might make a difference in terms of a city’s
environmental equity? Can the structure and processes of city government make-up for
the lack of federal and state regulation?
In this chapter, I rank the cities based on their equitable distribution of
disamenities and then, amenities. I use those equity rankings to explore the institutional
propositions I proposed in Chapter 6.

Equity and the City

I hoped to rank all 18 cities on a continuum to denote their overall level of equity,
but their diverse sizes, populations, incomes, economies, and amounts of hazards and
amenities made it very difficult to do. For example, trying to compare a small city like
Albany on the same scale as a large city like San Diego is fraught with complications.
Therefore, the best way to organize the ranking was according to city size (e.g., large,

209
medium, or small). I also ranked the disamenities and amenities separately because the
former have attendant risks and externalities, while the latter are public goods.

Disamenities and the City

I ranked the cities on individual indicators (e.g., mean minority percentage of hot
spots, hazards per person, amount of hazard) for both TRI facilities and LQGs. The cities
are ranked from least equitable to most equitable, (i.e., 1 = least, 6 = most). In this way,
those cities with the lowest or worst equity rating—where vulnerable populations are
most over-represented near hazards—would be the least-desired outcome. I illustrate the
ranking system by discussing the six largest cities here (see Tables 7.1-7.2), and present
the overall hazard ranking for the big cities in Table 7.3. I used the same ranking system
for the medium and small cities, but the ranking discussion and tables are presented in the
Appendix; only the overall rankings are presented here (see Tables 7.4-7.5).
Of the six largest cities, the ones with the greatest amount of TRI facilities are, in
order: Philadelphia, San Jose, San Diego, Austin, Seattle, and Boston. Table 7.1 ranks the
cities from least to most equitable (i.e., 1 = least, 6 = best) according to their summary
statistics. Boston emerges as one of the most equitable cities in terms of these numbers,
though admittedly it has the fewest number of TRI sites. Boston’s TRI facilities are
randomly distributed and do not register any hot spots—sites in the other five cities
cluster globally and locally. Boston also has the fewest releases and its largest-producing
facility is located in a tract that is only 10% minority, with a median income of $57,292.
Boston’s mean center is located in a tract with similar demographics. Philadelphia, with

210
the most TRI sites, ranks worst in four out of six categories for which there are data. San
Jose emerges as the second-worst city, followed closely by Seattle. Austin and Boston
have minority percentages below 40% and incomes above $30,000 for both their biggest
producing tract and mean center tract. However Austin ranks worst in terms of mean
minority percentage of hot spots. San Jose and Seattle also had high minority
representation among their hot spots, while no city had incomes lower than $36,000 in
their hot spots. None of the six largest cities exhibited TRI cold spots.

Table 7.1
The Six Largest Cities and TRI Sites
Total
Pounds
Persons
per Site
Biggest
Producer
Tract
Minority %
Biggest
Producer
Tract
Income
Mean
Center
Minority
%
Mean
Center
Income
Hot
Spots
Minority
%
Hot
Spots
Income
Philadelphia 1 5 No data No data 1 1 6 2
San Diego 3 4 1 1 4 3 5 5
San Jose 2 3 2 4 2 5 3 6
Austin 4 2 4 3 5 2 2 3
Boston 6 6 5 5 6 6 1 1
Seattle 5 1 3 2 3 4 4 3

Of the six largest cities, the cities with the greatest amount LQGs are, in order:
San Diego, Seattle, San Jose, Philadelphia, Boston, and Austin. Table 7.2 ranks those
cities from least to most equitable (i.e., 1 = least, 6 = best) according to their summary
statistics. San Jose and Seattle, which appear to be the least equitable cities, are the two
largest producers with only a 364-ton difference. Seattle is the next highest producer, yet
it has the most startling persons per LQG ratios of 1 LGQ for every 7,316 people. The
best persons per LQG ratio belongs to Philadelphia, but with Philadelphia’s population
over 1.5 million, this is no surprise. Although no data were available for the minority

211
percentage of Philadelphia’s biggest-producing tract, only San Jose has a minority
percentage greater than 50% in that category. In terms of the mean center tract,
Philadelphia, San Jose, and Seattle have minority percentages of 83%, 76%, and 53%,
respectively. Income figures are generally higher for LQGs than TRI facilities; although
Philadelphia’s income is $33,065, the other four cities for which there are data have
incomes over $46,000. The hot spot mean incomes range from $30,023 in Boston to
$77,628, with five cities having incomes over $39,000.

Table 7.2
The Six Largest Cities and LQGs
Total
Tons
Persons
per LQG
Biggest
Producer Tract
Minority %
Biggest
Producer
Tract
Income
Mean
Center
Minority
%
Mean
Center
Income
Hot
Spots
Minority
%
Hot
Spots
Income
Philadelphia 2 6 No data 1 1 1 6 3
San Diego 4 2 5 No data 5 2 5 5
San Jose 1 3 1 3 2 6 1 6
Austin 5 5 2 5 6 3 4 4
Boston 6 4 4 4 4 5 3 1
Seattle 3 1 3 2 3 4 2 2

Table 7.3 displays the average rankings (combined from Tables 7.1 and 7.2) for
each type of hazard, and then ranks the cities overall. Recall that a ranking of 1 for
hazards indicates the least equitable city, while a ranking of 6 indicates the most
equitable city. Overall, Philadelphia had the worst hazard ranking, followed by Seattle,
San Jose, Austin, San Diego, and Boston. In other words, Boston is a more
environmentally equitable city in terms of the hazards as I defined them.

212
Table 7.3
Average Hazard Ranks for the Six Largest Cities
TRI
Rating
Ordinal
Rank
LQG
Rating
Ordinal
Rank
Overall
Hazard
Rank
Philadelphia 2.29 1 2.86 2 1
San Diego 3.25 4 4 5 5
San Jose 3.38 5 2.88 3 3
Austin 3.13 2 / 3 4.25 6 4
Boston 4.5 6 3.88 4 6
Seattle 3.13 2 / 3 2.5 1 2

Table 7.4 displays the average rankings for each type of hazard in the medium-
sized cities, and then ranks the cities overall with 1 being the least equitable and 6 being
the most equitable (see the Appendix for individual TRI and LQG rankings). Overall,
Little Rock is the least equitable city, ranking as the second worst in both TRI facilities
and LQGs. This was followed by Grand Rapids, Norfolk, Miami and Salt Lake City, and
Dayton. Norfolk presents the most stark contrast between its TRI rank (i.e., least
equitable) and its LQG ranking (i.e., most equitable); the city’s biggest-producing tract
has the highest percentage of minorities and the lowest income. As well, Norfolk hot
spots are located in high minority, low-income tracts.

Table 7.4
Average Hazard Ranks for the Six Medium Cities
TRI
Rating
Ordinal
Rank
LQG
Rating
Ordinal
Rank
Overall
Hazard
Rank
Miami 4.33 5 3.13 3 4 / 5
Norfolk 2.13 1 4.63 6 3
Grand
Rapids
3.75 4 2.75 1 2
Little Rock 2.75 2 2.88 2 1
Salt Lake
City
2.88 3 4.0 5 4 / 5
Dayton 4.5 6 3.5 4 6

213
Table 7.5 displays the average hazard rankings for the five smallest cities (recall
that Santa Fe has no hazards). With a ranking of one in both TRI sites and LQGs, Decatur
is least equitable city overall (see the Appendix for the explanation of rankings). The
remaining cities, ranked from least equitable to most equitable, are Albany and Boulder
in a tie for second and third, and Charleston and Flagstaff tied for fourth and fifth.
Charleston and Flagstaff are interesting because in the case of the former, the economy
revolves around government services; most people imagine that the coal mines and coal
processing industries would dampen Charleston’s rank, but those sites are not within the
city proper. Flagstaff has only one TRI facility and one LQG.

Table 7.5
Average Hazard Ranks for the Five Smallest Cities
TRI
Rating
Ordinal
Rank
LQG
Rating
Ordinal
Rank
Overall
Hazard
Rank
Albany 2.63 3 2.38 2 2 / 3
Boulder 2.0 2 2.5 3 2 / 3
Decatur 1.88 1 2.25 1 1
Charleston 3.38 5 3.0 4 4/5
Flagstaff 3.5 4 4.75 5 4/5

Amenities and the City

I ranked the amenities in the same way as the hazards, in terms of least equitable
to most (i.e., 1 = least, 6 = most). The least equitable city would be one in which minority
and/or poor populations were under-represented near amenities. A more equitable city
would earn a higher ranking (e.g., a 5 or 6) and would be more desirable, just as a city
that ranked more equitably on hazards (i.e., earning a 5 or a 6) would be more desirable.

214
As with the hazards discussion, I present the expanded tables and discussion for the six
largest cities, but only present the overall ranks for the medium and small cities. (See the
Appendix for additional discussion of the medium and small cities.)
Of the six largest cities, Seattle has the most parks, followed by Boston, San Jose,
Austin, Philadelphia, and San Diego (see Table 7.6). Boston has the best equity overall,
with no clustering of parks, and ranking next best on nearly everything else. Seattle may
have the most parks and the best persons per park ratio, but minorities are not well-
represented, particularly in terms of mean center or hot spots. However, Seattle was the
only large city that registered park cold spots. The mean minority percentage for Seattle’s
10 cold-spot tracts was 16%, while mean income was $53,776, indicating that where
parks are dispersed or lacking, they are so in predominantly white and upper-middle
income tracts. San Diego scores poorly on several indicators, but its hot spots are in tracts
with a mean minority percentage of 77% and a mean income of $25,751.

Table 7.6
The Six Largest Cities and Parks
Parks Persons
Per Park
Mean
Center
Minority
%
Mean
Center
Income
Hot
Spots
Minority
%
Hot
Spots
Income
Cold Spots
Minority
%
Cold
Spots
Income
Philadelphia 2 1 6 6 1 3 6 6
San Diego 1 2 2 1 5 5 6 6
San Jose 4 3 4 2 4 1 6 6
Austin 3 4 1 4 3 4 6 6
Boston 5 5 5 3 6 6 6 6
Seattle 6 6 3 5 2 2 5 5

Table 7.7 indicates that community gardens are most plentiful in Boston,
Philadelphia, Seattle, San Jose, San Diego, and Austin, respectively. Boston again shines
as the most equitable city when it comes to community gardens. It has the most, with the
best ratio of persons per garden, and even though it does not rank well in terms of hot

215
spots, it boasts a mean center tract with a minority percentage of 96% and median income
of $27,165. Seattle and Philadelphia are the next most equitable cities, while San Diego
and San Jose do not rank very well. Recall that they do not have many gardens. Austin
also has few gardens, but does represent minorities and low-income people well. These
cities registered no garden cold spots.

Table 7.7
The Six Largest Cities and Community Gardens
Gardens Persons Per
Garden
Mean
Center
Minority
%
Mean Center
Income
Hot Spots
Minority %
Hot Spots
Income
Philadelphia 5 4 2 2 6 6
San Diego 2 1 3 4 4 4
San Jose 3 3 4 1 2 1
Austin 1 2 5 5 6 5
Boston 6 6 6 6 3 3
Seattle 4 5 1 3 5 2

Table 7.8 displays the average rankings (from Tables 7.6 and 7.7) for each type of
amenity, and then ranks the cities overall. Recall that a ranking of 1 for amenities
indicates the least equitable city, while 6 indicates the most equitable city, so a 6 is
desirable. Overall, Boston has the best ranking, followed by Seattle, Philadelphia and
Austin, with San Diego and San Jose being tied for least equitable.

Table 7.8
Average Amenity Rankings for the Six Largest Cities
Parks
Rating
Ordinal
Rank
Community
Gardens
Rating
Ordinal
Rank
Overall
Amenity
Ranking
Philadelphia 3.13 3 / 4 2.83 4 4 / 3
San Diego 3.5 1 4.0 2 1 / 2
San Jose 3.25 2 4.67 1 1 / 2
Austin 3.13 3 / 4 3.0 5 4 / 3
Boston 1.75 6 2.0 6 6
Seattle 2.75 5 3.67 3 5

216

Table 7.9 displays the overall park and community garden ratings for the
medium-sized cities, and then ranks the cities overall with 1 being the least equitable and
6 being the most equitable (see the Appendix for expanded park and community garden
ratings). Overall, Miami ranked as most equitable, followed by Little Rock, Dayton,
Norfolk and Grand Rapids, and Salt Lake City.

Table 7.9
Average Amenity Rankings for the Six Medium Cities
Parks
Rating
Ordinal
Rank
Community
Gardens
Rating
Ordinal
Rank
Overall
Amenity
Ranking
Miami 4.0 6 / 5 4.5 5 6
Norfolk 3.67 4 1.0 1 3 / 2
Grand
Rapids
3.25 3 2.42 2 3 / 2
Little Rock 4.0 6 / 5 4.0 4 5
Salt Lake
City
3.0 1 3.75 3 1
Dayton 3.08 2 5.33 6 4

Table 7.10 displays the overall amenity rankings for the smallest cities Overall,
Santa Fe ranked as the most equitable, followed by Albany and Decatur, Boulder,
Flagstaff, and Charleston.

Table 7.10
Average Amenity Ranks for the Six Smallest Cities
Parks
Rating
Ordinal
Rank
Community
Gardens
Rating
Ordinal
Rank
Overall
Amenity
Ranking
Albany 3.67 3 5.0 6 5 / 4
Boulder 3.83 4 3.25 4 3
Decatur 4.17 6 3.0 3 5 / 4
Santa Fe 4.0 5 4.0 5 6
Charleston 2.5 1 2.5 1 2
Flagstaff 2.83 2 2.6 2 1

217
My attempt to rank cities is rudimentary and does not accomplish my original
intent of ranking all 18 cities along a single index. The diversity of the cities in terms of
size and economy made ranking all cities together difficult. As well, trying to rank
disamenities exactly as amenities was also difficult. Further, no toxicity or risk measures
are included in the hazard rankings. Still, this basic ranking system is a starting point for
exploring the institutional propositions, which I do next.

Institutional Variables and Environmental Equity

In Chapter 6, I outlined eight propositions related to the political structure and
processes of city government that may impact a city’s distribution of environmental
disamenities or amenities. Due to the small number of cities in my study, I was unable to
conduct traditional hypothesis testing. Thus, I addressed each proposition—four related
to the opportunity structure or form of government and four related to the political
processes at work—in a descriptive manner. I begin with disamenities.

The Opportunity Structure and Disamenities

The opportunity structure of a city government may affect a city’s ability to
prevent, mitigate, or abolish environmental inequity. I discuss the four structure-related
propositions according to city size (e.g., large, medium, small). The propositions are:

218
P
1
: Cities with the mayor-council form of government will have a better equity
distribution than cities with a council-manager form of government.
P
2
: Cities with mayoral term limits will have better equity distributions than cities
without mayoral term limits.
P
3
: Cities without mayoral veto power will have better equity distributions than
cities with veto power.
P
4
: Cities with formal neighborhood councils will have better equity distributions
than cities without formal neighborhood councils.

Table 7.11 lists the six largest cities according to their disamenity ranking derived
above, along with their political opportunity structure variables. In all but Austin, the
mayor-council form of government is prevalent in big cities, although San Diego was a
council-manager city until 2006 when it switched to mayor-council for a five-year trial.
Austin has a mayor, but it is a ceremonial position and the mayor is limited to one term
and has no veto power. Mayoral term limits operate in four of the six cities; although
Philadelphia and San Jose’s mayors are limited to two four-year terms, they can both run
again after sitting out one four-year term. Mayoral veto power is present in four of the six
cities. Given these variables, the strongest mayoral post is in Boston, and the weakest, not
counting Austin’s ceremonial mayor, is in San Jose.

219
Table 7.11
Disamenity Ranks and Political Opportunity Variables in the Six Largest Cities
a

Disamenity
Ranking
Form of
Government
Additional
Executive
Mayoral
Term
Limits
Mayoral
Veto
Formal
Neighborhood
Organizations
Boston 6 Mayor-
Council
No Yes Yes
San Diego
b
5 Mayor-
Council
Yes Yes Yes
Austin 4 / 3 Council-
Manager
Ceremonial
Mayor
Yes No Yes
San Jose 4 / 3 Mayor-
Council
Yes No Yes
Seattle 2 Mayor-
Council
No Yes Yes
Philadelphia 1 Mayor-
Council
Yes Yes Yes
a
Cities are arrayed in terms of most equitable to least equitable.
b
San Diego began a five-year trial of the mayor-council form of government in 2006.

For disamenities in the six largest cities, Proposition One is not supported.
Though Boston is a strong mayor city, San Diego (which was a council-manager city at
the time the data in this study were collected) ranked second in terms of equity. San Jose,
Seattle, and Philadelphia all have mayor council governments, but had poor equity
ratings. Proposition Two, that cities with mayoral term limits will have better equity
outcomes is partially supported. Although Boston has no term limit, San Diego and
Austin do. Of the cities with poorer equity ratings, Seattle has no term limits, but
Philadelphia does. Proposition Three argues that cities without a mayoral veto will have
better equity distributions and this is definitely not supported. Finally, Proposition Four
suggests that cities with formal neighborhood organizations will have better equity
distributions. This proposition is partially supported, but should be qualified. Boston has
a thriving and forceful collection of neighborhood organizations, which supports the
proposition, but so do Philadelphia and Seattle have strong neighborhood organizations

220
and those cities do not rank well in terms of equity. San Diego and Austin’s
neighborhood programs are not that strong, though they ranked high on equity. This
suggests further investigation into how the neighborhood organizations are formalized
and in which city programs they are most active.
Table 7.12 lists the disamenity rankings and the institutional variables for the six
medium cities. Proposition One is partially supported. Although Dayton is a council-
manager city, the cities with the three lowest equity ratings are (or were at the time of
data collection) council-manager governments. Proposition Two is not supported, and
results for the mayoral veto proposition are mixed; Miami and Salt Lake City’s mayoral
veto supports the proposition. Proposition Three is also partially supported. All of these
cities have neighborhood organizations, though none are strong like Boston’s or
Philadelphia’s. Thus, support for Proposition Four is mixed.

Table 7.12
Disamenity Ranks and Political Opportunity Variables in the Six Medium Cities
a

Disamenity
Ranking
Form of
Government
Additional
Executive
Mayoral
Term
Limits
Mayoral
Veto
Formal
Neighborhood
Organizations
Dayton 6 Council-
Manager
Ceremonial
Mayor
No No Yes
Miami 5 / 4 Mayor-
Council
Yes Yes Yes
Salt Lake
City
5 / 4 Mayor-
Council
No Yes Yes
Norfolk 3 Council-
Manager
Ceremonial
Mayor
No No Yes
Grand
Rapids
2 Commission-
Manager
Yes
Little Rock
b
1 Mayor-
Council
No Yes Yes
a
Cities are arrayed in terms of most equitable to least equitable.
b
Little Rock switched from a council-manager form of government in 2007.

221
The rankings and institutional variables for the smallest cities are arrayed in Table
7.13. Recall that Santa Fe has no disamenities as I defined them. Proposition One is
partially supported; two of the three most equitable cities are mayor-council cities, while
the two least equitable are council-manager cities. As for mayoral term limits,
Proposition Two is not supported, and Proposition Three is partially supported; the most
equitable cities do not have mayoral veto power. Neighborhood organizations are not as
prevalent in smaller cities, which might be expected given that a small city of only
50,000-60,000 people might have an easier time integrating citizens into city programs
than a large and densely populated city. In this case, Proposition Four is mixed.

Table 7.13
Disamenity Ranks and Political Opportunity Variables in the Five Smallest Cities
a

Disamenity
Ranking
Form of
Government
Additional
Executive
Mayoral
Term
Limits
Mayoral
Veto
Formal
Neighborhood
Organizations
Charleston 5 / 4 Mayor-
Council
City
Manager
No No No
Flagstaff 5 / 4 Council-
Manager
Ceremonial
Mayor
No No Yes
Albany 3 / 2 Mayor-
Council
No Yes No
Boulder 3 / 2 Council-
Manager
No
Decatur 1 Council-
Manager
Mayor
with some
power
No No Yes
a
Cities are arrayed in terms of most equitable to least equitable.

Overall, the support for the opportunity structure propositions was mixed at best
for disamenities. Casting aside the propositions and city size, I compared the more
equitable cities in the study (e.g., Boston, Dayton, Charleston) with those less equitable
(e.g., Philadelphia, Grand Rapids, Decatur) to determine if some combination of political

222
opportunity structure was evident. Surprisingly, five of the seven more equitable cities in
terms of hazards had in common the mayor-council form of government and
neighborhood organizations, though the strength of the neighborhood organizations
varied. This combination should be examined thoroughly—it could be that even strong
mayors need neighborhood alliances to promote an environmental equity agenda and
limit the exposure of poor and minority populations to hazards. No other consistent
pattern or combination was evident for either the most equitable or the least equitable
cities. This is disappointing, but I only examined 18 cities in the study. Additional cities
may yield more conclusive results.

Disamenities and the Political Processes of the City

In this section, I discuss the political process propositions according to city size
(e.g., large, medium, small) in terms of how they relate to disamenities. These
propositions include the following:

P
5
: Cities with district- or ward-based elections will have better equity
distributions.
P
6
: Cities using partisan ballots will have better equity distributions than cities
with nonpartisan ballots.
P
7
: Cities with multiple parties vying for election will have better equity
distributions than cities with two dominant parties.

223
P
8
: Majority Democrat cities will have better equity distributions than majority
Republican cities.

Table 7.14 presents the process variables and hazard ranks for the six largest
cities. Proposition Five is partially supported; the two most equitable cities have some
variation of district-based elections. However, Philadelphia, which ranks last in terms of
equity does as well, though it has a combination of 10 district-based and 7 at-large
members. Proposition Six not supported, while Proposition Seven is partially supported.
Proposition Eight is also partially supported, but weakly so. With the exception of San
Diego’s vote in the 2004, each of the six largest cities voted Democrat in the last two
presidential elections.

Table 7.14
Disamenity Ranks and Political Process Variables in the Six Largest Cities
a

Disamenity
Rank
Type of Election Ballot
Type
Multiple
Parties
Operating
Presidential
Vote in
2008
Presidential
Vote in
2004
Boston 6 9 district-based; 4
at-large; Council
President elected
at-large
Non-
partisan
Yes Democrat Democrat
San Diego 5 District-based Non-
partisan
Yes Democrat Republican
Austin 4 At-large Non-
partisan
Yes Democrat Democrat
San Jose 3 District-based Non-
partisan
Yes Democrat Democrat
Seattle 2 At-large Non-
partisan
Yes Democrat Democrat
Philadelphia 1 10 district-based;
7 at-large
Partisan Yes Democrat Democrat
a
Cities are arrayed in terms of most equitable to least equitable.

224
Regarding the medium-sized cities, Proposition Five not supported (see Table
7.15). Although Dayton elects its council members at large, Grand Rapids and Little
Rock rank last, yet both elect their council members based on districts or wards.
Proposition Six is mixed—each of the six medium-sized cities use a non-partisan ballot.
Proposition Seven is also mixed. Support for Proposition Eight is mixed. In the 2008
Presidential election, each of the cities voted Democrat. In 2004, Salt Lake City and
Norfolk, which ranked in the middle in terms of equity, voted Republican. As with the six
largest cities, there may not be enough variation in some of the process variables to gauge
their effect adequately. Variation exists in how councils are elected (e.g., ward-based
versus at-large), but ballot type, parties, and even voting lack variance.

Table 7.15
Disamenity Rank and Political Process Variables in the Six Medium Cities
a

Disamenity
Rank
Type of Election Ballot
Type
Multiple
Parties
Operating
Presidential
Vote in
2008
Presidential
Vote in
2004
Dayton 6 At-large Non-
partisan
Yes Democrat Democrat
Miami 5 / 4 Ward-based Non-
partisan
Yes Democrat Democrat
Salt Lake
City
5 / 4 Ward-based Non-
partisan
Yes Barely
Democrat
Republican
Norfolk 3 5 district-based;
2 at-large
Non-
partisan
Yes Democrat Republican
Grand
Rapids
2 Ward-based Non-
partisan
Yes Democrat Democrat
Little Rock 1 7 ward-based; 3
at-large
Non-
partisan
Yes Democrat Democrat
a
Cities are arrayed in terms of most equitable to least equitable.

Table 7.16 displays the political process variables for the smallest cities in the
study. Proposition Five is partially supported; two of the three most equitable cities have

225
primarily ward-based elections, while the two least equitable cities have at-large
elections. Proposition Six is mixed, given that the city with the best equity outcome (i.e.,
Charleston) uses a partisan ballot, while two of the least equitable cities use a non-
partisan ballot. Support for Proposition Seven is also mixed. Proposition Eight is not
supported; more variation in vote majorities exists in the smaller cities, and two of the
most equitable cities voted primarily Republican in the recent presidential elections.
Boulder and Decatur voted primarily Democrat, yet they rank last in equity. More
variation exists among the smaller cities. With the exception of how many parties operate
at the local level, the smaller cities diverge in the types of elections and ballots used and
in their voting patterns in Presidential elections.

Table 7.16
Disamenity Rank and Political Process Variables in the Five Smallest Cities
a

Disamenity
Rank
Type of Election Ballot
Type
Multiple
Parties
Operating
Presidential
Vote in
2008
Presidential
Vote in
2004
Charleston 5 / 4 21 Ward-based; 6
at-large
Partisan Yes Republican Republican
Flagstaff 5 / 4 At-large Non-
partisan
Yes Republican Democrat
Albany 3 / 2 15 Ward-based;
Council President
at-large
Partisan Yes Democrat Democrat
Boulder 3 / 2 At-large Non-
partisan
Yes Democrat Democrat
Decatur 1 At-large Non-
partisan
Yes Democrat Republican
a
Cities are arrayed in terms of most equitable to least equitable.

As with the political structure variables, the data show only mixed support for the
political process variables, though overall, the process-related propositions faired slightly

226
better. Examining the 17 cities collectively, no consistent pattern emerges regarding
disamenities and the political processes variables. The most equitable cities (e.g., San
Diego, Miami, Flagstaff) are a mixture of ward-based and at-large elections. As well, the
majority of most equitable cities and least equitable cities (e.g., Seattle, Little Rock) use
non-partisan ballots and voted Democrat in the last two presidential elections.

Political Opportunity Structure and Amenities

Below, I examine the four structure-related propositions vis-à-vis environmental
amenities according to city size. The four structure-related propositions are:

P
1
: Cities with the mayor-council form of government will have a better equity
distribution than cities with a council-manager form of government.
P
2
: Cities with mayoral term limits will have better equity distributions than cities
without mayoral term limits.
P
3
: Cities without mayoral veto power will have better equity distributions than
cities with veto power.
P
4
: Cities with formal neighborhood councils will have better equity distributions
than cities without formal neighborhood councils.

Table 7.17 arrays the largest cities and their structural variables according to the
amenity rankings. Note that Boston again ranks as the most equitable city. Proposition
One is partially supported; the three most equitable cities are mayor-council cities, while

227
two of the bottom three cities are (or were at the time of data collection) council-manager
cities. Proposition Two, that cities with mayoral term limits will have better equity
outcomes is not supported. Neither Boston nor Seattle has mayoral term limits, while the
least equitable cities do have term limits. Proposition Three argues that cities without a
mayoral veto will have better equity distributions and this is not supported—the three
most equitable cities have mayoral vetoes, while two of the three least equitable do not.
Finally, Proposition Four suggests that cities with formal neighborhood organizations
will have better equity distributions and is partially supported. Neighborhood alliances
are particularly important in promoting environmental amenities, as I discuss below.

Table 7.17
Amenity Ranks and Political Opportunity Variables in the Six Largest Cities
a

Amenity
Ranking
Form of
Government
Additional
Executive
Mayoral
Term
Limits
Mayoral
Veto
Formal
Neighborhood
Organizations
Boston 6 Mayor-
Council
No Yes Yes
Seattle 5 Mayor-
Council
No Yes Yes
Philadelphia 4 / 3 Mayor-
Council
Yes Yes Yes
Austin 4 / 3 Council-
Manager
Ceremonial
Mayor
Yes No Yes
San Diego
b
2 / 1 Mayor-
Council
Yes Yes Yes
San Jose 2 / 1 Mayor-
Council
Yes No Yes
a
Cities are arrayed in terms of most equitable to least equitable.
b
San Diego began a five-year trial of the mayor-council form of government in 2006.

The six medium cities are listed in Table 7.18. Proposition One is partially
supported, with the most equitable city having a strong mayor-council form of
government. Two of the three least equitable cities (e.g., Norfolk, Grand Rapids) have

228
council-manager governments. Little Rock ranks second in terms of equity, but at the
time of data collection, the city had a council-manager form of government. Proposition
Two is not supported—term limits are only evident in Miami. While two of the three
most equitable cities have a mayoral veto, so does Salt Lake City, which ranks last in
terms of equity. Thus, Proposition Three is partially supported; while Santa Fe and
Decatur have no veto, the least equitable city also has none. All of these cities have
neighborhood organizations, thus, support for Proposition 4 is mixed; each of the
medium-sized cities has formal neighborhood organizations so it is unclear if they have a
positive effect on amenity distribution.

Table 7.18
Amenity Rankings and Political Opportunity Variables in the Six Medium Cities
a

Amenity
Rankings
Form of
Government
Additional
Executive
Mayoral
Term
Limits
Mayoral
Veto
Formal
Neighborhood
Organizations
Miami 6 Mayor-
Council
Yes Yes Yes
Little Rock
b
5 Mayor-
Council
No Yes Yes
Dayton 4 Council-
Manager
Ceremonial
Mayor
No No Yes
Norfolk 3 / 2 Council-
Manager
Ceremonial
Mayor
No No Yes
Grand
Rapids
3 / 2 Commission-
Manager
Yes
Salt Lake
City
1 Mayor-
Council
No Yes Yes
a
Cities are arrayed in terms of most equitable to least equitable.
b
Little Rock switched from a council-manager form of government in 2007.

Table 7.19 arrays the amenity rankings and institutional variables for the smallest
cities. Proposition One is partially supported as both Santa Fe and Albany have strong
mayor-council forms of government and two of the three least equitable cities have

229
council-manager forms of government. Proposition Two is also mixed; Santa Fe is the
only city with mayoral term limits and also happens to be the most equitable. Proposition
Three is not supported; only Albany has a mayoral veto. Finally, support for Proposition
Four is very weak considering that the two most equitable cities do not have
neighborhood organizations, yet Flagstaff—the least equitable city—does. It is surprising
that only two of the six cities have neighborhood organizations. This could be a reflection
of city size; the majority of these cities have populations of 82,000 or less. Or, it could be
that smaller cities have smaller government structures that can serve their populations
without additional layers of governance.

Table 7.19
Amenity Ranks and Political Opportunity Variables in the Six Smallest Cities
a

Amenity
Ranks
Form of
Government
Additional
Executive
Mayoral
Term
Limits
Mayoral
Veto
Formal
Neighborhood
Organizations
Santa Fe 6 Mayor-
Council
City
Manager
Yes No No
Albany 5 / 4 Mayor-
Council
No Yes No
Decatur 5 / 4 Council-
Manager
Mayor
with some
power
No No Yes
Boulder 3 Council-
Manager
No
Charleston 2 Mayor-
Council
City
Manager
No No No
Flagstaff 1 Council-
Manager
Ceremonial
Mayor
No No Yes
a
Cities are arrayed in terms of most equitable to least equitable.
bSanta Fe’s mayor can break a tie of the City Council, however.

As with the disamenities, the support for the opportunity structure variables was
mixed at best for amenities. I compared the more equitable cities in the study (e.g.,

230
Boston, Miami, Santa Fe) with those less equitable (e.g., San Jose, Salt Lake City,
Flagstaff) to determine if some combination of political opportunity structure was
evident. Unfortunately, no consistent combination or pattern emerged.

Amenities and Political Processes

In this section, I discuss the amenity rankings and political processes variables
according to city size. Recall that these propositions include the following:

P
5
: Cities with district- or ward-based elections will have better equity
distributions.
P
6
: Cities using partisan ballots will have better equity distributions than cities
with nonpartisan ballots.
P
7
: Cities with multiple parties vying for election will have better equity
distributions than cities with two dominant parties.
P
8
: Majority Democrat cities will have better equity distributions than majority
Republican cities.

The amenity ranks and process variables for the six largest cities are presented in
Table 7.20. Proposition Five is not supported; though the most equitable city uses
district-based elections, so do the two least equitable cities, while the second-most
equitable city uses at-large elections. Proposition Six is not supported; only Philadelphia

231
has a partisan ballot. Proposition Seven is partially supported. Proposition Eight is
supported—the most equitable cities voted Democrat in recent presidential elections.

Table 7.20
Amenity Ranks and Political Process Variables in the Six Largest Cities
a

Disamenity
Rank
Type of Election Ballot
Type
Multiple
Parties
Operating
Presidential
Vote in
2008
Presidential
Vote in
2004
Boston 6 District-based;
Council President
Elected at-large
Non-
partisan
Yes Democrat Democrat
Seattle 5 At-large Non-
partisan
Yes Democrat Democrat
Philadelphia 4 / 3 10 district-based;
7 at-large
Partisan Yes Democrat Democrat
Austin 4 / 3 At-large Non-
partisan
Yes Democrat Democrat
San Diego 2 / 1 District-based Non-
partisan
Yes Democrat Republican
San Jose 2 / 1 District-based Non-
partisan
Yes Democrat Democrat
a
Cities are arrayed in terms of most equitable to least equitable.

Table 7.21 presents the amenity ranks and process variables for the medium-sized
cities. Proposition Five is not supported—while Miami and Little Rock, the most
equitable cities, use ward-based elections, the three least equitable cities use some form
of ward-based elections. Proposition Six is not supported as all of the medium-sized cities
use non-partisan ballots. The results for Proposition Seven are mixed. Support for
Proposition Eight is better than mixed given that the top three equitable cities voted
Democrat in both the recent presidential elections, while two of the three least equitable
voted Republican in 2004 and Salt Lake City barely voted Democrat in 2008.

232
Table 7.21
Amenity Rank and Political Process Variables in the Six Medium Cities
a

Amenity
Rank
Type of Election Ballot
Type
Multiple
Parties
Operating
Presidential
Vote in
2008
Presidential
Vote in
2004
Miami 6 Ward-based Non-
partisan
Yes Democrat Democrat
Little Rock 5 7 ward-based; 3
at-large
Non-
partisan
Yes Democrat Democrat
Dayton 4 At-large Non-
partisan
Yes Democrat Democrat
Norfolk 3 / 2 5 district-based;
2 at-large
Non-
partisan
Yes Democrat Republican
Grand Rapids 3 / 2 Ward-based Non-
partisan
Yes Democrat Democrat
Salt Lake City 1 Ward-based Non-
partisan
Yes Barely
Democrat
Republican
a
Cities are arrayed in terms of most equitable to least equitable.

The amenity ranks and political process variables for the smallest cities are
presented in Table 7.22. Proposition Five is partially supported, as is Proposition Six.
Support for Proposition Seven is also mixed. Proposition Eight is supported with the most
equitable cities voting Democrat and the least equitable voting primarily Republican.

Table 7.22
Amenity Rank and Political Process Variables in the Six Smallest Cities
a

Amenity
Rank
Type of Election Ballot
Type
Multiple
Parties
Operating
Presidential
Vote in
2008
Presidential
Vote in
2004
Santa Fe 6 Ward-based Non-
partisan
Yes Democrat Democrat
Albany 5 / 4 15 Ward-based;
Council President
at-large
Partisan Yes Democrat Democrat
Decatur 5 / 4 At-large Non-
partisan
Yes Democrat Republican
Boulder 3 At-large Non-
partisan
Yes Democrat Democrat
Charleston 2 21 Ward-based; 6
at-large
Partisan Yes Republican Republican
Flagstaff 1 At-large Non-
partisan
Yes Republican Democrat
a
Cities are arrayed in terms of most equitable to least equitable.

233
Similar to the disamenities, support for the process-related propositions was weak
or mixed at best. I examined the most equitable of the 18 cities (e.g., Seattle, Miami,
Santa Fe) and the least equitable cities (e.g., San Diego, Norfolk, Charleston) to see if any
combination of process variables emerged, though none did. As I noted earlier, the
process variables lack sufficient variation. Adding additional cities might yield different
or more conclusive results.

Summary

What do the results of the institutional analysis show? In the case of the structural
variables, only Proposition Four (e.g., neighborhood organization may correlate with
better equity distributions) was partially supported for all city sizes for both amenities
and disamenities. Proposition One (e.g., mayor-council governments may correlate with
better equity distributions) was partially supported in all city sizes for amenities.
Regarding the political process variables, Proposition Seven (e.g., cities with multiple
parties will have better equity distributions) showed weak support for all city sizes for
both disamenities and amenities. Proposition Eight (e.g., cities voting Democrat will have
better equity distributions) was fairly well-supported for all city sizes for amenities. Some
conclusions that can be drawn from the analysis are: (1) that the propositions operated
differently for disamenities and amenities, and (2) other forces might explain the equity
distributions for disamenities and amenities.
For disamenities, perhaps the distribution of hazards is: mired in historical
decisions about planning and economic growth, constrained by budget limits, and less of

234
a priority than other urban problems (e.g., crime). Thus, the historical development of the
city may have as much to do with modern equity distributions as do current political
interventions. Trying to focus solely on current interventions or discrete political
variables misses the larger context; or, as Pierson describes it, “the significance of such
variables is frequently distorted when they are ripped from their temporal context,”
(2002b, p. 72). Historically, cities have made thousands of decisions about their structure,
development, property rights, and policies. These decisions, North argues, create the
scaffolding of the city, which “not only constrains the choice set at a moment of time but
is the source of path dependence,” (2005b, p. 26). Even a city’s culture, norms, and
values are welded into the scaffolding. Thus, no matter what changes a particular city or
its institutional actors may try to effect, those changes are constrained by previous policy
makers and historical antecedents.
The distribution of environmental hazards is very much tied to the history of
economic growth in a city (c.f., Krieg, 1995). That growth is based on large fixed costs,
information gathering, coordination, and adaptation of both process and expectations
(Arthur, 1994). Arthur (1989) argues that “increasing returns can cause the economy
gradually to lock itself in to an outcome not necessarily superior to alternatives, not easily
altered, and not entirely predictable,” (p. 128). Pierson (2000c) and North (1990) note
that political institutions operate in very much the same way. Therefore, as the economic
and political processes of a city unfold over time into a set of institutional arrangements,
they may make change “difficult” and “unattractive” (Pierson, 2000a, p. 491). Historical
processes also create “increasing social complexity,” which generates “problems of
overload” and “growing interaction effects,” (Pierson, 2000a, p. 483). Thus, a change in

235
the distribution of environmental hazards may be desirable, but a city’s historical path
and social complexity may make change exceedingly difficult, particularly in large cities
with myriad urban problems. Of course, the distribution of disamenities is also affected
by a city’s infrastructure (e.g., transportation) and housing policies. In the case of
disamenities, workers may need to live closer to work because housing is more affordable
and public transportation is available. Lower-skilled positions offered at many industrial
locations may often be filled by poor and/or minority populations, which may create a
cycle of environmental inequity for workers.
Can the same be said for environmental amenities? Certainly, path dependency is
at work for environmental amenities as well. However, two things separate amenities
from hazards. In the case of the latter, toxic releases and hazardous waste are the by-
products of industry and economic externalities. Governments must try to constrain those
externalities, while still maintaining the very industries that create them for economic
purposes. Amenities are public goods that cities find desirable and ends in themselves.
Another difference is that neighborhood governance and nonprofit organizations play a
very important role in promoting and maintaining environmental amenities. Thus, the
“path” of environmental amenities may be easier to alter than that of environmental
disamenities. In other words, disamenities may be intractable, while amenities may be
more governable and open to change.

236
Conclusion

While the results of the institutional analysis were disappointing, one bright spot
emerged: Boston was ranked as the most equitable city in both the distribution of hazards
and the distribution of goods. No other city in the study ranked similarly in the
distribution of disamenities and amenities. For example, Little Rock was the least
equitable of the medium-sized cities in its disamenities distribution, but was the second-
most equitable in terms of amenities. Likewise, Flagstaff rated well on hazards, but was
the least equitable in terms of amenities. What is it about Boston that makes it a more
equitable city in both categories? I conclude this chapter with a brief look at Boston.
I chose Boston for this study because it was a large city that represented the Rust
Belt and Northeast. With a population of 589,141, it was on the small end of the large
cities. I knew it would have a good number of community gardens and it fit my need for a
mayor-council government. I also knew that Boston’s current economy revolved around
education and research (e.g., educational, medical); high-technology, government, and
fishing and fish processing, a change from its more industrial past. And, according to the
2000 Census, Boston’s minority population was made up primarily of African
Americans, Hispanics, and to a smaller degree, Asians. Beyond that, I knew very little
about Boston. While Boston only had 32 hazards, it boasted 397 amenities, the most of
any city in the study, even though it ranks in the middle in terms of square miles (89.66
land miles).
As for Boston’s governing structure, it is a strong mayor-council government,
allowing its mayor veto power and imposing no term limits. The city council has 13

237
elected members (9 district-based and 4 at-large). The interesting thing about Boston’s
city government is that it is based very much on neighborhood districts. Twenty-three
neighborhoods exist in Boston, though several are combined into one political district,
and they are coordinated by the Office of Neighborhood Services. It is through
neighborhood councils that citizens are encouraged to provide input to city policy,
participate formally, and provide constant feedback through a neighborhood liaison.
Boston also has an Office of Neighborhood Development. Boston’s focus on
neighborhoods creates strong alliances not only within neighborhoods, but between
neighborhoods and city government. Activism thrives, even in some of the poorer
neighborhoods. In their book, Streets of Hope, Medoff and Sklar (1994) chronicled the
revitalization of the Dudley Street neighborhood, which at one time was dilapidated,
riddled with urban problems, and rife with low-income citizens who felt peripheralized
and powerless to improve the neighborhood. Through the Dudley Street Neighborhood
Initiative and related neighborhood-level alliances, the citizens worked with city
government to improve housing, local infrastructure, the environment, and the well-being
of residents. For example, the neighborhood was once a leader in lead poisoning and
home to 60 hazardous sites (Medoff & Sklar, 1994), but the Dudley Street group worked
to clean it, prevent further waste industries from locating there, and instituted a greening
program that focused on parks, community gardens, and urban art.
Other local initiatives showed similar successes. Groups like Alternatives for
Community and Environment (ACE) and the Roxbury Environmental Empowerment
Project (REEP) worked with local government to improve environmental health
outcomes for issues like pollution-related asthma (Loh & Sugerman-Brozan, 2002).

238
Another example of a neighborhood-led effort is the Chelsea Creek Action Group, made
up of residents from two Boston neighborhoods and various city liaisons. The group
worked to revitalize Chelsea Creek and the surrounding neighborhoods through clean-up,
restoration, and greening projects (Chelsea Creek Action Group, 2003). Boston has a
long history of citizen involvement in local government, an effort that is mutually
reinforcing for both citizens and government.
Specifically, what are Boston’s environmental justice policies? Boston has an
Environment Department that provides conservation, restoration, pollution control, and
environmental review services. Boston’s city council also has a Committee on
Environment and Health, and environmental justice is incorporated in many of the city’s
initiatives and redevelopment plans. The city works with ACE and The Greater Boston
Environmental Justice Network to promote environmental justice (Faber, 2007). While
there are no specific city-level environmental-justice laws, Boston does have an
Environmental Justice Action Plan, which is supported by state-level initiatives.
Massachusetts has an Executive Office of Environmental Affairs, which has a formal
Environmental Justice Policy that moves beyond simple recognition or consideration of
the problem to specific priorities like cleaning-up waste sites, decreasing pollution,
redeveloping brownfields, improving compliance with EPA’s pollution requirements, and
providing grants to local governments for related efforts (American Bar Association,
2004). Massachusetts has a series of pollution reduction laws (e.g., The Massachusetts
Toxics Use Reduction Act), though no specific environmental justice law. Boston and
local organizations have also been the recipient several EPA Environmental Justice Small
Grants. These grants have funded projects to reduce the use of pesticides and chemicals

239
in public housing, to clean up the Chelsea creek area, to host roundtable discussions of
business owners and citizens, to coordinate education on toxins and contaminants, and to
conduct outreach to residents in Boston’s Chinatown. Though the grant amounts are
small, ranging from $15,000-$25,000, they offer financial support to Boston’s poor and
minority neighborhoods. This environmental justice infrastructure offers support and
resources for the state’s cities, like Boston, to combat environmental inequity. That
infrastructure, combined with Boston’s robust neighborhood organizations and citizen
participation, encourages and facilitates prevention, mitigation, and remediation of
environmental inequity, even without formal environmental equity laws.
Boston’s amenities, 249 city parks and 148 community gardens in this study, are
also managed through neighborhood organizations and nonprofit groups. Boston’s parks
and community gardens have played an integral role in the city’s history. Boston has
some of the oldest and most prestigious parks in the country, including the Boston
Commons, founded in 1634, which is the oldest public park in the United States.
According to the Trust for Public Land (2008), Boston has nearly 3,000 acres of city
parkland (this does not include national or state parklands), which translates into nearly 4
acres per 1,000 people. The parks are managed primarily by the Boston Department of
Parks and Recreation, but the department works in conjunction with the Massachusetts
Department of Conservation and Recreation, the Boston Conservation Commission, and
the Massachusetts Port Authority. Boston’s Department of Parks and Recreations also
works closely with the Boston’s Parks and Recreation Commission to maintain the
viability and safety of Boston’s parks. This mayoral commission includes citizens and

240
city staff, and makes policy recommendations to the mayor and council based on citizen
needs and interests.
Community gardens have enjoyed a rich history in Boston as well, dating back to
1895 when the Industrial Aid Society for the Prevention of Pauperism leased a farm to
the city’s poor so they could grow their own produce (Lawson, 2005). Following that
initial public garden, other city departments (e.g., The School Department) and nonprofit
groups (e.g., The Massachusetts Horticultural Society) initiated community gardens in
local schools. During both World Wars, Victory Gardens were popular in Boston, and the
gardens became popular once again in the 1970s (Lawson, 2005). Today, it is estimated
that 3,000 gardeners participate in 148 gardens that generate $1.5 million in fruits and
vegetables each year (American Community Gardening Association, 1998). Eleven
different state and local agencies and nonprofit organizations (e.g., Boston Urban
Gardens, Boston Natural Areas Network) provide land trust for the community gardens
and oversee their management. Several of the community garden nonprofit groups are
neighborhood specific (e.g., Dorchester Gardenlands Preserve, South End lower Roxbury
Open Space Land Trust).
Boston’s city parks and community gardens are both included in Boston’s
comprehensive plan, Open Space Plan 2002-2006: Restoring the Legacy…Fulfilling the
Vision (Boston Parks and Recreation Department, 2002). The plan offers a vision for
parks and community gardens that not only capitalizes on the natural habitat, it also
integrates lands not traditionally owned by the city (e.g., community gardens owned by
nonprofit land trusts) and promotes nonprofit partnerships and citizen participation to
maintain and promote Boston’s natural spaces. In keeping with the other city

241
departments, this plan for action calls for integration of all parks, gardens, etc. into a
single network, but is built around Boston’s local neighborhoods.
Boston, settled in 1630 by Puritans and incorporated in 1822, has a long and rich
history. Its institutional scaffolding is complex, yet it has revolved around a central theme
of citizen participation and active governance. From the Revolutionary War to today’s
environmental struggles, several keys to Boston’s success are evident. First, participation
and citizen input allow policy makers to respond to citizen needs. Second, the
involvement of nonprofit organizations in city politics allows for broader governance and
public management than traditional top-down structures do. Third, the mobilization and
strength of local neighborhoods, represented informally in neighborhood alliances and
formally in council representation and city departments, allows for collective action that
is tailored to those neighborhoods. Miller (2000) argues that social movements exist on
different scales and that “place-specific” efforts fuel more successful movements:

Political opportunity structures, potential organizational resources, social
and cultural institutions, collective identities, and the effects of broad-
scale social change vary from place to place. In short, the characteristics of
places affect the ability of organizations to mobilize and campaign
effectively. (p. 167)

Boston’s 23 neighborhoods allow for policy makers to address place-specific issues, like
environmental justice, by responding to specific needs and working with neighborhood
residents. In this way, local residents and activists can develop strategies that best address
their particular problem or need. Finally, Boston’s rich history of neighborhood
organizations and the city government’s cooperation with those organizations creates an

242
environment where both bridging social capital and linking social capital flourish. As I
discussed in Chapter 6, social capital creates norms of trust that promote collective action
and strengthens the meso-level networks that support social movement activities. It also
affords formerly independent or marginalized groups the social cohesion necessary to for
collective action.
Environmental equity, like other urban problems, is complex and path dependent.
Of the cities in this study, Boston, with its neighborhood-based governance structure and
citizen activism, is an exemplar that other cities, particularly the larger cities in the study,
might look to for solutions to a seemingly intractable problem.

243
CHAPTER EIGHT: CONCLUSION

What would we do, I ask, if we acted on the basis of the knowledge we
have, imperfect as it is?
Aaron Wildavsky, But Is It True?

In this dissertation, I examined the distribution of two environmental disamenities
and two environmental amenities in 18 cities in an attempt to understand if minority and
poor people were over-represented near disamenities and under-represented near
amenities. I also investigated the city institutional variables that may explain those
distributions. In this chapter, I review the major findings, identify the policy implications,
discuss the dissertation’s limitations, and suggest future research.

Summary of Major Empirical Findings

In this study, I used GIS technology, geostatistical analysis, and regression
analysis to identify the distribution of TRI facilities, large-quantity generators, city parks
and community gardens in 18 cities nationwide. While GIS is not new to the study of
environmental equity, I introduced new techniques (e.g., mean center analysis, Moran’s I,
Ripley’s K, hot-spot analysis) to identify if sites clustered within each city and if so,
where. In identifying clusters, I also described the sociodemographic information of the
residents closest to the disamenities and amenities.

244
Regarding TRI facilities, the mean centers for TRI facilities were in
predominantly minority tracts (with some tracts being as high as 98% minority) in eight
cities, and in low-income tracts in six cities. In Albany, Little Rock, Norfolk, and San
Diego, the tracts housing the biggest-producing TRI facilities were predominantly
minority, had median incomes of less than $30,000, and only had 15% of the population
with a bachelor’s degree or better. TRI sites clustered globally in 13 of 17 cities
according to the Moran’s I test, with five cities clustering at the .001 level or better.
Getis-Ord Gi* analysis showed local clustering in 12 cities; in some cases, the mean
minority percentage for tracts in the local “hot spots” was as high as 83%. “Cold spots”
were found in two cities, and in Salt Lake City, the cold spots were in predominantly
white and middle-income tracts. Though the traditional OLS regression models showed
mixed results, density, median income, and owner-occupied housing value were often
inversely related to TRI facilities. Due to the low R
2
and high variance inflation factors
for the race variables in most regressions, no pattern was explained by the traditional
variables used in environmental equity analysis. R
2
was improved in the geographically
weighted regressions in three cities, but no pattern was evident.
Turning to LQGs, in six cities, the biggest-producing LQGs were in tracts that
were predominantly minority (in one case, the tract was 99% minority), while in five
cities, the biggest-producing LQGs were in low-income tracts (as low as $8,853 median
income). LQGs clustered in 14 of 17 cities according to Moran’s I analysis. Hot spots
were identified in 12 of 17 cities, 4 of which had mean minority percentages greater than
60%. The OLS regressions resulted in a low R
2
for most models, but median income,
density, and owner-occupied housing value were often inversely related to LQG

245
locations. In only one city, the geographically weighted regression returned a higher R
2
,
but nothing conclusive can be said about the pattern of LQG location using the usual
explanatory variables.
I used the same GIS analyses to examine the distribution of parks, which were
found to cluster globally in 83% of cities according to Moran’s I analysis, with
probability levels greater than .001 in 11 cities. Ripley’s K analysis also showed global
clustering in 72% of cities. Hot spots were identified in half of the cities; only three cities
had hot spots or park clusters in predominantly minority areas, while only two cities had
clusters in low-income areas. In some cities, the clusters were located in areas with mean
percentages of white residents as high as 72%. The OLS regressions produced low R
2
,
although owner-occupied house value was often positively correlated with park locations,
while median income was often inversely related. The geographically weighted
regressions did not return significantly better results, except in the case of one city.
Community gardens were found to cluster globally in 76% of cities using Moran’s
I analysis. Garden hot spots were found in 15 cities (where Z ≥1.96, p ≥ .05); mean
minority percentages ranged from 14%-73% and median income ranged from $22,000 to
nearly $74,000. Regressions were run in only Boston, Philadelphia, and Seattle, because
they had a sufficient number of gardens. A low R
2
resulted in nearly all of the OLS
models, while the geographically weighted regression in Boston produced an R
2
of .21.
Interestingly, percent minority was positively correlated in all three cities.
The GIS analysis suggests that in a majority of cities, the distributions of hazards
and amenities are not random, and often, the hazards cluster in minority and/or poor
areas, while amenities only sometimes cluster in minority and/or poor areas. Ideally, we

246
would see a random distribution (i.e., no global or local clustering) of hazards and
amenities. Still, the usual suspects (e.g., median income, owner occupied housing value,
race and ethnic variables, manufacturing jobs) failed to explain adequately the pattern of
hazards and amenities in most cities, even though they often correlated significantly. For
this reason, I examined the city institutional variables that might correlate with better
equity outcomes.
The institutional analysis was based on four structural propositions and four
process propositions. I ranked the cities based on the results of the GIS analyses I
performed and then examined the eight institutional propositions, first for disamenities
and then amenities. No clear pattern was evident in the institutional analysis. Overall, it
appeared that neighborhood organizations were important for increased equity in most
cities, though I did not include any measure of how strong or formalized those
neighborhood organizations were. Further, increased equity was seen in cities with both
the mayor council form of government and neighborhood organizations. Whether or not a
mayor had term limits or veto power did not seem to affect equity very strongly. For the
process variables, no clear pattern emerged, though overall, cities that had voted
Democrat in the two previous presidential elections had increased equity in the
distribution of amenities and often in the distribution of disamenities. Based on the
analysis, I concluded that a city’s history and institutional scaffolding lock it on a certain
trajectory (i.e., what Pierson [2000a] calls path dependence) and make change very
difficult to effect. As North (2005a) suggests, change will only occasionally be radical,
even for an important issue like the distribution of environmental hazards and
environmental goods.

247
Implications and Policy Recommendations

If, as I suggest, environmental equity is subject to the path dependency of city
governments, or the historical decisions made about economic growth, housing,
transportation, and zoning, how can a city promote better equity distributions? I
examined Boston, which had the best equity ratings of all cities in both disamenities and
amenities, and found a very strong mayor-council government backed by very strong
neighborhood organizations. The city government was organized around neighborhood
districts, and citizens worked within neighborhood organizations to solve environmental
issues. The city also worked in conjunction with nonprofit organizations to solve urban
issues, and in the case of amenities, Boston had a very detailed master plan that included
goals for green spaces of all types and ownership, and also incorporated formal citizen
participation. In the case of disamenities, Boston was one of the few cities that worked
closely with the state to implement environmental equity programs, even without formal
environmental justice legislation. Boston’s networks of neighborhood organizations and
their formal involvement in city government also created bridging and linking social
capital, which strengthens meso-level structures and fosters collective action efforts.
What does Boston’s success suggest for policy makers elsewhere?
First, environmental equity is a local problem that needs to be resolved locally.
Sometimes, action needs to be taken specifically in neighborhoods. Local constituents
must be involved so that reforms target specific needs. As in the case of Boston’s Dudley
Street neighborhood or Chelsea Creek neighborhood, local residents formed groups and
brought their issues to the attention of a very responsive city government. Environmental

248
justice has been on the national policy agenda for 25 years and is even the subject of an
Executive Order, yet not a single environmental justice bill has made it out of committee.
Thus, change must be made locally and must start at the neighborhood level.
Second, cities must encourage citizen participation and citizens must become
involved in solving environmental equity problems, which Irazábal (2005) notes can be a
challenge for urban governance. However, Boston is an example of a city that has
mayoral commissions made up of residents, neighborhood organizations composed of
residents, and city departments that are receptive to working with citizens. Empowering
citizens of all ethnicities and races, all education levels, and all income backgrounds is
the only way to promote social change (Brulle, 2000; Macedo & Karpowitz, 2006),
particularly with a complex social problem like environmental equity.
Third, Boston’s success in improving environmental equity in local areas was
often based on cleaning hazards, but also on improving the area with natural habitat
restoration or urban art. Thus, a focus on the distribution of amenities and increasing
those amenities may provide a cultural or frame shift, in which the dialogue on
environmental equity becomes one of promoting public goods and their attendant benefits
to all citizens rather than the early environmental justice discourse that often employed
incendiary language and accusations about intentional racism and classism. A shift to a
more positive frame may lead to enhanced cooperation between hazard producers, local
governments, and citizens.
Fourth, as this study has shown, no one is immune to environmental hazards or
the negative externalities of industry and economic growth. While working to reduce
exposure to environmental risks, policy makers should also implement programs that

249
mitigate those risks. For example, in their sustainability programs, cities should include
plans for reducing exposure to hazards, as well as increasing amenities that provide
health benefits. At the federal level, the EPA can opt for violators to implement
Supplemental Environmental Projects (SEPs), or environmentally beneficial projects that
improve and safeguard public health (e.g., greenbelt construction, brownfield
restoration), instead of simply imposing a fine.
Finally, the information on environmental hazards is woeful. The rules for
reporting on pollution and compliance with federal pollution standards should be
strengthened, as should EPAs enforcement of those rules. Until the EPA has sufficient
data on pollution amounts, researchers—environmental justice researchers,
epidemiologists, public health researchers, risk analysts—will continue to struggle to
produce information needed by policy makers to implement new legislation.

Limitations and Future Research

This dissertation is limited in several ways. First, the hazard distribution analysis
is limited by the EPA data on TRI facilities and releases and LQGs and their hazards
managed. Data are often missing or firms do not file reports with the EPA, so there may
be active TRI facilities or LQGs that are not reflected in the data presented here. In
addition, geolocating facilities has become much easier with Google Earth, but
sometimes the EPA address does not reflect additional facilities or firms that have
moved. Similarly, the amenity distribution is limited in that no federal agency keeps track
of community gardens. In many cases, even cities do not maintain records on community

250
gardens. Although I searched government records, nonprofit records, and even conducted
aerial surveys of some cities, I may not have accounted for all of the community gardens
in existence. Future research should include field visits to cities to verify the existence
and function of both hazards and amenities.
Next, the dissertation is limited by its national scope. What I tried to achieve in
terms of the breadth of the analysis, I lost in terms of depth. GIS is a very labor-intensive
technology. In trying to analyze four types of sites in 18 cities, I sacrificed a more
intricate analysis for a more general one. Similarly, I had hoped to provide comparability
by selecting 18 diverse cities; however, in the end the cities were so diverse (e.g., in size)
that it was hard to compare all 18 together. Future research should separate cities
according to size and perhaps geographic region. In this way, more detailed analyses, like
adding a layer of cadastral data to identify land ownership and natural features (e.g.,
lakes, ponds) can be added. Additional analyses on wind and water trajectories of TRI
releases could also be done with fewer cities.
This dissertation is also limited in its institutional analysis. The limited number of
cities did not allow for traditional hypothesis testing. As well, with fewer cities,
insufficient variance was seen in the institutional variables, which made it difficult to
analyze the institutional propositions. Further research should be expanded to additional
cities to identify sufficient variance and additional institutional variables that may factor
into environmental equity. From there, select variables can be examined using in-depth
case study analysis.
Finally, this dissertation is limited in its cross-sectional approach. Although the
decisions on economic development, housing, transportation, etc., were decades in the

251
making, this dissertation viewed only a snapshot of environmental data and Census data.
Others (c.f., Boone, 2002; Been, 1994) have argued that environmental justice must be
examined using historical analysis. Future research should trace the steps of economic
development, housing patterns, segregation, and other urban phenomena to find the root
causes of environmental equity.
In spite of the dissertation’s limitations, I examined environmental equity on a
national scale and included two amenities and two disamenities in its analysis, using
different spatial statistics than are normally used in environmental equity. Future research
may correct for the limitations here and create additional knowledge about the
distribution of hazards and amenities and the causes of those distributions. Still, cities can
begin to follow the lead of Boston and ameliorate their environmental equity problems,
one change at a time. Short an environmental disaster like the Union Carbide accident in
Bhopal, India, or an “urban mega-crisis” (Waste, 1998, p. 24), cities will continue to
receive little guidance from the federal government on the critical issue of environmental
justice. Thus, they must look within for solutions using the existing information as
imperfect as it may be.

252
BIBLIOGRAPHY
Adeola, F. O. (1994). Environmental hazards, health, and racial inequity in hazardous
waste distribution. Environment and Behavior, 26(1), 99-126.

Alaimo, K., Packnett, E., Miles, R. A., & Kruger, D. J. (2008). Fruit and vegetable intake
among urban community gardeners. Journal of Nutrition Education and Behavior, 40(2),
94-101.

Allen, B. L. (2003). Uneasy alchemy: Citizens and experts in Louisiana’s chemical
corridor disputes. Cambridge, MA: The MIT Press.

Ali, K., Partridge, M. D., & Olfert, M. R. (2007). Can geographically weighted
regressions improve regional analysis and policy making? International Regional Science
Review, 30(3), 300-329.

Amenta, E., & Caren, N. (2007). The legislative, organizational, and beneficiary
consequences of state-oriented challengers. In D. A. Snow, S. A. Soule, & H. Kriesi
(Eds.), The Blackwell companion to social movements (pp. 461-488). Malden, MA:
Blackwell Publishing Ltd.

Amenta, E. & Young, M. P. (1999). Making an impact: Conceptual and methodological
implications of the collective goods criterion. In M. Giugni, D. McAdam, & C. Tilly
(Eds.), How social movements matter (pp. 22-41). Minneapolis: University of Minnesota
Press.

American Bar Association. (2004). Environmental justice for all: A fifty-state survey of
legislation, policies, and initiatives. Chicago: Author.

American Community Gardening Association. (1998). National community gardening
survey. Philadelphia: Author.

Anderson, A. B., Anderton, D. L., & Oakes, J. M. (1994). Environmental equity:
Evaluating TSDF siting over the past two decades. Waste Age, 25, 83-100.

Anderton, D. L. (1996). Methodological issues in the spatiotemporal analysis of
environmental equity. Social Science Quarterly, 77(3), 508-515.

Anderton, D. L., Anderson, A. B., Oakes, J. M., & Fraser, M. R. (1994). Environmental
equity: The demographics of dumping. Demography, 31(2), 229-248.

Anderton, D. L., Anderson, A. B., Rossi, P. H., Oakes, J. M., Fraser, M. R., Weber, E.
W., et al. (1994). Hazardous waste facilities: “Environmental equity” issues in
metropolitan areas. Evaluation Review, 18(2), 123-140.

253
Andranovich, G. D., & Riposa, G. (1993). Doing urban research. Newbury Park, CA:
SAGE Publications.

Anselin, L., & Getis, G. (1992). Spatial statistical analysis and geographic information
systems. The Annals of Regional Science, 26, 19-33.

Armstrong, D. (2000). A survey of community gardens in upstate New York:
Implications for health promotion and community development. Health & Place, 6, 319-
327.

Arora, S., & Cason, T. N. (1999). Do community characteristics influence environmental
outcomes? Evidence from the Toxics Release Inventory. Southern Economic Journal,
65(4), 691-716.

Arthur, W. B. (1989). Competing technologies, increasing returns, and lock-in by
historical events. The Economic Journal, 99, 116-131.

Arthur, W. B. (1994). Increasing returns and path dependence in the economy. Ann
Arbor: University of Michigan Press.

Ash, M., & Fetter, T. R. (2004). Who lives on the wrong side of the environmental
tracks? Evidence from the EPA’s Risk-Screening Environmental Indicators model. Social
Science Quarterly, 85(2), 441-462.

Asch, P., & Seneca, J. J. (1978). Some evidence on the distribution of air quality. Land
Economics, 54(3), 278-297.

Atlas, M. (2002a). There goes the neighborhood: Environmental equity and the location
of new hazardous waste management facilities. Policy Studies Journal, 30(2), 171-192.

Atlas, M. (2002b). Few and far between? An environmental equity analysis of the
geographic distribution of hazardous waste generation. Social Science Quarterly, 83(1),
365-378.

Ayers, J. G. (2002). Chronic effects of air pollution. Occupational Environmental
Medicine, 59, 147-148.

Bachrach, P., & Baratz, M. (1962). Two faces of power. American Political Science
Review, 56(4), 947-952.

Babbie, E. (1992). The practice of social research (6
th
ed.). Belmont, CA: Wadsworth
Publishing Company.

254
Baden, B. M., Noonan, D. S., & Turaga, R. M. A. (2007). Scales of justice: Is there a
geographic bias in environmental equity analysis? Journal of Environmental Planning
and Management, 50(2), 163-185.

Banfield, E. (1961). Political influence. Glencoe, IL: Free Press.

Baybeck, B., & DeLorenzo, L. (2001). Reevaluating environmental justice: Assessing
unequal toxin and hazard dispersion in the United States. Paper presented at the meeting
of the Western Political Science Association, Las Vegas, NV.

Beatley, T. (2000). Green urbanism: Learning from European cities. Washington, D. C.:
Island Press.

Beck, U. (1992). Risk society: Towards a new modernity. London: Sage Publications, Inc.

Beck, U. (1995). Ecological politics in an age of risk. Cambridge, MA: Blackwell
Publishers Inc.

Been, V. (1994). Locally undesirable land uses in minority neighborhoods:
Disproportionate siting or market dynamics? Yale Law Journal, 103, 1383-1422.

Been, V. (1995). Analyzing evidence of environmental justice. Journal of Land Use and
Environmental Law, 11(1), 1-36.

Been, V. with Gupta, F. (1997). Coming to the nuisance or going to the Barrios? A
longitudinal analysis of environmental justice claims. Ecology Law Quarterly, 24(1), 1-
56.

Been, V., & Voicu, I. (2007). The effect of community gardens on neighboring property
(Working Paper 07-04). New York: New York University Furman Center for Real Estate
& Urban Policy.

Bevc, C. A., Marshall, B. K., & Picou, J. S. (2007). Environmental justice and toxic
exposure: Toward a spatial model of physical health and psychological well-being. Social
Science Research, 36, 48-67.

Bishaw, A. (2005). Areas with concentrated poverty: 1999 (CENSR-16). Washington, D.
C.: U.S. Department of Commerce.

Black, T., & Stewart, J. A. (2003). Burning and burying in Connecticut: Are regional
solutions to solid waste disposal equitable? In G. R. Visgilio, & D. M. Whitelaw (Eds.),
Our backyard: A quest for environmental justice (pp. 61-81). Lanham, MD: Rowman &
Littlefield Publishers, Inc.

255
Blanco, G. A., & Cooper, E. L. (2004). Immune systems, Geographic Information
Systems (GIS), environment, and health effects. Journal of Toxicology and
Environmental Health Part B, 7, 465-480.

Boer, J. T., Pastor, M., Jr., Sadd, J. L., & Snyder, L. D. (1997). Is there environmental
racism? The demographics of hazardous waste in Los Angeles County. Social Science
Quarterly, 78(4), 793-810.

Bolin, B., Nelson, A., Hackett, E. J., Pikawka, K. D., Smith, C. S., Sicotte, D., et al.
(2002). The ecology of technological risk in a Sunbelt City. Environment and Planning A,
34, 317-339.

Bolin, B., Grineski, S., & Collins, T. (2005). The geography of despair: Environmental
racism and the making of South Phoenix, Arizona, USA. Human Ecology Review, 12(2),
156-168.

Boone, C. G. (2002). An assessment and explanation of environmental inequity in
Baltimore. Urban Geography, 23(6), 581-595.

Boone, C. G., & Modarres, A. (1999). Creating a toxic neighborhood in Los Angeles
County: A historical examination of environmental inequity. Urban Affairs Review,
35(2), 163-187.

Boone, C. G., & Modarres, A. (2006). City and environment. Philadelphia: Temple
University Press.

Boston Parks and Recreation Department (2002). Open Space Plan 2002-2006: Restoring
the Legacy…Fulfilling the Vision. Retrieved March 1, 2009, from
http://www.cityofboston.gov/parks/openspace_doc.asp

Bowen, W. M. (1999). Comments on “‘Every breath you take…’: The demographics of
toxic air releases in Southern California.” Economic Development Quarterly, 13(2), 124-
134.

Bowen, W. M. (2001). Environmental justice through research-based decision-making.
NY: Garland Publishing.

Bowen, W. (2002). An analytical review of environmental justice research: What do we
really know? Environmental Management, 29(1), 3-15.

Bowen, W. M., Salling, M. J., Haynes, K. E., & Cyran, E. J. (1995). Toward
environmental justice: Spatial equity in Ohio and Cleveland. Annals of the Association of
American Geographers, 85(4), 641-663.

256
Bowen, W. M., & Wells, M. V. (2002). The politics and reality of environmental justice:
A history and consideration for public administrators and policy makers. Public
Administration Review, 62(6), 688-698.

Bowman, A. O’M., & Crews-Meyer, K. A. (1997). Locating southern LULUs: Race,
class, and environmental justice. State and Local Government Review, 29(2), 110-119.

Brajer, V., & Hall, J. V. (1992). Recent evidence of the distribution of air pollution
effects. Contemporary Policy Issues, X, 63-71.

Brehm, J. M., Eisenhauer, B. W., & Krannich, R. S. (2006). Community attachments as
predictors of local environmental concern. The American Behavioral Scientist, 50(2),
142-165.

Brooks, N. & Sethi, R. (1997). The distribution of pollution: Community characteristics
and exposure to air toxics. Journal of Environmental Economics and Management, 32,
233-250.

Brown, B. B., Perkins, D. D., & Brown, G. (2003). Place attachment in a revitalizing
neighborhood: Individual and block levels of analysis. Journal of Environmental
Psychology, 23, 259-271.

Brulle, R. J. (2000). Agency, democracy, and nature: The U.S. environmental movement
from a critical theory perspective. Cambridge, MA: The MIT Press.

Brulle, R. J., & Jenkins, J. C. (2006). Spinning our way to sustainability? Organization
and Environment, 19(1), 82-97.

Brunsdon, C. (2008). Inference and spatial data. In J. P. Wilson & A. S. Fotheringham
(Eds.), The handbook of geographic information science (pp. 337-351). Malden, MA:
Blackwell Publishing.

Bryant, B. (Ed.). (1995). Environmental justice: Issues, policies, and solutions.
Washington, D. C.: Island Press.

Bullard, R. D. (1983). Solid waste sites and the Black Houston community. Sociological
Inquiry, 53, 273-288.

Bullard, R. D. (1990). Ecological inequities and the New South: Black communities
under siege. The Journal of Ethnic Studies, 17(4), 101-115.

Bullard, R. D. (1993a). Waste and racism: A stacked deck? Forum for applied research
and public policy. Knoxville, TN: University of Tennessee.

257
Bullard, R. D. (1993b). Confronting environmental racism: Voices from the grassroots.
Boston: South End Press.

Bullard, R. D. (1994). Unequal protection: Environmental justice and communities of
color. San Francisco: Sierra Club Books.

Bullard, R. D. (1996). Environmental justice: It’s more than waste facility siting. Social
Science Quarterly, 77(3), 491-499.

Bullard, R. D. (2000). Dumping in Dixie: Race, class, and environmental quality (3
rd

ed.). Boulder, CO: Westview Press.

Burke, L. M. (1993). Race and environmental equity: A geographic analysis in Los
Angeles. Geographic Information Systems, 44-50.

Buzzelli, M. (2007). Bourdieu does environmental justice? Probing the linkages between
population health and air pollution epidemiology. Health and Place, 13, 3-13.

Cable, S., Mix, T., & Hastings, D. (2005). Mission impossible? Environmental justice
activists’ collaborations with professional environmentalists and with academics. In D. N.
Pellow, & R. J. Brulle (Eds.), Power, justice, and the environment: A critical appraisal of
the environmental justice movement (pp. 55-75). Cambridge, MA: The MIT Press.

Cannavò, P. F. (2007). The working landscape: Founding, preservation, and the politics
of place. Cambridge: The MIT Press.

Capek, S. M. (1993). The “environmental justice” frame: A conceptual discussion and an
application. Social Problems, 40(1), 5-24.

Center for Policy Alternatives. (2007). Environmental justice. Retrieved December 26,
2007, from http://www.stateaction.org/issues/issuecfm/issue/EnvironmentalJustice.xml

Chakraborty, J., & Armstrong, M. P. (1997). Exploring the use of buffer analysis for the
identification of impacted areas in environmental equity assessment. Cartography and
Geographic Information Systems, 24(3), 145-157.

Chakraborty, J., Schweitzer, L. A., & Forkenbrock, D. J. (1999). Using GIS to assess the
environmental justice consequences of transportation system changes. Transactions in
GIS, 3(3), 239-258.

Checker, M. (2005). Polluted promises: Environmental racism and the search for justice
in a southern town. New York: New York University Press.

258
Chelsea Creek Action Group. (2003). Chelsea Creek Community Based Comparative
Risk Assessment. Washington, D. C.: U.S. Environmental Protection Agency. Retrieved
March 5, 2007, from
http://www.epa.gov/region01////eco/uep/boston/TableSummaryIntro.pdf

Clark, R. D., Lab, S. P., & Stoddard, L. (1995). Environmental equity: A critique of the
literature. Social Pathology, 1(3), 253-269.

Clarke, L. (1989). Acceptable risk: Making decisions in a toxic environment. Berkeley,
CA: University of California Press.

Clayton, S. (2000). Models of justice in the environmental debate. Journal of Social
Issues, 56(3), 459-474.

Clingenmayer, J. C., & Feiock, R. C. (2001). Institutional constraints and policy choice:
An explanation of local governance. Albany, NY: State University of New York Press.

Cole, L. W., & Foster, S. R. (2001). From the ground up: Environmental racism and the
rise of the environmental justice movement. New York: New York University Press.

Council on Environmental Quality. (1997). Environmental justice: Guidance under the
National Environmental Policy Act. Retrieved February 15, 2009, from
http://www.lm.doe.gov/env-justice/pdf/justice.pdf

Cromley, E. K., & McLafferty, S. L. (2002). GIS and public health. NY: The Guilford
Press.

Cutter, S. L. (1995). Race, class, and environmental justice. Progress in Human
Geography, 19(1), 111-122.

Cutter, S. L. (2006a). The forgotten casualties: Women, children, and environmental
change. In S. L. Cutter (Ed.), Hazards, vulnerability, and environmental justice (pp. 49-
67). London: Earthscan.

Cutter, S. L. (Ed.). (2006b). Hazards, vulnerability, and environmental justice. London:
Earthscan.

Cutter, S. L. (2006c). Issues in environmental justice. In S. L. Cutter (Ed.), Hazards,
vulnerability, and environmental justice (pp. 263-269). London: Earthscan.

Cutter, S. L., Boruff, B. J., & Shirley, W. L. (2003). Social vulnerability to environmental
hazards. Social Science Quarterly, 84(2), 243-261.

Cutter, S. L., Holm, D., & Clark, L. (1996). The role of geographic scale in monitoring
environmental justice. Risk Analysis, 16(4), 517-526.

259

Cutter, S. L., Scott, M. S., & Hill, A. A. (2002). Spatial variability in toxicity indicators
used to rank chemical risks. American Journal of Public Health, 92(3), 420-422.

Cutter, S. L., & Solecki, W. D. (1996). Setting environmental justice in space and place:
Acute and chronic airborne toxic releases in the Southeastern United States. Urban
Geography, 17(5), 380-399.

Dahl, R. (1961). Who governs? New Haven, CT: Yale University Press.

Daniels, G., & Friedman, S. (1999). Spatial inequality and the distribution of industrial
toxic releases: Evidence from the 1990 TRI. Social Science Quarterly, 80(2), 244-262.

Davidson, P. R. (2003). Risky business? Relying on empirical studies to assess
environmental justice. In G. R. Visgilio & D. M. Whitelaw (Eds.), Our backyard: A quest
for environmental justice (pp. 83-103). Lanham, MD: Rowman & Littlefield Publishers,
Inc.

Davidson, P., & Anderton, D. L. (2000). Demographics of dumping II: A national
environmental equity survey and the distribution of hazardous materials handlers.
Demography, 37(4), 461-466.

Davis, D. (2002). When smoke ran like water: Tales of environmental deception and the
battle against pollution. NY: Basic Books.

Davis, D. L., & Webster, P. S. (2002). The social context of science: Cancer and the
environment. The ANNALS of the American Academy of Political and Social Science,
584(1), 13-34.

de Souza Medeiros, R. (2008). Between conflict and cooperation: Dilemmas in the
relations between non-governmental organizations and the state in Brazil. Unpublished
doctoral dissertation, Boston University.

DeSocio, M. (2007). Business community structures and urban regimes: A comparative
analysis. Journal of Urban Affairs, 29(4), 339-366.

DeWitt, J. (2004). Civic environmentalism. In R. F. Durant, D. J. Fiorino, & R. O’Leary
(Eds.), Environmental governance reconsidered: Challenges, choices, and opportunities
(pp. 219-254). Cambridge, MA: The MIT Press.

Diani, M. (2003). Introduction: Social movements, contentious actions, and social
networks: “From metaphor to substance”? In M. Diani & D. McAdam (Eds.), Social
movements and networks: Relational approaches to collective action (pp. 1-18). Oxford,
England: Oxford University Press.

260
Dobson, A. (1993). Critical theory and green politics. In A. Dobson & P. Lucardie (Eds.),
The politics of nature: Explorations in green political theory (pp. 190-206). New York:
Routledge.

Dolinoy, D. C., & Miranda, M. L. (2004). GIS modeling of air toxics releases from TRI-
reporting and non-TRI-reporting facilities: Impacts for environmental justice.
Environmental Health Perspectives, 112(17), 1717-1724.

Donati, P. (1997). Environmentalism, postmaterialism, and anxiety. ARENA Journal, 8,
147-172.

Downey, L. (1998). Environmental injustice: Is race or income a better predictor? Social
Science Quarterly, 79(4), 766-778.

Dreier, P., Mollenkopf, J., & Swanstrom, T. (2004). Place matters: Metropolitics for the
twenty-first century (2
nd
edition, revised). Lawrence: University of Kansas Press.

Dryzek, J. S., Downes, D., Hunold, C., & Schlosberg, D. (2003). Green states and social
movements: Environmentalism in the United States, United Kingdom, Germany, and
Norway. New York: Oxford University Press.

Dryzek, J, & Goodin, R. E. (1986). Risk-sharing and social justice: The motivational
foundations of the post-war welfare state. British Journal of Political Science, 16(1), 1-
34.

Durkheim, E. (1982). The rules of sociological method and selected texts on sociology
and its method. New York: The MacMillan Press Ltd.

Eady, V. (2003). Environmental justice in state policy decisions. In J. Agyeman, R. D.
Bullard, & B. Evans (Eds.), Just sustainabilities: Development in an unequal world (pp.
168-181). Cambridge, MA: The MIT Press.

Eisinger, P. K. (1973). The conditions of protest behavior in American cities. American
Political Science Review, 67(1), 11-28.

Eisinger, P. (1998). City politics in an era of federal devolution. Urban Affairs Review,
33(3), 308-325.

Elkin, S. L. (1987). City and regime in the American republic. Chicago: The University
of Chicago Press.

Elster, J. (1992). Local justice: How institutions allocate scarce goods and necessary
burdens. New York: Russell Sage Foundation.

261
Elster, J. (2007). Explaining social behavior: More nuts and bolts for the social sciences.
New York: Cambridge University Press.

Environmental Protection Agency (EPA). (1992). Environmental equity: Reducing risk
for all communities (EPA-230-R-92-008). Washington, D. C.: Author.

Environmental Protection Agency (EPA). (1994). Presidential documents: Executive
Order 12898 of February 11, 1994. Retrieved December 27, 2007, from
http://www.epa.gov/fedregstr/eo/eo12898.pdf

Environmental Protection Agency (EPA). (2001). The Emergency Planning and
Community Right-to-Know Act: Section 313 release and other waste management
reporting requirements (EPA 260/K-01-001). Washington, D. C.: Author.

Environmental Protection Agency (EPA). (2006). United States Environmental
Protection Agency Charter: Local Government Advisory Committee. Retrieved
December 5, 2008, from http://www.epa.gov/ocir/scas_lgac/pdf/2006-0919_reneal-of-
lgac_charter.pdf
Environmental Protection Agency (EPA) Web site. (2007). Retrieved November 30,
2007, from http://www.epa.gov

Faber, D. (2007). A more “productive” environmental justice politics: Movement
alliances in Massachusetts for clean production and regional equity. In R. Sandler & P.C.
Pezzullo (Eds.), Environmental justice and environmentalism: The social justice
challenge to the environment (pp. 134-164). Cambridge, MA: The MIT Press.

Faber, D. (2008). Capitalizing on environmental justice: The polluter-industrial complex
in the age of globalization. Lanham, MD: Rowman & Littlefield Publishers, Inc.

Faber, D. R., & Krieg, E. J. (2002). Unequal exposure to ecological hazards:
Environmental injustices in the Commonwealth of Massachusetts. Environmental Health
Perspectives, 110(Supplement 2), 277-288.

Faber, D., & McCarthy, D. (2001). The evolving structure of the EJM in the United
States: New models for democratic decision-making. Social Justice Research, 14(4), 405-
421.

Faber, D. R., & McCarthy, D. (2005). Foundations for social change: Critical
perspectives on philanthropy and popular movements. Lanham, MD: Rowman &
Littlefield Publishers, Inc.

Feiock, R. C., Jeong, M. G, & Kim, J. (2003). Credible commitment and council-
manager government: Implications for policy instrument choices. Public Administration
Review, 63(5), 616-625.

262
Feiock, R. C., & Stream, C. (2001). Environmental protection versus economic
development: A false trade-off? Public Administration Review, 61(3), 313-321.

Ferris, D. (1994). A call for justice and equal protection. In R. D. Bullard (Ed.), Unequal
protection: Environmental justice and communities of color (pp. 298-319). San
Francisco: Sierra Club Books.

Field, R. C. (1998). Risk and Justice: Capitalist production and the Environment. In D.
Faber (Ed.), The struggle for ecological democracy: Environmental justice movements in
the United States (pp. 81-103). New York: The Guilford Press.

Figueroa, R. M. (1995). Debating the paradigms of justice: The bivalence of
environmental justice. Unpublished doctoral dissertation, University of Colorado at
Boulder.

Fletcher, R. H. (2003). From Love Canal to environmental justice: The politics of
hazardous waste on the Canada-U.S. border. Peterborough, Canada: Broadview Press.

Floyd, M. F., & Johnson, C. Y. (2002). Coming to terms with environmental justice in
outdoor recreation: A conceptual discussion with research implications. Leisure Sciences,
24, 59-77.

Ford, J. M., & Beveridge, A. A. (2004). “Bad” neighborhoods, fast food, “sleazy”
businesses, and drug dealers: Relations between the location of licit and illicit businesses
in the urban environment. Journal of Drug Issues, 34(1), 51-76.

Foreman, C. H., Jr. (1998). The promise and peril of environmental justice. Washington,
D. C.: Brookings Institution Press.

Foster, S. (2002). Environmental justice in an era of devolved collaboration. Harvard
Environmental Law Review, 26, 459-499.

Frazier, J. W., Margai, F. M., & Tettey-Flo, E. (2003). Race and place: Equity issues in
urban America. Boulder, CO: Westview Press.

Frederickson, H. G., Johnson, G. A., & Wood, C. H. (2004). The adapted city:
Institutional dynamics and structural change. Armonk, NY: M. E. Sharpe.

Freeman, A. M., III. (1972). Distribution of environmental quality. In A. V. Kneese & B.
T. Bower (Eds.), Environmental quality analysis: Theory and method in the social
sciences. Baltimore, MD: The Johns Hopkins University Press.

Fricker, R. D., Jr., & Hengartner, N. W. (2001). Environmental equity and the
distribution of toxic release inventory and other environmentally undesirable sites in
metropolitan New York city. Environmental and Ecological Statistics, 8, 33-52.

263

Frumkin, H. (2001). Beyond toxicity: Human health and the natural environment.
American Journal of Preventative Medicine, 20(3), 234-240.

Gamson, W. A., & Meyer, D. S. (1996). Framing political opportunity. In D. McAdam, J.
D. McCarthy, & M. N. Zald (Eds.), Comparative perspectives on social movements:
Political opportunities, mobilizing structures, and cultural framings (pp. 275-290). New
York: Cambridge University Press.

Gauna, E. (1995). Federal environmental citizen provisions: Obstacles and incentives on
the road to environmental justice. Ecology Law Quarterly, 22(1), 1-88.

Gelobter, M. (1992). Toward a model of “environmental discrimination.” In B. Bryant &
P. Mohai (Eds.), Race and the incidence of environmental hazards (pp. 64-81). Boulder,
CO: Westview Press.

Gelobeter, M. (1994). The meaning of urban environmental justice. Fordham Urban Law
Journal, XXI, 841-856.

General Accounting Office (GAO). (1983). Siting of hazardous waste landfills and their
correlation with racial and economic status of surrounding communities (GAO/RCED-
83-168). Washington, D. C.: Author.

Gering, J. (2004). What is a case study and what is it good for? American Political
Science Review, 98(2), 341-354.

Geschwind, S. A., Stolwijk, J. A. J., Bracken, M., Fitzgerald, E., Stark, A., Olsen, C., et
al. (1992). Risk of congenital malformations associated with proximity to hazardous
waste sites. American Journal of Epidemiology, 135, 1197-1207.

Getis, A. (2008). A history of the concept of spatial autocorrelation: A geographer’s
perspective. Geographical Analysis, 40, 297-309.

Getis, A., & Ord, J. K. (1992). The analysis of spatial association by use of distance
statistics. Geographical Analysis, 24(3), 189-206.

Gies, E. (2008). Playing it smart. Land & People, 20(2), 22-31.

Girdner, E. J., & Smith, J. (2003). Killing me softly: Toxic waste, corporate profit, and
the struggle for environmental justice. New York: Monthly Review Press.

Giugni, M. (1999). How social movements matter: Past research, present problems,
future developments. In M. Giugni, D. McAdam, & C. Tilly (Eds.), How social
movements matter (pp. xiii-xxxiii). Minneapolis: University of Minnesota Press.

264
Glickman, T. S., Golding, D., & Hersh, R. (1995). GIS-based environmental equity
analysis: A case study of TRI facilities in the Pittsburgh area. In W. A. Wallace and E. G.
Beroggi (Eds.), Computer supported risk management (pp. 95-114). Dordrecht, The
Netherlands: Kluwer.

Glickman, T. S. (1994, Summer). Measuring environmental equity with Geographic
Information Systems. RESOURCES, 2-6.

Glickman, T. S., & Hersh, R. (1995). Evaluating environmental equity: The impacts of
industrial hazards on selected social groups in Allegheny County, Pennsylvania
(Discussion Paper 95-13). Washington, D. C.: Resources for the Future.

Glover, T. D. (2004). Social capital in the lived experiences of community gardeners.
Leisure Sciences, 26, 143-162.

Goodchild, M. F., Anselin, L., Appelbaum, R.P., & Hartorn, B. H. (2000). Toward
spatially integrated social science. International Regional Science Review, 23(2), 139-
159.

Gragg, R. D., III., Gasana, J., & Christaldi, R. A. (2002). Molecular biomarkers as
measures of environmental justice: A proposed health-assessment paradigm.
International Journal of Public Administration, 25(2 & 3), 281-303.

Graham, J. D., Beaulieu, N. D., Sussman, D., Sadowitz, M., & Li, Y. (1999). Who lives
near coke plants and oil refineries? An exploration of the environmental inequity
hypothesis. Risk Analysis, 19(2), 171-186.

Greenberg, M., & Cidon, M. (1997). Broadening the definition of environmental equity:
A framework for states and local governments. Population Research and Policy Review,
16, 397-413.

Greenberg, M. R., & Renne, J. (2005). Where does walkability matter the most? An
environmental justice interpretation of New Jersey Data. Journal of Urban Health:
Bulletin of the New York Academy of Medicine, 82(1), 90-100.

Grineski, S., Bolin, B., & Boone, C. (2007). Criteria air pollution and marginalized
populations: Environmental inequity in metropolitan Phoenix, Arizona. Social Science
Quarterly, 88(2), 535-554.

Hamilton, J. T. (1993). Politics and social costs: Estimating the impact of collection
action on hazardous waste facilities. RAND Journal of Economics, 24(1), 101-125.

Hamilton, J. T. (1995). Testing for environmental racism: Prejudice, profits, political
power? Journal of Policy Analysis and Management, 14(1), 107-132.

265
Hamilton, J. T. (1999). Exercising property rights to pollute: Do cancer risks and politics
affect plant emission reductions? Journal of Risk and Uncertainty, 18(2), 105-124.

Hamilton, J. T., & Viscusi, W. K. (1999). Calculating risks? The spatial and political
dimensions of hazardous waste policy. Cambridge, MA: The MIT Press.

Harner, J., Warner, K., Pierce, J., & Huber, T. (2002). Urban environmental justice
indices. The Professional Geographer, 54(3), 318-331.

Harnik, P. (2003). The excellent city park system. San Francisco: Trust for the Public
Land.

Harvey, D. (1985). The urbanization of capital: Studies in the history and theory of
capitalist urbanization. Baltimore, MD: The Johns Hopkins University Press.

Harvey, D. (1989). The urban experience. Baltimore, MD: The Johns Hopkins University
Press.

Harvey, D. (1996). Justice, nature, and the geography of difference. Malden, MA:
Blackwell Publishers Inc.

Harvey, D. (2000). Spaces of hope. Berkeley, CA: University of California Press.

Haynes, K. E. (2004). Transportation location and environmental justice: A US
perspective. In D. A. Hensher, K. J. Button, K. E. Haynes, & P. Stopher (Eds.),
Handbook of transport geography and spatial systems 5 (pp. 43-63). London: Elsevier
Ltd.

Haynes, K. E., Lall, S. V., & Trice, M. P. (2001). Spatial issues in environmental equity.
International Journal of Environmental Technology and Management, 1(1/2), 17-31.

Hayward, S., Fowler, E., & Steadman, L. (2000). Index of leading environmental
indicators (5
th
ed.). San Francisco: Pacific Research Institute for Public Policy.

Heiman, M. K. (1996). Race, waste, and class: New perspectives on environmental
justice. Antipode, 28(2), 111-121.

Heitgard, J. L., Burg, J. R., & Strictland, H. G. (1995). A geographic information systems
approach to estimating and assessing National Priorities List site demographics: Racial
and Hispanic origin composition. International Journal of Occupational Medicine and
Toxicology, 4(3), 343-363.

Helfand, G. E., & Peyton, L. J. (1999). A conceptual model of environmental justice.
Social Science Quarterly, 80(1), 68-83.

266
Hero, R. E. (2005). Crossroads of equality: Race/ethnicity and cities in American
democracy. Urban Affairs Review, 40(6), 695-705.

Hersh, R. (1995). Race and industrial hazards: An historical geography of the Pittsburgh
region – 1900-1990 (Discussion Paper 95-18). Washington, D. C.: Resources for the
Future.

Hird, J. A. (1993). Environmental policy and equity: The case of Superfund. Journal of
Policy Analysis and Management, 12(2), 323-343.

Hird, J. A., & Reese, M. (1998). The distribution of environmental quality: An empirical
analysis. Social Science Quarterly, 79(4), 693-716.

Hiskes, R. P. (2006). Environmental rights, intergenerational justice, and reciprocity with
the future. Public Affairs Quarterly, 19(3), 177-194.

Hite, D. (2000). A random utility model of environmental equity. Growth and Change,
31, 40-58.

Hobbs, F., & Stoops, N. (2002). Demographic trends in the 20
th
century (CENSR-4).
Washington, D. C.: U.S. Department of Commerce.

Hofrichter, R. (Ed.). (2000). Reclaiming the environmental debate: The politics of health
in a toxic culture. Cambridge, MA: The MIT Press.

Hoidal, S. (2003). Returning to the roots of environmental justice: Lessons from the
inequitable distribution of municipal services. Minnesota Law Review, 88(1), 193-221.

Holmes, A., Slade, B. A., Cowart, L. (2000). Are minority neighborhoods exposed to
more environmental hazards? Allegations of environmental racism. Real Estate Review,
30(2), 50-57.

Hudnut III, W. H. (2004). Mayors’ forum: The unifying lens of green infrastructure.
Urban Land, 47-53.

Hurley, A. (1995). Environmental inequalities: Class, race and industrial pollution in
Gary, IN, 1945-1980. Chapel Hill, NC: The University of North Carolina Press.

Hurley, A. (1997). Fiasco at Wagner Electric: Environmental justice and urban
geography in St. Louis. Environmental History, 2(4), 460-481.

Hynes, H. P., & Howe, G. (2004). Urban horticulture in the contemporary United States:
Personal and community benefits. Acta Horticulturae, 643, 171-181.

267
Inoguchi, T., Newman, E., & Paoletto, G. (1999). Introduction: Cities and the
environment—toward ecopartnerships. In T. Inoguchi, E. Newman, & G. Paoletto (Eds.),
Cities and the environment: New approaches for eco-societies (pp. 1-16). New York:
United Nations University Press.

Institute of Medicine Committee on Environmental Justice. (1999). Toward
environmental justice: Research, education, and health policy needs. Washington, D. C.:
National Academy Press.

Irazábal, C. (2005). City making and urban governance in the Americas: Curitiba and
Portland. Aldershot, England: Ashgate Publishing Limited.

Jacobs, J. (1961). The death and life of great American cities. New York: Random
House.

Jacobson, J. O., Hengartner, N. W., & Louis, T. A. (2005). Inequity measures for
evaluations of environmental justice: A case study of close proximity to highways in
New York City. Environment and Planning A, 37, 21-43.

Jenkins, J. C. (1983). Resource mobilization theory and the study of social movements.
American Review of Sociology, 9, 527-553.

Jenkins, J. C., & Form, W. (2005). Social movements and social change. In T. Janoski, R.
Alford, A. Hicks, & M. S. Schwartz (Eds.), The handbook of political sociology (pp. 331-
349). New York: Cambridge University Press.

Jenkins, J. C., & Perrow, C. (1977). Insurgency of the powerless: Farm worker
movements (1946-1972). American Sociological Review, 42(2), 249-268.

Johnson, S. (2006). The ghost map: The story of London’s most terrifying epidemic—and
how it changed science, cities, and the modern world. New York: Riverhead Books.

Johnson, M., & Neiman, M. (2004). Courting business: Competition for economic
development among cities. In R. C. Feiock, (Ed.), Metropolitan governance: Conflict,
competition, and cooperation (pp. 124-126). Washington, D. D.: Georgetown University
Press.

Judd, D. R. (1997). Cities and the environment. In R. K. Vogel (Ed.), Handbook of
research on urban politics and policy in the United States (pp. 369-409). Westport, CT:
Greenwood Press.

Judd, D. R., & Swanstrom, T. (2008). City politics: The political economy of modern
America. New York: Pearson Longman.

268
Kamieniecki, S., & Steckenrider, J. (1997). Two faces of equity in Superfund
implementation. In S. Kamieniecki, G. A. Gonzalez, & R. O. Vos (Eds.), Flashpoints in
environmental policymaking: Controversies in achieving sustainability (pp. 129-154).
Albany, NY: State University of New York Press.

Kay, J. (1994). California’s endangered communities of color. In R. D. Bullard (Ed.),
Confronting environmental racism: Voices from the grass roots (pp. 155-188). Boston:
South End Press.

Kennedy, J. F. (1962). Special Message to the Congress Transmitting Reorganization
Plan I of 1962. Washington, D. C., January 30, 1962. Electronic document,
http://www.presidency.ucsb.edu/ws/print.php?pid=8699, accessed July 21, 2007.

Kennedy, L. G. (2004). Transportation and environmental justice. In K. Lucas (Ed.),
Running on empty: Transport, social exclusion, and environmental justice (pp. 155-179).
Bristol, U.K. The Policy Press.

King, G. Keohane, R. O., & Verba, S. (1994). Designing social inquiry: Scientific
inference in qualitative research. Princeton, NJ: Princeton University Press.

Kitschelt, H. P. (1986). Political opportunity structures and political protest: Anti-nuclear
movements in four democracies. British Journal of Political Science, 16(1), 57-85.

Klandermans, B., & Staggenborg, S. (Eds.). (2002). Methods of social movement
research. Minneapolis: University of Minnesota Press.

Klandermans, B., Staggenborg, S., & Tarrow, S. (2002). Conclusion: Blending methods
and building theories in social movement research. In B. Klandermans & S. Staggenborg,
(Eds.). Methods of social movement research (pp. 314-349). Minneapolis: University of
Minnesota Press.

Koenig, T. H., & Rustad, M. L. (2004). Toxic torts, politics, and environmental justice:
The case for criminal torts. Law & Policy, 26(2), 189-207.

Konisky, D. M. (2009). Inequities in enforcement? Environmental justice and
government performance. Journal of Policy Analysis and Management, 28(1), 102-21.

Konisky, D. M., Milyo, J., & Richardson Jr., L. E. (2008). Environmental policy
attitudes: Issues, geographical scale, and political trust. Social Science Quarterly, 89(5),
1066-1085.

Krieg, E. J. (1995). Toxic wastes, race, and class: A historical interpretation of greater
Boston. Unpublished doctoral dissertation, Northeastern University.

269
Krieg, E. J. (1998). Methodological considerations in the study of toxic wastes. The
Social Science Journal, 35(2), 191-201.

Krieg, E. J. (2005). Race and environmental justice in Buffalo, NY: A zip code and
historical analysis of ecological hazards. Society and Natural Resources, 18, 199-213.

Kriesel, W., Centner, T. J., & Keeler, A. G. (1996). Neighborhood exposure to toxic
releases: Are there racial inequities? Growth and Change, 27, 479-499.

Kriesi, H. (2007). Political context and opportunity. In D. A. Snow, S. A. Soule, & H.
Kriesi (Eds.), The Blackwell companion to social movements (pp. 67-91). Malden, MA:
Blackwell Publishing Ltd.

Kuo, F. E. (2001). Coping with poverty: Impacts of environment and attention in the
inner city. Environment & Behavior, 33(1), 5-34.

Lavalle, M., & Coyle, M. (1992). Unequal protection—The racial divide in
environmental law. National Law Journal, 21, S1-S12.

Lave, L. B., & Seskin, E. P. (1970). Air pollution and human health. Science, 169, 723-
733.

Lawson, B. (1995). Living for the city: Urban United States and environmental justice. In
L. Westra, & P.S. Wenz (Eds.), Faces of environmental racism: Confronting issues of
global justice (pp. 41-56). Lanham, MD: Rowman & Littlefield Publishers, Inc.

Lawson, L. J. (2005). City bountiful: A century of community gardening in America.
Berkeley, CA: University of California Press.

Lazarus, R. J. (1993). Environmental justice: The distributional effects of environmental
protection. Northwestern University Law Review, 87, 787-857.

Lester, J. P., & Allen, D. W. (1999). Environmental justice in the U.S.: Myths and
realities. Paper presented at the 1999 Western Political Science Association Meeting,
Seattle, WA.

Lester, J. P., Allen, D. W., & Hill, K. M. (2001). Environmental injustice in the United
States: Myths and realities. Boulder, CO: Westview Press.

Lewandowski, J. D. (2005). Capitalising sociability: Rethinking the theory of social
capital. In R. Edwards, J. Franklin, & J. Holland (Eds), Assessing social capital: Concept,
policy, and practice (pp. 14-28). Newcastle, United Kingdom: Cambridge Scholars
Publishing.

270
Lichterman, P. (1996). The search for political community: American activists
reinventing commitment. New York: Cambridge University Press.

Lindblom, C.E. (1977). Politics and markets: The world’s political-economic systems.
NY: Basic Books.

Liu, F. (1996). Urban ozone plumes and population distribution by income and race: A
case study of New York and Philadelphia. Journal of the Air & Waste Management
Association, 46, 207-215.

Liu, F. (1997). Dynamics and causation of environmental equity, locally unwanted land
uses, and neighborhood changes. Environmental Management, 21(5), 643-656.

Liu, F. (2001). Environmental justice analysis: Theories, methods, and practice. Boca
Raton, FL: Lewis Publishers.

Loh, P. & Sugerman-Brozan, J. (2002). Environmental justice organizing for
environmental health: Case study on asthma and diesel exhaust in Roxbury,
Massachusetts. The ANNALS of the American Academy of Political and Social Science,
584, 110-124.

Long, D. A., & Perkins, D. D. (2007). Community social and place predictors of sense of
community: A multilevel and longitudinal analysis. Journal of Community Psychology,
35(5), 563-581.

Lopez, R. (2002). Segregation and black/white differences in exposure to air toxics in
1990. Environmental Health Perspectives, 110(2), 289-295.

Lubell, M., Feiock, R. C., & Ramirez, E. (2005). Political institutions and conservation
by local governments. Urban Affairs Review, 40(6), 706-729.

Lynch, K. (1960). The image of the city. Cambridge, MA: The MIT Press.

Maantay, J. (2002). Mapping environmental injustices: Pitfalls and potential of
Geographic Information Systems in assessing environmental health and equity.
Environmental Health Perspectives, 110(Supplement 2), 161-171.

Maantay, J. (2007). Asthma and air pollution in the Bronx: Methodological and data
considerations in using GIS for environmental justice and health research. Health and
Place, 13, 32-56.

Maantay, J. A., Maroko, A. R., & Herrmann, C. (2007). Mapping population distribution
in the urban environment: The Cadastral-based Expert Dasymetric System (CEDS).
Cartography and Geographic Information Science, 34(2), 77-102.

271
Maantay, J., & Ziegler, J. (2006). GIS for the urban environment. Redlands, CA: ESRI
Press.

MacDonald, H. (2006). The American Community Survey: Warmer (more current), but
fuzzier (less precise) than the Decennial Census. Journal of the American Planning
Association, 72(4), 491-503.

Macedo, S., & Karpowitz, C. F. (2006). The local roots of American inequality. PS:
Political Science & Politics, 39(1), 59-64.

Main, K. D. (2007). Place attachment and MacArthur Park: A case study of the
importance of public space in an immigrant neighborhood and the implications for local
planning practice. Unpublished doctoral dissertation. University of California, Los
Angeles.

Mallin, K. (1990). Investigation of a bladder cancer cluster in northwestern Illinois.
American Journal of Epidemiology, 132, 96-106.

Markham, W. T., & Rufa, E. (1997). Class, race, and the disposal of urban waste:
Locations of landfills, incinerators, and sewage treatment plants. Sociological Spectrum,
17, 235-248.

Massey, D. (1994). Space, place, and gender. Minneapolis: University of Minnesota
Press.

Massey, D. S., & Denton, N. A. (1993). American apartheid: Segregation and the
making of the underclass. Cambridge, MA: Harvard University Press.

May, P. J., Burby, R. J., Ericksen, J. J., Handmer, J. W., Dixon, J. E., Michaels, S., et al.
(1996). Environmental management and governance: Intergovernmental approaches to
hazards and sustainability. London: Routledge.

McAdam, D. (1983). Tactical innovation and the pace of insurgency. American
Sociological Review, 48, 735-754.

McAdam, D. (1996). Conceptual origins, current problems, future developments. In D.
McAdam, J. D. McCarthy, and M. N. Zald (Eds.), Comparative perspectives on social
movements: Political opportunities, mobilizing structures, and cultural framings (pp. 23-
40). New York: Cambridge University Press.

McAdam, D. (1999). Political process and the development of black insurgency, 1930-
1970 (2
nd
ed.). Chicago: The University of Chicago Press.

272
McAdam, D. (2003). Beyond structural analysis: Toward a more dynamic understanding
of social movements. In M. Diani & D. McAdam (Eds.), Social movements and
networks: Relational approaches to collective action (pp. 281-298). Oxford, England:
Oxford University Press.

McAdam, D., McCarthy, J. D., & Zald, M. N. (1988). Social movements. In N. J.
Smelser (Ed.), Handbook of sociology (pp. 695-738). Newbury Park, CA: SAGE.

McAdam, D., McCarthy, J. D., & Zald, M. N. (Eds). (1996). Comparative perspectives
on social movements: Political opportunities, mobilizing structures, and cultural
framings. New York: Cambridge University Press.

McAdam, D., & Snow, D. A. (1997). Social movements: Readings in their emergence,
mobilization, and dynamics. Los Angeles: Roxbury Publishing.

McCabe, B. C., & Feiock, R. C. (2005). Nested levels of institutions: State rules and city
property taxes. Urban Affairs Review, 40(5), 634-654.

McCarthy, D., & King, L. (2005). Introduction: Environmental problems requiring social
solutions. In L. King & D. McCarthy (Eds.), Environmental sociology: From analysis to
action (pp. xi-xxx). Lanham, MD: Rowman & Littlefield Publishers, Inc.

McGurtry, E. M. (1997). From NIMBY to civil rights: The origins of the environmental
justice movement. Environmental History, 2(3), 301-323.

McMaster, R. B., Leitner, H., & Sheppard, E. (1997). GIS-based environmental equity
and risk assessment: Methodological problems and prospects. Cartography and
Geographic Information Systems, 24(3), 172-189.

Medoff, P. & Sklar, H. (1994). Streets of hope: The fall and rise of an urban
neighborhood. Boston: South End Press.

Mennis, J. (2002). Using GIS to create and analyze statistical surfaces of population and
risk for environmental justice analysis. Social Science Quarterly, 83(1), 281-297.

Mennis, J. L., & Jordan, L. (2004). The distribution of environmental equity: Exploring
spatial nonstationarity in multivariate models of air toxic releases. Annals of the
Association of American Geographers, 95(2), 249-268.

Middleton, J. (2003). Health, environment, and social justice. Local Environment, 18(2),
155-165.

Miller, B. A. (2000). Geography and social movements: Comparing antinuclear activism
in the Boston area. Minneapolis: University of Minnesota Press.

273
Milman, A. (2006). Geographic pollution mapping of power plant emissions to inform
ex-ante environmental justice analysis. Journal of Environmental Planning and
Management, 49(4), 587-604.

Minkoff, D. C. (2002). Macro-organizational analysis. In B. Klandermans & S.
Staggenborg (Eds.), Methods of social movement research (pp. 260-285). Minneapolis:
University of Minnesota Press.

Mitchell, A. (2005). The ESRI guide to GIS analysis: Spatial measurements and
statistics. Redlands, CA: ESRI Press.

Mitchell, J. T, Thomas, D. S. K., & Cutter, S. L. (1999). Dumping in Dixie revisited: The
evolution of environmental injustices in South Carolina. Social Science Quarterly, 80(2),
229-243.

Mohai, P., & Saha, R. (2006). Reassessing racial and socioeconomic disparities in
environmental justice research. Demography, 43(2), 383-399.

Mollenkopf, J. (1994). A phoenix in the ashes: The rise and fall of the Koch coalition in
New York City politics. Princeton, NJ: Princeton University Press.

Morello-Frosch, R., & Jesdale, B. M. (2006). Separate and unequal: Residential
segregation and estimated cancer risks associated with ambient air toxics in U.S.
metropolitan areas. Environmental Health Perspectives, 114(3), 386-393.

Morello-Frosch, R., Pastor, M., Jr., Porras, C., & Sadd, J. (2002) Environmental justice
and regional inequity in Southern California: Implications for future research.
Environmental Health Perspectives, 110(Supplement 2), 149-154.

Morello-Frosch, R., Pastor, M., & Sadd, J. (2001). Environmental justice and Southern
California’s “riskscape”: The distribution of air toxics exposures and health risks among
diverse communities. Urban Affairs Review, 36(4), 551-578.

Morello-Frosch, R., Pastor, M., Jr., & Sadd, J. (2002). Integrating environmental justice
and the precautionary principle in research and policy making: The case of ambient air
toxics exposures and health risks among schoolchildren in Los Angeles. The ANNALS of
the American Academy of Political and Social Science, 584(1), 47-68.

Morgan, D. R., England, R. E., & Pelissero, J. P. (2007). Managing urban America (6
th

ed.). Washington, D. C.: CQ Press.

Morland, K, & Wing, S. (2007). Food justice and health in communities of color. In R. D.
Bullard (Ed.), Growing smarter: Achieving livable communities, environmental justice
and regional equity (pp. 171-188). Cambridge, MA: The MIT Press.

274
Most, M. T., Sengupta, R., & Burgener, M. A. (2004). Spatial scale and population
assignment choices in environmental justice analyses. The Professional Geographer,
56(4), 574-586.

Napton, M. L., & Day, F. A. (1992). Polluted neighborhoods in Texas: Who lives there?
Environment and Behavior, 24(4), 508-526.

National Research Council. (2004). The 2000 Census: Counting under adversity.
Washington, D. C.: The National Academies Press.

Neumann, C. M., Forman, D. L., & Rothlein, J. E. (1998). Hazard screening of chemical
releases and environmental equity analysis of populations proximate to Toxic Release
Inventory Facilities in Oregon. Environmental Health Perspectives, 106(4), 217-226.

Nickelsburg, S. M. (1998). Mere volunteers? The promise and limits of community-based
environmental protection. Virginia Law Review, 84, 1371-1409.

Nicolaides, B. M. (2003). Suburbia and the sunbelt. Organization of American Historians
Magazine of History, 18(1), 21-26.

North, D. C. (1990). Institutions, institutional change, and economic performance. New
York: Cambridge University Press.

North, D. C. (2005a). Understanding the process of economic change. Princeton, NJ:
Princeton University Press.

North, D. C. (2005b). Institutions and the performance of economies over time. In C.
Menard and M. M. Shirley (Eds.), Handbook of New Institutional Economics (pp. 21-30).
The Netherlands: Springer.

Novotny, P. (1998). Popular epidemiology and the struggle for community health in the
environmental justice movement. In D. Faber (Ed.), The struggle for ecological
democracy. Environmental justice movements in the United States (pp. 137-158). New
York: The Guilford Press.

Novotny, P. (2000). Where we live, work, and play: The environmental justice movement
and the struggle for a new environmentalism. Westport, CT: Praeger.

Nowak, D. J. (2008). The effects of urban trees on air quality. Washington D. C.: U.S.
Forest Service. Retrieved May 16, 2008, from www.fs.fed.us/he/syracuse/gif/trees.pdf

Oakes, J. M., Anderton, D. L., & Anderson, A. B. (1996). A longitudinal analysis of
environmental equity in communities with hazardous waste facilities. Social Science
Research, 25, 125-148.

275
O’Neill, M. S., Veves, A., Zanobetti, A., Sarnat, J. A., Gold, D. R., Panayiotis, A. E., et
al. (2005). Diabetes enhances vulnerability to particulate air pollution—Associated
impairment in vascular reactivity and endothelial function. Circulation, III, 2913-2920.

Ostrander, S. A. (2005). Legacy and promise for social justice funding: Charitable
foundations and progressive social movements past and present. In D. R. Faber and D.
McCarthy (Eds.), Foundations for social change: Critical perspectives on philanthropy
and popular movements (pp. 33-59). Lanham, MD: Rowman & Littlefield Publishers,
Inc.

Ostrom, E. (2005). Understanding institutional diversity. Princeton, NJ: Princeton
University Press.

Paigen, B., Goldman, L. R., Highland, J. H., Magnant, M. M., & Steegman, A. T. (1987).
Growth of children living near the hazardous waste site, Love Canal. Human Biology, 59,
489-508.

Pastor, M. (2001). Building social capital to protect natural capital: The quest for
environmental justice (Working Paper No. 11). Amherst, MA: Political Economy
Research Institute.

Pastor, M., Jr. (2007). ¿Quién es más urbanista? Latinos and smart growth. In R. D.
Bullard (Ed.), Growing smarter: Achieving livable communities, environmental justice,
and regional equity (pp. 73-101). Cambridge, MA: The MIT Press.

Pastor, M., Morello-Frosch, R., & Sadd, J. L. (2005). The air is always cleaner on the
other side: Race, space, and ambient air toxics exposures in California. Journal of Urban
Affairs, 27(2), 127-148.

Pastor, M., Jr., Morello-Frosch, R., & Sadd, J. L. (2006). Breathless: Schools, air toxics,
and environmental justice in California. The Policy Studies Journal, 34(3), 337-362.

Pastor, M., Jr., Sadd, J., & Hipp, J. (2001). Which came first? Toxic facilities, minority
move-in, and environmental justice. Journal of Urban Affairs, 23(1), 1-21.

Pastor, M., Jr., Sadd, J. L., & Morello-Frosch, R. (2004a). Reading, writing, and toxics:
Children’s health, academic performance, and environmental justice in Los Angeles.
Environment and Planning C: Government and Policy, 22, 271-290.

Pastor, M., Jr., Sadd, J. L., & Morello-Frosch, R. (2004b). Waiting to inhale: The
demographics of toxic air release facilities in 21
st
-century California. Social Science
Quarterly, 85(2), 420-440.

Pellow, D. N. (2000). Environmental inequality formation: Toward a theory of
environmental injustice. American Behavioral Scientist, 43(4) 581-601.

276

Pellow, D. N. (2002). Garbage wars: The struggle for environmental justice in Chicago.
Cambridge, MA: The MIT Press.

Pellow, D. N. (2004). The politics of illegal dumping: An environmental justice
framework. Qualitative Sociology, 27(4), 511-525.

Pellow, D. N. (2007). Resisting global toxics: Transnational movements for
environmental justice. Cambridge, MA: The MIT Press.

Pellow, D. N., & Brulle, R. J. (2005). Power, justice, and the environment: Toward
critical environmental justice studies. In D. N. Pellow & R. J. Brulle (Eds.), Power,
justice, and the environment: A critical appraisal of the environmental justice movement
(pp. 1-19). Cambridge, MA: The MIT Press.

Pellow, D. N., & Park, L. S. (2002). The Silicon Valley of dreams: Environmental
injustice, immigrant workers, and the high-tech global economy. New York: New York
University Press.

Peña, D. G. (2005). Autonomy, equity, and environmental justice. In D. N. Pellow & R.
J. Brulle (Eds.), Power, justice, and the environment: A critical appraisal of the
environmental justice movement (pp. 131-151). Cambridge, MA: The MIT Press.

Perlin, S. A., Setzer, R. W., Creason, J., & Sexton, K. (1995). Distribution of industrial
air emissions by income and race in the United States: An approach using the Toxic
Release Inventory. Environmental Science and Technology, 29(1), 69-80.

Perlin, S. A., Sexton, K., & Wong, D. W. S. (1999). An examination of race and poverty
for populations living near industrial sources of air pollution. Journal of Exposure
Analysis and Environmental Epidemiology, 9(1), 29-48.

Peters, B. G. (1998). Comparative politics: Theory and methods. New York: New York
University Press.

Peters, A., & MacDonald, H. (2004). Unlocking the Census with GIS. Redlands, CA:
ESRI Press.

Peterson, P. E. (1981). City limits. Chicago: University of Chicago Press.

Peterson, P. E. (1995). The price of federalism. Washington, D. C.: The Brookings
Institution.

Petts, J. (2005). Enhancing environmental equity through decision-making: Learning
from waste management. Local Environment, 10(4), 397-409.

277
Phillips, C. V., & Sexton K. (1999). Science and policy implications of defining
environmental justice. Journal of Exposure Analysis and Environmental Epidemiology, 9,
9-17.

Phillips, R. (1995). Evanston community and environmental racism: A case study in
social philosophy. In L. Westra & P. S. Wenz (Eds.), Faces of environmental racism:
Confronting issues of global justice (pp. 93-112). Lanham, MD: Rowman & Littlefield
Publishers, Inc.

Pierre, J. (2005). Comparative urban governance: Uncovering complex causalities. Urban
Affairs Review, 40(4), 446-462.

Pierson, P. (2000a). The limits of design. Explaining institutional origins and change.
Governance: An International Journal of Policy and Administration, 13(4), 475-499.

Pierson, P. (2000b). Not just what, but when: Timing and sequence in political processes.
Studies in American Political Development, 94(2), 72-92.

Pierson, P. (2000c). Increasing returns, path dependence, and the study of politics.
American Political Science Review, 94(2), 251-267.

Pincetl, S., Wolch, J., Wilson, J., & Longcore, T. (2003). Toward a sustainable LA: A
“nature’s services” approach. Los Angeles: USC Center for Sustainable Cities.
Retrieved May 16, 2008, from
www.usc.edu/dept/geography/ESPE/documents/report_haynes.pdf

Pinderhughes, R. (1996). The impact of race on environmental quality: An empirical and
theoretical discussion. Sociological Perspectives, 39(2), 231-248.

Pinderhughes, R. (1997). Who decides what constitutes a pollution problem? Race,
Gender, and Class, 5(1), 130-152.

Platt, R. H. (1996) Land use and society: Geography, law, and public policy.
Washington, D. C.: Island Press.

Platt, R. H. (2004). Toward ecological cities. Environment, 46(5), 10-27.

Platt, H. L. (2005). Shock cities: The environmental transformation of Manchester and
Chicago. Chicago: The University of Chicago Press.

Pollack, P. H., III., & Vittas, M. E. (1995). Who bears the burdens of environmental
pollution? Race, ethnicity, and environmental equity in Florida. Social Science Quarterly,
76(2), 294-310.

278
Porter, R., & Tarrant, M. A. (2001). A case study of environmental justice and federal
tourist sites in Southern Appalachia: A GIS application. Journal of Travel Research, 40,
27-40.

Portney, K. E. (2003). Taking sustainable cities seriously: Economic development, the
environment, and quality of life in American cities. Cambridge, MA: The MIT Press.

Prakash, S. R. (2007). Beyond dirty diesels: Clean and just transportation in northern
Manhattan. In R. D. Bullard (Ed.), Growing smarter: Achieving livable communities,
environmental justice, and regional equity (pp. 273-298). Cambridge, MA: The MIT
Press.

Prow, T. (1999). The power of trees. Illinois Steward, 7(4), 2.

Pulido, L. (1996a). A critical review of the methodology of environmental racism
research. Antipode, 28(2), 142-159.

Pulido, L. (1996b). Introduction: Environmental racism. Urban Geography, 17(5), 377-
379.

Pulido, L. (2000). Rethinking environmental racism: White privilege and urban
development in Southern California. Annals of the Association of American Geographers,
90(1), 12-40.

Pulido, L., Sidawi, S., & Vos, R. O. (1996). An archaeology of environmental racism in
Los Angeles. Urban Geography, 17(5), 419-439.

Putnam, R. D. (2000). Bowling alone: The collapse and revival of American community.
New York: Simon & Schuster.

Rae, D. W. (2006). Making life work in crowded places. Urban Affairs Review, 41(3),
271-391.

Ralston, B. (2004). GIS and public data. Clifton, NY: Delmar Learning.

Ramos-Pinto, P. (2005). Social capital as a capacity for collective action. In R. Edwards,
J. Franklin, & J. Holland (Eds.), Assessing social capital: Concept, policy, and practice
(pp. 53-69). Newcastle, United Kingdom: Cambridge Scholars Publishing.

Rawls, J. (1971). A theory of justice. Cambridge, MA: The Belknap Press of Harvard
University Press.

Rawls, J. (1980). A Kantian conception of equality. In V. Held (Ed.), Property, profits,
and economic justice (pp. 198-208). Belmont, CA: Wadsworth Publishing.

279
Rawls, J. (1985). Justice as fairness: Political not metaphysical. Philosophy and Public
Affairs, 14, 223-251.

Rechtschaffen, C. & Gauna, E. (2002). Environmental justice: Law, policy, and
regulation. Durham, NC: Carolina Academic Press.

René, A. A., Daniels, D. E., & Martin, S. A. (2000). Impact of environmental inequity on
health outcome: Where is the epidemiological evidence. Journal of the National Medical
Association, 92, 275-280.

Rhodes, E. L. (2003). Environmental justice in America: A new paradigm. Bloomington,
IN: Indiana University Press.

Ringquist, E. J. (1993). Environmental protection at the state level: Politics and progress
in controlling pollution. Armonk, NY: ME Sharpe.

Ringquist, E. J. (1995). Is “effective regulation” always oxymoronic?: The states and
ambient air quality. Social Science Quarterly, 76(1), 69-87.

Ringquist, E. J. (1997). Equity and the distribution of environmental risk: The case of
TRI facilities. Social Science Quarterly, 78(4), 811-829.

Ringquist, E. J. (2004). Environmental justice. In R. F. Durant, D. J. Fiorino, & R.
O’Leary (Eds.), Environmental governance reconsidered: Challenges, choices, and
opportunities (pp. 255-287). Cambridge, MA: The MIT Press.

Ringquist, E. J. (2005). Assessing evidence of environmental inequities: A meta-analysis.
Journal of Policy Analysis and Management, 24(2), 223-247.

Ringquist, E. J., & Clark, D. H. (1999). Local risks, states’ rights, and federal mandates:
Remedying environmental inequities in the U.S. federal system. Publius: The Journal of
Federalism, 29(2), 73-93.

Romley, J. A., Cohen, D., Ringel, J., & Strum, R. (2007). Alcohol and environmental
justice: The density of liquor stores and bars in urban neighborhoods in the United States.
Journal of Studies on Alcohol and Drugs, 68, 48-55.

Rootes, C. (2007). Environmental movements. In D. A. Snow, S. A. Soule, & H. Kriesi
(Eds.), The Blackwell companion to social movements (pp. 608-640). Malden MA:
Blackwell Publishing Limited.

Ryan, R. L. (2005). Exploring the effects of environmental experience on attachment to
urban natural areas. Environment & Behavior, 37(1), 3-42.

280
Sadd, J. L., Pastor, M., Jr., Boer, J. T., & Snyder, L. D. (1999a). “Every breath you take”:
The demographics of toxic air releases in Southern California. Economic Development
Quarterly, 13(2), 107-123.

Sadd, J. L., Pastor, M., Jr., Boer, J. T., & Snyder, L. D. (1999b). Response to comments
by William M. Bowen. Economic Development Quarterly, 13(2), 135-140.

Saha, R., & Mohai, P. (2005). Historical context and hazardous waste facility siting:
Understanding temporal patterns in Michigan. Social Problems, 52(4), 618-648.

Salvidar-Tanaka, L., & Krasny, M. E. (2004). Culturing community development,
neighborhood open space, and civic agriculture: The case of Latino community gardens
in New York City. Agriculture and Human Values, 21, 399-412.

Sampson, R. J., & Raudenbush, S. W. (1999). Systemic social observation of public
spaces: A new look at disorder in urban neighborhoods. American Journal of Sociology,
105(3), 603-651.

Sandercock, L. (1998). Towards cosmopolis: Planning for multicultural cities. New
York: John Wiley & Sons.

Sartori, G. (1991). Comparing and miscomparing. Journal of Theoretical Politics, 3(3),
243-257.

Saxton, G. D., & Benson, M. A. (2005). Social capital and the growth of the nonprofit
sector. Social Science Quarterly, 86(1), 16-35.

Schlosberg, D. (2004). Reconceiving environmental justice: Global movements and
political theories. Environmental Politics, 13(3), 517-540.

Schlosberg, D. (2007). Defining environmental justice: Theories, movements, and nature.
New York: Oxford University Press.

Schnaiberg, A., & Gould, K. A. (2000). Environment and society: The enduring conflict.
Caldwell, NJ: The Blackburn Press.

Schweitzer, L. (2006). Environmental justice and hazmat transport: A spatial analysis in
southern California. Transportation Research Part D, 11, 408-421.

Schweitzer, L., & Stephenson, M., Jr. (2007). Right answers, wrong questions:
Environmental justice as urban research. Urban Studies, 44(2), 319-337.

Scott, L., & Pratt, M. (2009). Answering why questions: An introduction to using
regression analysis with spatial data. ArcUser, 12(2), 40-43.

281
Scott, M. S., & Cutter, S. L. (2006). Using relative risk indicators to disclose toxic hazard
information to communities. In S. L. Cutter (Ed.), Hazards, vulnerability, and
environmental justice (pp. 307-326). London: Earthscan.

Scott, M., Cutter, S. L., Menzel, C., Ji, M., & Wagner, D. (1997). Spatial accuracy of the
EPA’s environmental hazards databases and their use in environmental equity analyses.
Applied Geographic Studies, 1(1), 45-61.

Sellers, J. M. (1999). Public goods and the politics of segregation: An analysis and cross-
national comparison. Journal of Urban Affairs, 21(2), 237-262.

Sellers, J. M. (2002). Governing from below: Urban regimes and the global economy.
New York: Cambridge University Press.

Sellers, J. M., & Lidström, A. (2007). Decentralization, local government, and the
welfare state. Governance: An International Journal of Policy, Administration, and
Institutions, 20(4), 609-632.

Sexton, K., & Adgate, J. L. (1999). Looking at environmental justice from an
environmental health perspective. Journal of Exposure Analysis and Environmental
Epidemiology, 9(1), 3-8.

Shaikh, S. L., & Loomis, J. B. (1999). An investigation into the presence and causes of
environmental inequity in Denver, Colorado. The Social Science Journal, 36(1), 77-92.

Shapiro, M. D. (2005). Equity and information: Information regulation, environmental
justice, and risks from toxic chemicals. Journal of Policy Analysis and Management,
24(2), 373-398.

Sharp, E. B. (1997). Policy process. In R. K. Vogel (Ed.), Handbook of research on
urban politics and policy in the United States (pp. 275-289). Westport, CT: Greenwood
Press.

Sheppard, E., Leitner, H., McMaster, R. B., & Tian, H. (1999). GIS-based measures of
environmental equity: Exploring their sensitivity and significance. Journal of Exposure
Analysis and Environmental Epidemiology, 9(1), 18-28.

Sherer, P. M. (2006). The benefit of parks: Why America needs more city parks and open
space. San Francisco: The Trust for the Public Land.

Shinew, K. J., Glover, T. D., & Parry, D. C. (2004). Leisure spaces as potential sites for
interracial interaction: Community gardens in urban areas. Journal of Leisure Research,
36(3), 336-355.

282
Shrader-Frechette, K. (2002). Environmental justice: Creating equality, balancing
democracy. New York: Oxford University Press.

Shaikh, S. L., & Loomis, J. B. (1999). An investigation into the presence and causes of
environmental inequity in Denver, Colorado. The Social Science Journal, 36(1), 77-92.

Shutkin, W. A. (2000). The land that could be: Environmentalism and democracy in the
twenty-first century. Cambridge, MA: The MIT Press.

Sibley, D. (1995). Geographies of exclusion: Society and difference in the West. London:
Routledge.

Sicotte, D., & Swanson, S. (2007). Whose risk in Philadelphia? Proximity to unequally
hazardous industrial facilities. Social Science Quarterly, 88(2), 515-534.

Slovic, P. (1997). Trust, emotion, sex, politics, and science. In M. H. Bazerman, D. M.
Messick, A. E. Tenbunsel, & K. A. Wade-Benzoni (Eds.), Environment, ethics, and
behavior: The psychology of environmental valuation and degradation (pp. 277-313).
San Francisco: The New Lexington Press.

Sobotta, R. R., Campbell, H. E., & Owens, B. J. (2007). Aviation noise and
environmental justice: The barrio barrier. Journal of Regional Science, 47(1), 125-154.

Squires, G. D., & Kubrin, C. E. (2005). Privileged places: Race, uneven development,
and the geography of opportunity in urban America. Urban Studies, 42(1), 47-68.

Staeheli, L. A., & Mitchell, D. (2007). Locating the public in research and practice.
Progress in Human Geography, 31(6), 792-811.

Stallings, R. A. (1991). Media discourse and the social construction of risk. Social
Problems, 37(1), 80-94.

Stein, R. M. (1990). Urban alternatives: Public and private markets in the provision of
local services. Pittsburgh, PA: University of Pittsburgh Press.

Stone, C. (1989). Regime politics: Governing Atlanta, 1946-1988. Lawrence, KS:
University of Kansas Press.

Stretesky, P. B. (1998). Testing a broader model of environmental justice. Social
Pathology, 4(2), 73-86.

Stretesky, P. B., Johnston, J. E., & Arney, J. (2003). Environmental inequity: An analysis
of large-scale hog operations in 17 states, 1982-1997. Rural Sociology, 68(2), 231-252.

283
Stretesky, P., & Lynch, M. J. (1999). Environmental justice and the predictions of
distance to accidental chemical releases in Hillsborough County, Florida. Social Science
Quarterly, 80(4), 830-846.

Stretesky, P. B., & Lynch, M. J. (2002). Environmental hazards and school segregation in
Hillsborough County, Florida: 1987-1999. Sociological Quarterly, 43(4), 553-573.

Strong, D., & Hobbs, K. A. (2002). Administrative responses to environmental racism.
International Journal of Public Administration, 25(2 & 3), 391-417.

Stuart, S. M. (2005). Lifting spirits: Creating gardens in California domestic violence
shelters. In P.F. Barlett (Ed.), Urban place: Reconnecting with the natural world (pp. 61-
88). Cambridge, MA: The MIT Press.

Sui, D. (1999). GIS, environmental equity analysis, and the modifiable areal unit problem
(MAUP). In M. Craglia & H. Onsrud (Eds.), Geographic information research: Trans-
Atlantic perspectives (pp. 41-54). London: Taylor and Francis Ltd.

Swanstrom, T. (1988). Semi-sovereign cities: The politics of urban development. Polity,
21, 83-110.

Swanston, S. F. (1999). Environmental justice and environmental quality benefits: The
oldest, most pernicious struggle and hope for burdened communities. Vermont Law
Review, 23, 545-566.

Szazs, A., & Meuser, M. (2000). Unintended, inexorable: The production of
environmental inequalities in Santa Clara County, California. American Behavioral
Scientist, 43(4), 602-632.

Sze, J. (2002). From environmental justice literature to the literature of environmental
justice. In J. Adamson, M. M. Evans, & R. Stein (Eds.), The environmental justice
reader: Politics, poetics, and pedagogy (pp. 163-180). Tucson, AZ: The University of
Arizona Press.

Sze, J. (2007). Noxious New York: The racial politics of urban health and environmental
justice. Cambridge, MA: The MIT Press.

Talen, E., & Anselin, L. (1998). Assessing spatial equity: An evaluation of measures of
accessibility to public playgrounds. Environment and Planning A, 30, 595-613.

Taquino, M., Parisi, D., & Gill, D. A. (2002). Units of analysis and the environmental
justice hypothesis: The case of industrial hog farms. Social Science Quarterly, 83(1),
298-316.

284
Tarrant, M. A., & Cordel, H. K. (1999). Environmental justice and the spatial distribution
of outdoor recreation sites: An application of Geographic Information Systems. Journal
of Leisure Research, 31(1), 18-34.

Tarrow, S. (1998). Power in movement: Social movements and contentious politics (2
nd

ed.). New York: Cambridge University Press.

The City Parks Forum. (2002). How cities use parks for economic development. Chicago:
American Planning Association. Retrieved May 16, 2008, from
www.planning.org/cpf/pdf/economicdevelopment/pdf

The Trust for Public Land (2001). New York’s community gardens: A resource at risk.
San Francisco: Author.

The Trust for Public Land. (2008). Acres of parkland by city and agency. Retrieved June
12, 2009 from www.tpl.org/ccpe

Tiebout, C. M. (1956). A pure theory of local expenditures. Journal of Political
Economy, 64, 415-424.

Tiefenbacher, J. P., & Hagelman, R. R., III. (1999). Environmental equity in urban
Texas: Race, income, and patterns of acute and chronic toxic air releases in metropolitan
counties. Urban Geography, 19(6), 516-533.

Tilly, C. (1978). From mobilization to revolution. Reading, MA: Addison-Wesley.

Tilly, C. (1999). From interactions to outcomes in social movements. In M. Giugni, D.
McAdam, & C. Tilly (Eds.), How social movements matter (pp. 253-270). Minneapolis:
University of Minnesota Press.

Todd, H., & Zagrafos, C. (2005). Justice for the environment: Developing a set of
indicators of environmental justice for Scotland. Environmental Values, 14, 483-501.

Touché, G. E., & Rogers, G. O. (2005). Environmental equity and electric power
generation: Disparate community outcomes within Texas? Journal of Environmental
Planning and Management, 48(6), 891-915.

United Church of Christ (UCC). (1987). Toxic wastes and race in the United States: A
national report on the racial and socio-economic characteristics with hazardous waste
sites. New York: United Church of Christ Commission for Racial Justice.

Van Der Heijden, H. A. (1999). Environmental movements, ecological modernization,
and political opportunity structures: Environmental Politics, 8(1), 199-221.

285
Vrijheid, M. (2000). Health effects of residence near hazardous waste landfill sites: A
review of epidemiologic literature. Environmental Health Perspectives, 108(Supplement
1), 101-112.

Wagner, C. L., & Fernandez-Gimenez, M. E. (2008). Does community-based
collaborative resource management increase social capital? Society and Natural
Resources, 21, 324-344.

Warner, K. (2001). Managing to grow with environmental justice. Public Works
Management and Policy, 6(2), 126-138.

Waste, R. J. (1989). The ecology of city policymaking. NY: Oxford University Press.

Waste, R. J. (1998). Independent cities: Rethinking U.S. urban policy. Oxford, England:
Oxford University Press.

Westra, L. (2006). Environmental justice and the rights of unborn and future
generations: Law, environmental harm, and the right to health. London: Earthscan.

Wildavsky, A. (1995). But is it true? A citizen’s guide to environmental health and safety
issues. Cambridge, MA: Harvard University Press.

Wiland, H., & Bell, D. (2006). Edens lost and found: How ordinary citizens are restoring
our great cities. White River Junction, VT: Chelsea Green Publishing Company.

Wilson, S. M., Howell, F., Wing, S., & Sobsey, M. (2002). Environmental injustice and
the Mississippi hog industry. Environmental Health Perspectives, 110(Supplement 2),
195-201.

Wolch, J. Wilson, J. P., & Fehrenbach, J. (2005). Parks and park funding in Los Angeles:
an equity mapping analysis. Urban Geography, 26(1), 4-35.

Woolcock, M. (2001). The place of social capital in understanding social and economic
outcomes. The Canadian Journal of Policy Research, 2(1), 11-17.

Wu, S., Qiu, X, & Wang, L. (2005). Population estimation methods in GIS and remote
sensing: A review. GIScience and Remote Sensing, 42(1), 80-96.

Yamamoto, E. K., & Lyman, J. W. (2001). Racializing environmental justice. University
of Colorado Law Review, 72, 311-360.

Yandle, T., & Burton, D. (1996). Reexamining environmental justice: A statistical
analysis of historical hazardous waste landfill siting patterns in metropolitan Texas.
Social Science Quarterly, 77(3), 477-492.

286
Yin, R. K. (2003). Case study research: Design and methods (3
rd
edition). Thousand
Oaks, CA: SAGE Publications, Inc.

Young, M. K. (2002). Aligning San Jose’s community gardens with city plans and
policies: An Analysis of resource management and community participation.
Unpublished master’s thesis. San Jose State University.

Zimmerman, R. (1993). Sociological and environmental risk. Risk Analysis, 13(6), 649-
666.

Zimmerman, R. (1994). Issues of classification in environmental equity: How we manage
is how we measure. Fordham Urban Law Journal, 21, 633-669.

287
APPENDIX

This appendix highlights the procedures I used to create the city maps in GIS
(with ArcView 9.3). It also displays detailed results from the disamenities and amenities
analyses for each city. Results were summarized in Chapters 4 and 5, respectively, while
the research design was specified in Chapter 3. It also includes the detailed amenity and
disamenity rankings used in Chapter 6 for the medium and small cities.

Procedures for Creating a GIS Map

I created a GIS for each city in the study using Census data stored in Excel files
17
,
TIGER shapefiles, and USGS landcover data
18
. I downloaded the Census 2000 Place and
Census 2000 Census Tract shapefiles for the appropriate county from the Census website:
http://www.census.gov/geo/www/tiger/tgrshp2007/tgrshp2007.html. Place files can only
be downloaded at the state level, while tract files can only be downloaded at the county
level. Once downloaded, I projected these shapefiles with the Project tool in ArcCatalog,
using the USA Contiguous Albers Equal Area Conic projection coordinates. Once
projected, I added the Census 2000 Place shapefile into ArcMap and used the Select By
Attributes tool to select the appropriate city. Once selected, I exported the data into a new

17
I downloaded the relevant data from the SF1 and SF3 files into comma-delimited .csv files, which I then
sorted and coded in Excel.
18
I downloaded the NLCD 2001 Land Cover data directly from the USGS website
(http://seamless.usgs.gov) using the “View & Download United States Data” link. These data are in a .tif
raster file, which I converted to polygon data using the Raster to Polygon tool in ArcMap. I clipped the
polygon file with the CityBoundary file and selected the polygons with the gridcodes 21 (low-intensity
residential); 22 (high-intensity residential); and 23 (industrial/commercial/transportation). Once
highlighted, I exported only these values to a new landcover shapefile to use for geoprocessing.

288
shapefile called CityBoundary and added it as a layer to ArcMap. Following that, I added
the projected tract shapefile and an Excel file containing the Census sociodemographic
data to the map. Using the Join function, I attached the Excel file to the tract shapefile,
populating the tracts with sociodemographic data. I exported the populated shapefile to a
new shapefile. Using the Clip tool, I clipped the county shapefile to the CityBoundary
file, which deleted all of the Census tracts and associated data that were outside the city
boundary. This resulted in a Census tract shapefile and attribute table that stored
attendant the sociodemographic data.
Next, I added the disamenity and amenity locations. For each type of site (e.g.,
LQG, park), I added an Excel file, and used the Display XY Data function to display the
sites by their longitude (“X”) and latitude (“Y”) coordinates, according to the WGS 1984
geographic coordinate system. As stated in Chapter 3, these coordinates were verified
using Google Earth. GIS has a geocoding function, but I opted to do the process
manually, which some argue provides for more accurate coordinates than an automated
process (c.f., Maantay, 2007). Once the sites were displayed, I exported them to a
shapefile using the original data frame for the projection. I then used the Spatial Join tool
to join the site locations with the Census tracts in order to count the number of sites
within each tract and add them to the attribute table for the Census tract layer. Once I
added, displayed, and joined and counted each of the four types of sites, I saved the map.
The next step in the process was to load the shapefiles into a geodatabase (GDB)
for processing and analysis. In ArcCatalog, I loaded Census tract shapefile, the city
boundary, the four types of sites, and the landcover data into a geodatabase template.
Once I created the geodatabase, I returned to ArcMap to measure the area of each tract

289
and to process the landcover data. I began by adding the Census tract shapefile and used
the Field Calculator to calculate the size of each tract in hectares, which was used to
calculate the population density of each tract. I then added the landcover shapefile for
processing. I used the Union tool to connect the tract and landcover shapefiles. With this
union, for each Census tract I was able to calculate the percentage of each category of
landcover (e.g., low-intensity residential, high-intensity residential,
industrial/commercial/transportation) using the Select By Attributes tool. After dissolving
the unioned layer, I summed the landcover percentages for each tract and exported this to
a .dbf table. I added this table to ArcMap and joined it with the original Census tract
shapefile, which added the Landcover percentage fields to the attribute table. Following
that, I removed the join of the .dbf table and the Census tracts.

Geoprocessing and Spatial Statistics Analysis

I began my analysis by checking for the mean center for each type of disamenity
and amenity in each city. The tool, which geolocates a mean center based on the means of
all the XY coordinates in the feature class, is found in the Spatial Statistics Toolbox.
For the geostatistical analyses, I used tools in the Spatial Statistics Toolbox,
which can be run from ArcMap or ArcCatalog. I used the Spatial Autocorrelation
(Moran’s I) tool to see if the different types of sites were clustered, dispersed, or random.
Moran’s I is a global statistic that examines the entire dataset (or city, in my study), and
helps identify if spatial patterns exist and to what extent they are not due to chance. I also
used the Multi-distance Spatial Cluster Analysis (Ripley’s K) tool to determine the

290
distances at which sites are clustered or dispersed. I opted for 10 distance bands and 99
permutations (i.e., 100 Monte Carlo simulations resulting in a probability level of .01). I
also wanted to determine where local clusters existed, so I used the Hot Spot Analysis
[Getis-Ord Gi*] with Rendering tool. The tool identifies where the clusters exist and are
the strongest, as well as produces a Z score and a p-value to determine statistical
significance. The output from these analyses, however, appears in an output window and
must be added as additional elements to the map and copied and pasted into Microsoft
Word.
The following section presents geoprocessing and statistical results that were not
included in Chapters 4 and 5. Regression analysis and results will be discussed later in
this Appendix.

Spatial Statistics Results

What follows are detailed tables displaying the results from different statistical
analyses that were not included in Chapters 4 and 5. (See those chapters for an
explanation of how to interpret the results.) As such, where complete results for a
particular analysis were included in Chapters 4 and 5, they are not repeated here. This
section is organized by analytic test and appears in the same order as in the text.

291
Ripley’s K Results

Tables A.1 and A.2 display the dispersion values for TRI facilities and LQGs
respectively. See Chapter 4 for the clustered values for these hazards. Dispersion is
significant at the .01 level only when the Observed K is less than the Expected K and
lower than the Lower Confidence Level.

Table A.1
Ripley’s K Dispersion Analysis for TRI Sites

Expected K

Observed K

Differential K

Lower Confidence

Higher Confidence

Albany error
Austin 9154.914 6829.296 -2325.617 6346.604 9547.702
Boston 2176.378 1244.546 -931.8314 964.0213 2006.77
Boulder 1560.377 948.096 -612.2807 865.4893 1896.192
Charleston error
Dayton clustered
Decatur error
Flagstaff error
Grand Rapids clustered
Little Rock 3279.137 1830.314* -1448.823 2005.009 4331.314
Miami 8178.519 5299.34 -2879.179 5299.34 9178.727
Norfolk 2870.567 1744.386 -1123.181 1560.226 2702.391
Philadelphia 6746.783 6104.093 -642.6907 5604.931 7787.088
Salt Lake City 5201.333 4762.423 -438.9092 4116.285 5946.021
San Diego clustered
San Jose 5715.591 3443.884 -2271.708 3157.171 4034.404
Seattle 8686.604 6023.501 -2663.104 4676.609 5677.774
*p < .01

292

Table A.2
Ripley’s K Dispersion Analysis for LQGs

Expected K

Observed K

Differential K

Lower Confidence

Higher Confidence

Albany 2608.65 1850.544* -758.1058 2087.696 3253.358
Austin 7148.226 5745.015 -1403.211 5147.727 7486.425
Boston clustered
Boulder 265.3889 0 -265.3889 0 877.3909
Charleston 3039.051 646.5822 -2392.469 457.2027 1165.643
Dayton 6024.762 5648.113 -376.6492 4206.026 5849.895
Decatur 1345.936 468.577* -877.3585 662.6679 1554.094
Flagstaff error
Grand Rapids 5050.425 4347.054 -703.3706 3964.963 5226.197
Little Rock 3193.846 2790.223 -403.6228 1881.168 4206.419
Miami 9095.988 9565.931 505.843 7399.931 10043.5
Norfolk 2868.966 2281.798 -587.1672 1803.92 3783.934
Philadelphia 6894.46 5919.335 -975.125 5028.589 6512.229
Salt Lake City 4208.297 2796.222* -1412.075 2838.273 4187.266
San Diego clustered
San Jose clustered
Seattle 11536.61 9900.791 -1635.823 7274.719 8348.26
*p < .01

Tables A.3 and A.4 display the dispersion values for parks and gardens,
respectively. See Chapter 5 for the clustered values for these amenities.

Table A.3
Ripley’s K Dispersion Analysis for Parks

Expected K

Observed K

Differential K

Lower Confidence

Higher Confidence

Albany 1136.505 1061.96 -74.54497 902.9107 1183.008
Austin clustered
Boston clustered
Boulder clustered
Charleston 6031.392 5241.995 -789.3962 3887.568 5621.414
Dayton 3074.206 2808.525 -265.681 2475.423 3259.713
Decatur 2877.069 2658.071 -218.9979 2321.061 3052.445
Flagstaff 2754.597 1789.562 -965.0347 1777.592 2488.285
Grand Rapids clustered
Little Rock 532.12 516.6132 -15.50686 0 789.1396
Miami 3282.517 2981.362 -301.1545 2685.591 3380.547
Norfolk 31059.23 20010.96 -11048.27 12240.98 15528.18
Philadelphia clustered
Salt Lake City clustered
San Diego clustered
San Jose 20942.86 18632.18 -2310.678 11943.56 12935.33
Santa Fe 2430.866 1896.597 -534.2689 1662.55 2244.084
Seattle 6670.824 6645.597 -25.22716 5519.954 5886.892
*p < .01

293
Table A.4
Ripley’s K Dispersion Analysis for Community Gardens

Expected K

Observed K

Differential K

Lower Confidence

Higher Confidence

Albany 150.1822 0 -150.1822 0 436.4877
Austin 56134.914 20202.492 -35932.422 17427.926 25959.713
Boston clustered
Boulder 3377.414 1704.377 -1673.037 1704.377 3408.754
Charleston error
Dayton 2379.646 1892.445* -487.2013 1929.923 3074.859
Decatur 1394.988 594.075 -800.9133 0 1782.225
Flagstaff error
Grand Rapids 2277.228 751.4063 -1525.821 1227.041 1988.034
Little Rock 2754.597 1789.562 -965.0347 1777.992 2488.285
Miami 8709.936 8141.537 -568.3989 6628.838 9374.593
Norfolk 2564.808 887.1818* -1677.626 958.2658 1402.758
Philadelphia 3490.469 3308.108 -182.3608 2848.83 3225.769
Salt Lake City 2116.521 937.0813* -1414.609 1047.689 2095.377
San Diego 1428.236 0 -1428.236 0 3468.202
San Jose 20606.37 13139.91 -7466.464 8686.633 11195.37
Santa Fe 3236.992 577.9944 -2658.998 577.9944 913.8895
Seattle 5566.288 4686.984 -879.3032 4274.611 4996.529
*p < .01

Getis-Ord Gi* Results

Tables A.5 and A.6 display the significant Getis-Ord Gi* results for TRI and
LQG “hot spots.” These results are significant where Z ≥ 1.96. Due to the number of
significant results returned, only the results for Z ≥ 2.56 were included and discussed in
Chapter 4.

Table A.5
TRI “Hot Spots” or Getis-Ord Gi* Results

City

Tract

Gi* Z Score

Gi* p Value

Percent
Minority

Median
Income

Austin 18.36 2.32 0.021 38 68,813
23.12 2.28 0.023 91 28,725
24.17 2.72 0.006 68 41,369
23.14 2.14 0.033 56 23,786
18.29 2.02 0.044 23 45,821
22.05 3.28 0.001 79 37,111
18.28 2.14 0.033 23 70,551

294
Table A.5: Continued

18.51 2.32 0.021 42 63,925
23.13 2.02 0.044 52 30,509
22.02

2.32

0.021

91

32,139

Boulder 127.08 2.90 0.004 7 87,797
127.07 2.11 0.035 15 66,705
127.05 2.82 0.005 13 51,087
127.01

1.98

0.048

4 71,820
Dayton 25 2.01 0.045 9 30,154
1001.01 2.10 0.035 17 47,228
1003.02

2.68

0.007

10 44,720
Decatur 22

3.24

0.001

9 45,693
Grand Rapids 133 3.22 0.001 27 41,290
26 2.33 0.020 84 29,500
38 2.21 0.027 83 30,163
27 2.50 0.012 28 32,607
39

2.81

0.005

80 35,640
Little Rock 40.07

3.82

0.000

66 25,838
Norfolk 64 2.04 0.041 78 35,991
50 2.25 0.024 98 21,131
52 2.65 0.008 98 12,813
51 2.32 0.020 94 17,283
53

2.02

0.044

98 22,176
Philadelphia 43 3.05 0.002 0 0
44 2.00 0.046 42 28,024
49 2.32 0.021 0 0
68 3.07 0.002 0 0
26 2.28 0.023 0 0
34 2.11 0.035 77 20,500
36 2.11 0.035 64 13,860
360 5.16 0.000 19 45,539
361 3.32 0.001 19 51,513
184 2.10 0.035 2 33,065
324 3.05 0.002 0 41,250
325 2.00 0.046 6 32,255
327 2.57 0.010 61 22,857
328 3.75 0.000 86 56,250
329 2.10 0.035 20 33,958
354 4.02 0.000 38 0
355 2.34 0.020 10 38,438
359

3.25

0.001

13

43,582

Salt Lake City 1003.02 3.00 0.003 20 13,750
1139.01 2.79 0.005 21 37,348
1003.03

3.91

0.000

16 69,250
San Diego 83.33 3.19 0.001 27 127,271
83.58 2.98 0.003 57 49,976
83.41 2.26 0.024 31 51,783
83.49 2.30 0.021 54 58,550
83.59 2.52 0.012 53 36,359
95.02 3.53 0.000 24 58,869

295
Table A.5: Continued

83.4 2.48 0.013 26 64,554
85.11 2.69 0.007 32 34,191
83.5 2.26 0.024 60 60,828
83.51 2.68 0.007 58 60,223
83.57 2.15 0.032 63 54,201
94 5.14 0.000 40 38,796
95.06 2.92 0.004 21 63,953
83.6 2.99 0.003 54 52,867
83.48 2.26 0.024 55 66,442
83.46 2.68 0.007 44 99,718
83.56

2.15

0.032

46

55,802

San Jose 5050.05 5.71 0.000 51 71,667
5006 2.03 0.043 33 73,918
5045.05 3.56 0.000 72 79,259
5043.1 3.02 0.003 78 78,531
5043.18 2.50 0.012 62 52,065
5043.07 2.18 0.029 71 80,199
5014 2.12 0.034 80 47,454
5043.11 2.29 0.022 83 100,837
5010 2.12 0.034 76 28,250
5037.09 2.32 0.020 93 41,814
5046.02 2.38 0.017 80 57,589
5044.1 3.81 0.000 72 78,501
5001 2.43 0.015 82 48,276
5002 2.39 0.017 64 49,041
5043.21 2.02 0.044 79 74,063
5043.15 2.94 0.003 70 84,764
5043.16 1.96 0.049 81 86,626
5050.07 3.43 0.001 65 74,911
5003 2.62 0.009 68 45,057
5043.14 2.43 0.015 73 88,792
5044.11 3.43 0.001 65 85,949
5012 2.12 0.034 79 48,542
5004 2.42 0.016 42 58,354
5050.06 3.56 0.000 47 97,098
5051 2.53 0.011 64 59,211
5036.01 2.19 0.029 67 38,750
5043.2 3.21 0.001 72 98,875
5011

2.25

0.024

76

51,646

Seattle 109 3.77 0.000 40 33,654
100 3.28 0.001 71 37,122
112 4.30 0.000 61 30,917
99 2.37 0.018 27 46,684
103 2.18 0.029 70 39,554
110 3.28 0.001 86 36,754
113 4.02 0.000 43 46,838
104 5.07 0.000 77 48,697
108 5.45 0.000 49 53,198
265 4.55 0.000 60 16,285
111.01 2.18 0.029 82 40,293
264 4.55 0.000 41 40,291

296
Table A.6
LQG “Hot Spots” or Getis-Ord Gi* Results

City

Tract

Gi* Z Score

Gi* p Value

Percent
Minority

Median
Income

Albany

4.01

3.16

0.002

13

51,214
Austin 17.22 2.86 0.004 21 46,591
17.06 2.71 0.007 12 63,854
24.17 3.83 0.000 68 41,369
17.45 3.02 0.003 21 70,096
17.54 2.32 0.020 21 56,360
24.11 2.32 0.020 88 37,314
18.17 2.10 0.035 23 41,661
23.14 2.71 0.007 56 23,786
18.29 3.21 0.001 23 45,821
18.28 3.36 0.001 23 70,551
17.53 2.21 0.027 23 70,462
23.13 2.57 0.010 52 30,509
18.2 2.10 0.035 68 32,367
24.13

3.19

0.001

89

34,219

Boston 504 2.06 0.039 54 31,583
505 2.22 0.027 55 33,207
507 2.44 0.015 55 32,134
506 2.19 0.029 66 37,576
512 2.47 0.013 35 31,763
403 2.06 0.039 17 65,673
509 2.96 0.003 47 31,903
406 2.09 0.036 5 64,470
502 2.50 0.012 62 29,183
503 2.09 0.036 73 13,925
501 2.60 0.009 53 28,143
601 2.03 0.042 1 49,104
408 2.22 0.027 36 33,487
402

2.06

0.039

47

32,275

Boulder 127.07 3.38 0.001 15 66,705
122.02 1.97 0.049 9 28,326
122.03

2.85

0.004

27 35,853
Dayton 32 2.19 0.028 10 32,358
1003.02

2.25

0.024

10 44,720
Decatur 22 2.52 0.012 9 45,693
21

2.32

0.020

10 25,317
Grand Rapids 133 3.22 0.001 27 41,290
38 2.21 0.027 83 30,163
39

2.81

0.005

80 35,640
Little Rock 40.07

4.32

0.000

66 25,838
Miami 37.01 2.74 0.006 85 9,628
37.02

5.89

0.000

80

26,196

Philadelphia 2 2.39 0.017 0 0
49 2.17 0.030 0 0

297
Table A.6: Continued

50 2.73 0.006 0 0
68 2.88 0.004 41 75,487
76 2.29 0.022 51 21,131
87 2.08 0.037 41 9,320
88 2.18 0.029 16 42,346
8 2.17 0.030 15 42,000
12 2.17 0.030 76 24,508
13 1.98 0.047 57 25,978
33 2.18 0.029 19 45,539
360 2.23 0.026 6 27,689
295 2.39 0.017 14 31,167
296 2.25 0.024 11 23,214
297 2.10 0.036 20 27,234
298 2.00 0.045 57 21,317
299 2.12 0.034 56 14,793
178 2.25 0.024 1 30,866
180 2.26 0.024 8 18,571
181 3.30 0.001 7 26,331
182 3.88 0.000 1 32,820
183 3.30 0.001 7 41,875
185 3.90 0.000 1 28,531
186 3.08 0.002 14 30,950
187

2.54

0.011

0

0

Salt Lake City 1003.02 2.58 0.010 20 13,750
1139.01 4.34 0.000 21 37,348
1003.03

3.79

0.000

16 69,250
San Diego 83.33 4.95 0.000 27 127,271
83.41 3.71 0.000 31 51,783
83.43 3.20 0.001 43 34,967
83.31 2.43 0.015 19 88,217
95.02 2.35 0.019 24 58,869
83.4 4.04 0.000 26 64,554
83.39 2.95 0.003 37 42,717
83.5 2.11 0.034 60 60,828
83.51 2.11 0.034 58 60,223
94 2.93 0.003 40 38,796
83.12 2.02 0.043 13 87,742
83.46

4.19

0.000

44

99,718

San Jose 5050.05 7.27 0.000 51 71,667
5045.05 4.11 0.000 72 79,259
5043.1 3.77 0.000 78 78,531
5043.18 2.95 0.003 62 52,065
5043.07 2.20 0.028 71 80,199
5043.11 2.84 0.005 83 100,837
5010 2.02 0.043 76 28,250
5037.09 2.13 0.034 93 41,814
5046.02 3.84 0.000 80 57,589
5052.03 2.18 0.030 36 59,107
5044.1 4.39 0.000 72 78,501
5001 2.24 0.025 82 48,276
5002 2.32 0.020 64 49,041
5043.21 2.36 0.018 79 74,063
5043.15 3.26 0.001 70 84,764
5043.16 2.22 0.026 81 86,626
5050.07 4.49 0.000 65 74,911

298
Table A.6: Continued

5003 2.45 0.014 68 45,057
5043.14 2.68 0.007 73 88,792
5044.11 3.64 0.000 65 85,949
5004 2.08 0.037 42 58,354
5050.06 4.11 0.000 47 97,098
5051

3.20

0.001

64

59,211

Seattle 109 3.30 0.001 40 33,654
100 3.56 0.000 71 37,122
93 3.15 0.002 53 42,208
112 3.39 0.001 61 30,917
99 2.68 0.007 27 46,684
110 3.11 0.002 86 36,754
113 2.89 0.004 43 46,838
104 4.82 0.000 77 48,697
108 4.58 0.000 49 53198
265 3.69 0.000 60 16,285
111.01 1.98 0.048 82 40,293
264 3.69 0.000 41 40,291

Tables A.7 and A.8 display the significant Getis-Ord Gi* results for TRI and
LQG “cold spots.” These results are significant where Z ≥ -2.56.

Table A.7
TRI “Cold Spots or Getis-Ord Gi* Results

City

Tract

Gi* Z Score

Gi* p Value

Percent
Minority

Median
Income

Albany

17

-2.25

0.025

15

42981
16 -2.79 0.005 19 36371
15 -2.35 0.019 20 21250
5.02 -2.49 0.013 21 36563
5.01

-2.25

0.025

41 18953
Boston 1402.02 -2.09 0.036 33 50460
1105.02 -2.09 0.036 33 51964
1404 -2.19 0.028 83 42986
1304.01

-2.01

0.044

35 34346
Norfolk 21 -2.11 0.035 12 67308
58 -2.30 0.022 89 25994
26 -2.04 0.041 47 17017
57.01 -2.30 0.022 74 22902
20 -2.04 0.041 19 41522
31 -2.04 0.041 60 30259
28

-2.35

0.019

40 44912
Salt Lake City 1042 -2.33 0.020 6 82,608

299
Table A.7: Continued

1013 -2.30 0.022 5 124,259
1015 -3.65 0.000 14 26,717
1046 -2.91 0.004 14 38,000
1041 -2.96 0.003 5 70,039
1039 -2.47 0.013 7 52,244
1035 -3.42 0.001 11 49,213
1012 -2.46 0.014 7 39,012
1030 -4.21 0.000 36 32,067
1045 -2.37 0.018 4 51,615
1049 -2.58 0.010 23 33,056
1029 -3.24 0.001 42 24,375
1032 -3.42 0.001 25 27,222
1036 -3.24 0.001 3 68,929
1033 -3.24 0.001 14 38,487
1038 -2.77 0.006 6 51,141
1048 -2.00 0.045 8 40,710
1019 -3.01 0.003 20 25,938
1020 -3.91 0.000 42 24,385
1037 -3.07 0.002 5 61,463
1043 -2.17 0.030 9 41,063
1047 -2.10 0.036 7 47,454
1040 -2.58 0.010 6 57,007
1017 -3.43 0.001 21 27,568
1018 -3.91 0.000 24 33,481
1034 -3.42 0.001 10 42,286
1031 -3.42 0.001 29 37,925
1044 -2.02 0.044 2 80,344
1023 -3.42 0.001 39 16,048
1011 -2.29 0.022 12 30,424
1024 -2.87 0.004 51 23,125
1014 -2.58 0.010 21 22,778
1021 -2.60 0.009 25 16,978
1025 -2.03 0.042 32 22,786
1022 -2.60 0.009 12 21,131
1010 -1.98 0.047 7 52,621
1016 -3.91 0.000 14 31,420
1114 -2.47 0.013 27 30,470

300
Table A.8
LQG “Cold Spots” or Getis-Ord Gi* Results

City

Tract

Gi* Z Score

Gi* p Value

Percent
Minority

Median
Income

Albany 19.02 -2.22 0.027

10

53,299
17 -3.71 0.000 15 42,981
16 -3.36 0.001 19 36,371
3 -2.61 0.009 43 30,874
19.01 -2.03 0.042 10 52,093
6 -2.28 0.022 46 19,865
15 -3.14 0.002 20 21,250
5.02 -2.90 0.004 21 36,563
5.01 -2.53 0.011 41 18,953

Boston

1402.02 -2.54 0.011

33

50,460
1401.02 -2.00 0.045 37 49,444
1304.02 -2.25 0.025 7 54,125
1304.01 -2.44 0.015 35 34,346

Salt Lake City 1042 -2.13 0.033 6 82,608
1015 -3.53 0.000 14 26,717
1046 -2.57 0.010 14 38,000
1041 -2.42 0.016 5 70,039
1035 -3.29 0.001 11 49,213
1012 -2.28 0.022 7 39,012
1030 -4.10 0.000 36 32,067
1049 -2.78 0.005 23 33,056
1029 -3.73 0.000 42 24,375
1032 -3.08 0.002 25 27,222
1036 -2.89 0.004 3 68,929
1033 -2.89 0.004 14 38,487
1038 -2.42 0.016 6 51,141
1048 -2.30 0.022 8 40,710
1019 -3.08 0.002 20 25,938
1020 -3.79 0.000 42 24,385
1037 -2.72 0.006 5 61,463
1040 -2.04 0.042 6 57,007
1017 -3.53 0.000 21 27,568
1018 -3.79 0.000 24 33,481
1118 -2.12 0.034 15 38,424
1034 -3.08 0.002 10 42,286
1031 -3.29 0.001 29 37,925
1023 -3.29 0.001 39 16,048
1024 -2.72 0.006 51 23,125
1014 -2.41 0.016 21 22,778
1021 -2.22 0.026 25 16,978
1022 -2.22 0.026 12 21,131
1016 -3.79 0.000 14 31,420
1114 -2.66 0.008 27 30,470

Tables A.9 and A.10 display the significant Getis-Ord Gi* results for park and
garden “hot spots.” These results are significant where Z ≥ 1.96. Due to the number of

301
significant results returned, only the results for Z ≥ 2.56 were included and discussed in
Chapter 5.

Table A.9
Park “Hot Spots” or Getis-Ord Gi* Results

City

Tract

Gi* Z Score

Gi* p Value

Percent
Minority

Median
Income

Austin 8.01 3.70 0.000 92 35,478
19.11 2.36 0.018 22 40,432
9.02 3.76 0.000 94 23,700
11 4.43 0.000 39 45,063
12 4.47 0.000 26 41,366
16.05 4.18 0.000 11 50,375
13.08 2.42 0.015 67 32,299
14.02 4.68 0.000 25 47,550
13.03 3.73 0.000 24 45,515
8.02 3.58 0.000 96 12,427
4.01 3.34 0.001 29 32,719
8.03 4.47 0.000 88 25,703
10 4.64 0.000 85 23,597
23.04 3.96 0.000 74 27,551
14.03 4.18 0.000 37 40,491
16.03 2.58 0.010 5 84,338
7 4.98 0.000 28 27,768
23.14 2.82 0.005 56 23,786
6.03 2.59 0.010 21 9,580
21.11 2.62 0.009 92 26,463
8.04 4.98 0.000 93 17,725
14.01 4.64 0.000 17 40,095
9.01 4.80 0.000 94 31,538
13.05 4.47 0.000 52 31,700
23.15 4.15 0.000 62 22,363
23.13 2.14 0.032 52 30,509
23.07 1.99 0.046 59 29,283
13.04 3.39 0.001 22 40,871
23.16 4.45 0.000 65 21,411
4.02 2.23 0.026 65 25,829
6.01 4.11 0.000 38 13,750
16.02 3.25 0.001 38 31,299
5 2.41 0.016 20 22,209
6.04 3.55 0.000 27 7,423
8.01

3.70

0.000

92

35,478

Boulder 122.05 2.43 0.015 12 31,315
124.01 1.97 0.048 7 19,634
121.02

2.21

0.027

17

49,444

Charleston 7 2.64 0.008 48 20,868
8 2.65 0.008 30 22,206
21 2.64 0.008 12 55,573
6 2.95 0.003 15 35,077
1

2.21

0.027

24 16,650

Dayton 19 2.24 0.025 7 25,238

302
Table A.9: Continued

21 2.52 0.012 17 22,927
22 2.76 0.006 10 20,799
24 2.18 0.030 4 32,708
26 1.97 0.049 4 35,931
32 2.73 0.006 10 32,358
37 2.26 0.024 97 15,951
33 2.69 0.007 7 34,537
34 2.90 0.004 7 20,972
35 2.58 0.010 95 19,467
36 2.33 0.020 92 14,688
906 1.97 0.049 9 27,979
6 2.10 0.036 86 50,991
7 2.12 0.034 95 21,195
8.02 2.26 0.024 72 30,207
10 2.33 0.020 83 17,094
12 2.33 0.020 38 19,576
13 2.33 0.020 68 21,571
14 2.20 0.028 25 23,929
15 2.33 0.020 53 8,253
16

1.96

0.050

15

38,818

Grand Rapids 15 2.47 0.014 44 26,146
21 2.50 0.012 27 26,250
19 2.79 0.005 34 35,489
20

2.16

0.031

36 9,746
Little Rock 40.01 2.36 0.018 75 23,021
2 2.30 0.022 88 18,099
4

2.16

0.031

89 17,500
Norfolk 50

2.09

0.037

98 21,131
Miami 37.01 2.97 0.003 9,628 85
37.02

2.39

0.017

26,196

80

Philadelphia 42.01 3.06 0.002 21 32,564
60 2.33 0.020 43 31,250
61 2.11 0.035 46 33,462
63 2.58 0.010 82 21,320
1 3.35 0.001 19 48,886
5 2.41 0.016 46 9,620
6 2.79 0.005 30 41,563
9 3.03 0.002 30 20,725
10 4.25 0.000 9 72,625
11 3.67 0.000 20 36,564
12 1.97 0.049 15 42,000
14 3.54 0.000 67 26,897
15 3.67 0.000 35 38,026
16 4.07 0.000 17 50,598
17 4.51 0.000 21 48,889
18 3.81 0.000 31 36,458
19 2.67 0.008 94 21,766
22 3.18 0.001 86 30,156
23 4.27 0.000 31 29,806
24 3.54 0.000 41 34,247
25 4.58 0.000 50 26,250
26 4.22 0.000 0 0
27 5.01 0.000 45 23,750

303
Table A.9: Continued

28 4.06 0.000 41 22,759
30 3.42 0.001 83 20,294
29 3.79 0.000 11 26,744
41.01 2.62 0.009 63 20,759
40.01 2.49 0.013 7 25,854
214 2.16 0.030 12 40,746
215 2.44 0.014 6 45,464
366

4.31

0.000

13

87,027

Salt Lake City 1003.05 2.22 0.026 56 42,649
1006 2.09 0.037 55 30,250
1012 2.54 0.011 7 39,012
1003.06 2.10 0.036 57 35,044
1019 2.58 0.010 20 25,938
1017 2.15 0.032 21 27,568
1007 2.71 0.007 28 31,265
1011 2.44 0.015 12 30,424
1005 2.54 0.011 48 33,500
1008 2.25 0.024 11 28,125
1027 2.33 0.020 58 29,455
1021 2.00 0.045 25 16,978
1022 2.00 0.045 12 21,131
1010 2.00 0.045 7 52,621
1002 2.92 0.003 8 67,841
1004

2.54

0.011
33 39,512
San Diego 50 4.29 0.000 95 22,802
45.02 2.15 0.032 81 30,692
54 2.17 0.030 22 44,810
49 3.42 0.001 97 23,728
60 1.96 0.050 16 42,064
52 2.69 0.007 39 18,057
64 2.44 0.015 23 90,957
40 2.22 0.026 97 22,878
53 2.58 0.010 46 19,522
39.01 2.40 0.016 96 24,266
59 2.35 0.019 26 29,453
63 3.43 0.001 37 90,957
36.01 3.14 0.002 96 23,750
39.02 2.75 0.006 96 21,477
47 3.05 0.002 90 18,850
99.02 2.00 0.045 24 0
35.01 2.37 0.018 94 18,828
57 2.43 0.015 35 21,975
51 3.83 0.000 63 11,535
48 2.59 0.010 95 19,925
46 2.48 0.013 56 26,961
58 2.05 0.040 50 24,583
113 2.19 0.029 39 123,997
1 2.10 0.035 10 84,618
35.02

3.13

0.002

96

23,667

San Jose 5120.29 2.84 0.005 26 102,652
5032.14 1.97 0.049 37 79,664
5032.12 2.33 0.020 87 45,515
5031.16 2.09 0.037 85 84,622
5120.3 2.18 0.029 40 72,025
5120.31 2.25 0.025 29 100,734

304
Table A.9: Continued

5031.03 2.48 0.013 30 90,835
5032.13 2.19 0.029 67 43,636
5030.01 2.01 0.044 88 46,597
5120.23 2.34 0.019 15 79,920
5031.11 2.02 0.044 59 61,055
5120.26 2.89 0.004 89 63,578
5120.28

2.47

0.013

51

73,631

Santa Fe 11.07 2.06 0.039 60 45,388
12.02

2.57

0.010

81 27,438
Seattle 56 2.76 0.006 7 87,578
100 2.62 0.009 71 37,122
95 3.30 0.001 44 53,447
89 3.46 0.001 55 47,431
63 2.23 0.025 4 75,034
78 1.98 0.048 25 82,635

Table A.10
Garden “Hot Spots” or Getis-Ord Gi* Results

City

Tract

Gi* Z
Score

Gi* p
Value

Percent
Minority

Median
Income

Albany 2 2.14 0.033

83

16,222
25

1.97

0.048

75 16.158
Austin 8.01 4.10 0.000 92 35,478
9.02 3.41 0.001 94 23,700
21.09 2.79 0.005 93 30,234
11 3.97 0.000 39 45,063
14.02 2.21 0.027 25 47,550
23.11 2.79 0.005 67 17,321
8.02 3.51 0.000 96 12,427
4.01 3.97 0.000 29 32,719
8.03 3.15 0.002 88 25,703
10 3.07 0.002 85 23,597
23.04 2.35 0.019 74 27,551
14.03 2.43 0.015 37 40,491
7 4.95 0.000 28 27,768
23.14 3.01 0.003 56 23,786
6.03 2.28 0.023 21 9,580
21.11 4.58 0.000 92 26,463
8.04 2.99 0.003 93 17,725
14.01 2.07 0.038 17 40,095
9.01 2.99 0.003 94 31,538
13.05 2.14 0.032 52 31,700
23.15 2.35 0.019 62 22,363
21.1 3.77 0.000 97 27,344
23.16 2.60 0.009 65 21,411
2.04 2.28 0.023 14 36,484
4.02 4.16 0.000 65 25,829
6.01 5.28 0.000 38 13,750
6.04 2.28 0.023 27 7,423

305
Table A.10: Continued

Boston 104.02 2.13 0.033 31 13,780
6.02 2.83 0.005 44 20,175
101.02 2.29 0.022 26 16,447
304 2.25 0.025 4 50,051
612 2.71 0.007 7 37,188
203 2.91 0.004 26 52,160
3.01 2.04 0.042 17 56,321
1105.02 3.04 0.002 33 51,964
606 2.55 0.011 10 57,292
1401.04 3.12 0.002 56 27,030
105 2.33 0.020 25 33,333
804 2.02 0.043 92 34,297
805 2.02 0.043 95 14,417
607 2.74 0.006 57 16,500
5.02 3.31 0.001 23 38,958
603 2.69 0.007 2 49,825
1404 3.31 0.001 83 42,986
704 2.44 0.015 92 12,165
1303 2.31 0.021 7 65,156
302 2.40 0.016 4 43,125
605 2.61 0.009 3 48,475
702 2.27 0.023 76 15,374
921 1.98 0.048 57 40,133
101.01 2.26 0.024 24 34,655
912 2.23 0.026 63 36,929
102.01 2.13 0.033 29 28,932
1 2.18 0.029 26 44,630
1011.02 2.55 0.011 92 30,496
913 2.23 0.026 80 27,153
706 2.43 0.015 19 89,056
5.01 2.18 0.030 18 39,530
1104.02 3.02 0.003 38 50,410
911 2.62 0.009 46 39,222
8.01 3.00 0.003 32 30,968
1010.01 3.05 0.002 92 37,635
1106.02 2.27 0.023 11 59,524
108 2.33 0.020 11 71,671
1008 2.80 0.005 36 42,392
611 2.71 0.007 59 12,132
6.01 2.40 0.016 26 43,804
1003 2.25 0.025 92 34,167
301 2.10 0.036 3 52,691
608 2.82 0.005 5 41,976
909 2.78 0.005 65 32,731
708 2.23 0.026 39 45,486
1106.01 2.37 0.018 8 65,764
709 2.23 0.026 54 39,969
7.02 2.60 0.009 35 27,675
305 2.17 0.030 6 63,488
707 2.34 0.019 48 58,843
3.02 2.31 0.021 17 64,750
602 2.78 0.005 0 55,952
104.01 2.23 0.026 27 26,576
1010.02 2.96 0.003 89 36,742
7.01 2.74 0.006 31 36,875
914 2.13 0.033 84 27,083
201 2.64 0.008 6 81,804
4.02 3.13 0.002 21 54,833

306
Table A.10: Continued

703 2.19 0.029 23 62,878
610 2.74 0.006 43 13,973
1002 2.14 0.032 95 35,825
2.01 2.01 0.044 19 62,305
1007 2.89 0.004 6 45,566
103 2.02 0.043 35 14,224
1401.03 3.09 0.002 46 51,071
801 2.13 0.033 79 27,930
2.02 2.18 0.029 33 56,803
1403 3.02 0.003 65 41,209
1006.01 2.01 0.044 45 41,870
1302 2.49 0.013 5 68,143
8.02 2.79 0.005 42 29,555
1304.01 2.43 0.015 35 34,346
711 2.23 0.026 51 32,303
1006.02 2.40 0.016 17 41,848
712 2.43 0.015 62 20,806
907 2.52 0.012 35 35,805
303 3.22 0.001 12 70,854
1009 3.07 0.002 71 47,109
1004 2.34 0.019 81 41,268
910 2.37 0.018 33 44,957
202 2.82 0.005 13 53,470
701 2.55 0.011 45 17,639
1011.01 2.43 0.015 95 35,867
1105.01 2.54 0.011 15 55,754
705 2.43 0.015 50 43,636
106 2.34 0.019 20 61,458
1104.01 2.05 0.041 57 45,938
604 2.83 0.005 3 43,015
1103 1.98 0.048 45 51,667
102.02 2.13 0.033 30 18,060
4.01 2.41 0.016 22 30,714
107 2.23 0.026 10 77,282
1102 2.27 0.023 59 32,891
915

2.23

0.026

80

32,004

Boulder 121.06

2.14

0.032

13 70,583
Dayton 19 2.22 0.027 7 25,238
37 2.17 0.030 97 15,951
38 2.22 0.027 96 23,604
39 2.32 0.020 98 25,833
40 2.53 0.011 95 10,563
41 2.46 0.014 98 14,921
34 2.11 0.035 7 20,972
35 2.07 0.038 95 19,467
42 2.43 0.015 97 23,274
43 2.32 0.020 94 19,650
44 2.36 0.019 96 21,821
806 2.32 0.020 1 30,551
7 2.07 0.038 95 21,195
16

2.01

0.045

15

38,818

Decatur 6 2.21 0.027 56 18,269
5 2.07 0.038 35 15,474
3 2.07 0.038 50 22,776
16 2.07 0.038 16 41,136

307
Table A.10: Continued

4

2.07

0.038

60 24,962
Flagstaff 2 2.11 0.035 19 43,041
1

2.47

0.014

15 57,760
Grand Rapids 9 2.84 0.005 15 36,783
15

2.50

0.012

44 26,146
Little Rock 12 2.25 0.024 83 24,014
11 2.25 0.024 97 26,512
6 2.20 0.028 56 29,261
7 2.38 0.017 95 21,552
1 2.06 0.040 49 10,556
13 2.38 0.017 94 20,528
10 2.38 0.017 94 17,639
3 2.36 0.018 62 16,164
15

2.20

0.028

12

36,706

Norfolk 48 2.23 0.025 94 10,746
38 2.23 0.025 12 48,750
39 2.71 0.007 14 50,398
40.02 2.16 0.030 11 36,310
25 2.16 0.030 69 20,284
49 2.16 0.030 48 46,081
42

2.03

0.043

98

6,913

Miami 52.02 2.56 0.010 94 19,271
37.02 3.20 0.001 85 9,628
29 2.77 0.006 93 24183
34 2.42 0.016 97 11,006
30.04 2.58 0.010 89 12,967
31 2.11 0.035 95 11,334
27.02 3.65 0.000 88 22,917
26 1.98 0.048 94 14,148
36.01 2.58 0.010 96 7,595
30.01 2.26 0.024 82 16,031
28

2.72

0.007

95

9,671

Philadelphia 80 3.46 0.001 97 28,366
106 2.59 0.010 96 15,952
107 2.61 0.009 97 15,417
71 2.11 0.035 96 26,790
174 2.92 0.003 97 15,050
72 1.97 0.049 96 26,955
86 4.31 0.000 83 19,612
85 3.34 0.001 97 22,077
21 2.23 0.026 99 21,875
69 1.97 0.049 95 13,569
70 1.98 0.047 98 21,821
73 2.59 0.010 97 25,778
74 2.93 0.003 96 17,604
75 2.97 0.003 0 0
76 5.13 0.000 41 75,487
77 3.98 0.000 66 14,628
78 3.52 0.000 67 28,885
79 4.31 0.000 70 32,380
176.01 3.04 0.002 95 11,909
87 4.37 0.000 51 21,131

308
Table A.10: Continued

88 4.74 0.000 41 9,320
89 3.38 0.001 50 6,311
90 2.15 0.031 32 13,792
91 3.22 0.001 68 17,500
92 4.37 0.000 95 20,083
93 2.50 0.012 98 21,503
7 2.11 0.035 22 34,536
8 2.92 0.003 16 42,346
9 2.23 0.026 30 20,725
11 2.53 0.011 20 36,564
12 2.52 0.012 15 42,000
14 2.25 0.024 67 26,897
15 2.73 0.006 35 38,026
18 2.83 0.005 31 36,458
19 2.75 0.006 94 21,766
20 2.15 0.031 98 19,240
22 2.58 0.010 86 30,156
23 2.39 0.017 31 29,806
13 2.92 0.003 76 24,508
24 2.45 0.014 41 34,247
28 2.34 0.019 41 22,759
30 2.17 0.030 83 20,294
33 2.25 0.024 57 25,978
34 3.08 0.002 77 20,500
29 2.17 0.030 11 26,744
108 2.61 0.009 98 17,310
176.02 2.95 0.003 92 12,076
141 4.03 0.000 92 12,165
142 2.23 0.026 52 30,862
143 2.84 0.004 11 40,293
144 4.68 0.000 69 23,720
145 4.11 0.000 99 12,092
146 4.43 0.000 78 26,295
153 2.82 0.005 90 14,826
154 3.53 0.000 64 31,923
155 4.26 0.000 98 12,333
156 5.09 0.000 93 14,524
157 5.83 0.000 69 22,478
158 5.74 0.000 8 31,862
159 2.93 0.003 5 28,871
160 5.45 0.000 6 29,621
161 4.44 0.000 44 19,098
162 4.59 0.000 96 13,833
163 4.29 0.000 94 15,865
164 3.88 0.000 98 12,314
165 3.85 0.000 98 16,964
166 2.82 0.005 96 14,250
167 2.43 0.015 98 16,367
175 2.82 0.005 97 14,294
177 2.97 0.003 81 15,687
178

3.27

0.001

56

14,793

Salt Lake City 1001 1.97 0.049 44 28,867
1003.06 2.25 0.024 57 35,044
1005

1.97

0.049

48 33,500
San Diego 27.06 2.01 0.044 69 29,313
19 2.62 0.009 22 47,866

309
Table A.10: Continued

75.02 2.80 0.005 10 34,934
23.02 2.49 0.013 88 20,473
25.02 2.50 0.012 68 37,744
18 2.57 0.010 57 30,738
25.01 2.65 0.008 84 25,963
16 2.89 0.004 72 26,216
75.01 2.37 0.018 16 33,868
22.01 2.36 0.018 88 20,697
20.02 2.01 0.044 19 60,000
21 2.44 0.015 57 29,234
26.01 2.24 0.025 88 24,849
24.01 2.98 0.003 81 24,274
6 2.40 0.016 21 36,088
27.05 2.36 0.018 70 41,429
7 2.32 0.020 18 35,129
28.03 2.06 0.040 43 28,803
24.02 2.20 0.028 88 19,205
22.02 2.44 0.015 89 18,389
15 2.73 0.006 44 31,543
28.04 2.45 0.014 23 44,683
27.02 2.34 0.019 50 33,553
17 3.11 0.002 56 26,761
74 2.19 0.028 15 42,042
26.02 2.08 0.037 87 24,447
27.08

2.10

0.035

85

17,171

San Jose 5019 2.60 0.009 62 36,332
5006 2.88 0.004 33 73,918
5008 2.42 0.016 70 36,364
5015.01 2.00 0.045 90 42,136
5017 2.17 0.030 89 43,614
5014 2.29 0.022 80 47,454
5015.02 1.96 0.050 83 44,489
5023 2.09 0.037 26 66,534
5009.01 2.29 0.022 63 37,829
5031.12 2.00 0.045 78 42,656
5013 2.17 0.030 48 61,739
5010 2.29 0.022 76 28,250
5052.03 2.03 0.042 36 59,107
5009.02 2.25 0.025 71 33,750
5002 2.18 0.030 64 49,041
5020.02 2.07 0.038 71 40,025
5003 2.45 0.014 68 45,057
5016 2.00 0.045 74 29,781
5012 2.29 0.022 79 48,542
5004 2.42 0.015 42 58,354
5011

2.00

0.045

76

51,646

Seattle 100 2.44 0.015 71 37,122
95 3.91 0.000 44 53,447
103 2.97 0.003 70 39,544
89 2.79 0.005 55 47,431
101 3.21 0.001 60 47,926

310
Tables A.11 and A.12 display the significant Getis-Ord Gi* results for park and
garden “cold spots.” These results are significant where Z ≥ -2.56. These expanded
results were not included in Chapter 5.

Table A.11
Park “Cold Spots” or Getis-Ord Gi* Results

City

Tract

Gi* Z Score

Gi* p Value

Percent
Minority

Median
Income

Albany

19.02

-2.09

0.036

10

53,299

Boston 812

-2.29

0.022

88 18,967

Boulder 132.05 -2.15 0.032 5 83,952
127.08 -2.26 0.024 7 87,797
127.09

-2.15

0.032

5 82,705
Norfolk 59.01

-2.07

0.038

78 26,427
Philadelphia 137

-2.00

0.045

93 23,806
Salt Lake City 1045 -2.65 0.008 4 51,615
San Jose 5079.04 -2.02 0.044 47 114,506
5063.05 -2.10 0.035 56 51,103
5066.06 -1.98 0.048 37 64,643
5062.04 -2.28 0.023 43 72,978
5074.02 -2.21 0.027 27 116,255
5062.03 -2.28 0.023 63 62,207
5066.04 -2.37 0.018 28 77,041
5066.05

-2.34

0.019

41 64,191
Seattle 46 -2.49 0.013 9 62,159
39 -2.42 0.015 13 58,114
21 -2.65 0.008 16 50,284
12 -2.64 0.008 30 32,463
13 -2.53 0.011 30 32,983
36 -2.96 0.003 13 47,547
17 -2.04 0.041 24 42,436
18 -2.54 0.011 24 41,941
22 -2.52 0.012 16 68,450
20 -3.20 0.001 14 50,351
34 -2.03 0.042 10 55,885
35 -2.06 0.040 8 56,000
38 -2.36 0.018 12 59,432
26 -2.75 0.006 11 66,066
7 -2.01 0.044 29 36,080
9 -2.01 0.045 11 81,118
11 -2.76 0.006 21 54,776
24 -2.86 0.004 13 62,784
28 -2.08 0.038 10 58,446
27 -2.94 0.003 10 63,952
25 -2.89 0.004 13 57,778
10 -2.52 0.012 21 50,536
19 -2.60 0.009 17 51,760

311
Table A.12
Garden “Cold Spots” or Getis-Ord Gi* Results

City

Tract

Gi* Z Score

Gi* p Value

Percent
Minority

Median
Income

Albany 18.02

-2.25

0.024

10

41,815

Norfolk 59.03 -2.16 0.030 49 30,335
55 -2.33 0.020 37 29,696
7

-2.01

0.044

20 42,601

The following section discusses the regression analysis and results for each city.

Regression Analysis

As explained in Chapter 3, four OLS regression models were tested for each
dependent variable in the six largest cities (though some cities did not have enough
dependent variables to support the analysis). The same OLS models were run on TRI
facilities and LQG facilities in the medium-sized cities. The OLS tool may be used with
the Spatial Analyst extension when using ArcMap (with an ArcView 9.3 license). It is in
the Spatial Statistics Toolbox. I specified the feature class from which data was drawn
and a Unique_ID assigned to each feature (i.e., Census tract). I then specified the
dependent variables, explanatory variables, and added optional output in .dbf format. The
results of the OLS appeared in a dialogue box, but those results were only temporary,
hence the need for capturing the results in .dbf format. Once the tool ran, a map of the
residuals was returned as a layer in ArcMap with attribute values on the regression
residuals. I ran Moran’s I on the regression residuals to test for spatial autocorrelation.

312
The OLS summary results include summary statistics on all the explanatory
variables, as well as diagnostic statistics to help with model interpretation. Some of those
statistics were explained in Chapters 4 and 5. Those statistics also include: the Joint F-
Statistic to gauge model significance; the Joint Wald Statistic, which indicates robust
model significance; the Koenker (BP) Statistic, which—when significant—indicates
when biased standard errors exist and when robust probability estimates should be used;
and the Jarque-Bera Statistic, which—when significant—indicates that the residuals are
not normally distributed.
GWR was run when possible on each dependent variable in the six largest cities.
The GWR tool (in the Spatial Statistics Toolbox) is available with the Geostatistical
Analyst extension when using ArcView 9.3. It is run from ArcMap, where the feature
class, dependent variable, and independent variables are specified. Several options are
available, but if none are specified, ArcGIS selects the most appropriate parameters (e.g.,
weights, distances) for the regression, which was the option I selected. To return an
individual raster layer for each coefficient (and its attendant residuals), an optional raster
workspace was specified. That raster layer shows where the coefficient was operating
most strongly within the geographic regression and also reports statistics on each
coefficient. These statistics include (1) a condition number, which indicates local
collinearity when larger than 30; (2) a local R
2
, where a higher values is preferable; and
(3) a coefficient standard error, where larger numbers indicate local collinearity. To
interpret the model, the AIC and Adjusted R
2
statistics in the summary report should be
considered. Sigma is a measure of standard deviation for the residuals, and a small
number is the goal. The other model diagnostics are indicative of the model parameters

313
used. One can compare the AIC number from the OLS results to the AIC number from the
GWR results, but no comparison of the Adjusted R
2
values from each type of regression
should be done.

Regression Results

Tables A.13-A.16 present the results for the GWR regressions on TRI data for
four of the six largest cities. In the case of Boston, no GWR was run due to the limited
number of dependent variables, and the GWR did not run successfully in San Jose. Due
to the large number of statistics returned in GWR (i.e., a full set of coefficient statistics
for each variable in each Census tract), only summary statistics on GWR model
performance are included here.

Table A.13
TRI/GWR Results for Austin
Bandwidth : 19160.08814
ResidualSquares : 14.95952748
EffectiveNumber : 13.32702902
Sigma : 0.296057838
AICc : 85.60716539
R2 : 0.158497996
R2Adjusted : 0.097719657

Table A.14
TRI/GWR Results for Philadelphia
Bandwidth : 3045.518557
ResidualSquares : 254603.0371
EffectiveNumber : 73.98312139
Sigma : 28.79722661
AICc : 3691.235613
R2 : 0.749457272
R2Adjusted : 0.689899015

314

Table A.15
TRI/GWR Results for San Diego
Bandwidth : 19160.08814
ResidualSquares : 14.95952748
EffectiveNumber : 13.32702902
Sigma : 0.296057838
AICc : 85.60716539
R2 : 0.158497996
R2Adjusted : 0.097719657

Table A.16
TRI/GWR Results for Seattle
Bandwidth : 6428.915821
ResidualSquares : 40.66482159
EffectiveNumber : 25.14643766
Sigma : 0.634985509
AICc : 263.0831751
R2 : 0.595917388
R2Adjusted : 0.499171617

The following Tables A.17-A.21 present the data for the best-fitting OLS
regressions on TRI data for five of the six medium-sized cities. In the case of Miami, no
TRI sites fell within Census tracts, so no regression was done.

Table A.17
TRI/OLS Regression Results for Dayton Model I
Variable Coefficient StdError t-Statistic Probability Robust_SE Robust_t Robust_Pr VIF [1]
Intercept 0.258023 0.441099 0.584954 0.560636 0.259097 0.995855 0.323067 --------
PCT_MIN -0.002403 0.002581 -0.930901 0.355399 0.002274 -1.056583 0.294671 1.207741
MED_INC_99 -0.000025 0.000009 -2.819682 0.006391* 0.00001 -2.412835 0.018701* 1.311653
PCT_MFG 0.047926 0.019758 2.425654 0.018108* 0.022107 2.167849 0.033894* 1.133533
POP_DENSITY -0.000045 0.000069 -0.648934 0.518702 0.00002 -2.182335 0.032760* 1.052645

Number of Observations: 69 Number of Variables: 5
Degrees of Freedom: 64 Akaike's Information Criterion (AIC) [2]: 150.927205
Multiple R-Squared [2]: 0.153936 Adjusted R-Squared [2]: 0.101057
Joint F-Statistic [3]: 2.911097 Prob(>F), (4,64) degrees of freedom: 0.028162*
Joint Wald Statistic [4]: 8.155386 Prob(>chi-squared), (4) degrees of freedom: 0.086049
Koenker (BP) Statistic [5]: 6.6098 Prob(>chi-squared), (4) degrees of freedom: 0.158002
Jarque-Bera Statistic [6]: 328.466067 Prob(>chi-squared), (2) degrees of freedom: 0.000000*

315
Table A.18
TRI/OLS Regression Results for Grand Rapids Model II
Variable Coefficient StdError t-Statistic Probability Robust_SE Robust_t Robust_Pr VIF [1]
Intercept 0.769007 0.538616 1.427747 0.159709 0.519937 1.479039 0.145532 --------
PCT_MIN 0.001196 0.004581 0.261125 0.795092 0.004235 0.282472 0.778774 2.49378
MED_INC_99 -0.000012 0.000009 -1.29483 0.201448 0.000008 -1.43213 0.158457 1.780181
PCT_MFG 0.027877 0.013332 2.090905 0.041745* 0.013249 2.104017 0.040532* 1.714466
LOG_DENSITY -0.206026 0.08392 -2.455042 0.017683* 0.061051 -3.374672 0.001454* 1.075213

Number of Observations: 54 Number of Variables: 5
Degrees of Freedom: 49 Akaike's Information Criterion (AIC) [2]: 101.745002
Multiple R-Squared [2]: 0.279612 Adjusted R-Squared [2]: 0.220805
Joint F-Statistic [3]: 4.754725 Prob(>F), (4,49) degrees of freedom: 0.002543*
Joint Wald Statistic [4]: 18.079659 Prob(>chi-squared), (4) degrees of freedom: 0.001191*
Koenker (BP) Statistic [5]: 17.399985 Prob(>chi-squared), (4) degrees of freedom: 0.001616*
Jarque-Bera Statistic [6]: 7.118195 Prob(>chi-squared), (2) degrees of freedom: 0.028465*

Table A.19
TRI/OLS Regression Results for Little Rock Model I
Variable Coefficient StdError t-Statistic Probability Robust_SE Robust_t Robust_Pr VIF [1]
Intercept -0.159206 0.380363 -0.418564 0.677668 0.360357 -0.441802 0.6609 --------
PCT_MIN 0.000446 0.003089 0.144463 0.885827 0.0028 0.159377 0.874137 2.341584
MED_INC_99 -0.000001 0.000005 -0.136806 0.891839 0.000004 -0.190045 0.850191 2.260059
PCT_MFG 0.053172 0.015093 3.523022 0.001044* 0.018882 2.81602 0.007376* 1.260743
POP_DENSITY -0.013004 0.010557 -1.231865 0.224852 0.007194 -1.807641 0.077829 1.047709

Number of Observations: 47 Number of Variables: 5
Degrees of Freedom: 42 Akaike's Information Criterion (AIC) [2]: 57.722148
Multiple R-Squared [2]: 0.326383 Adjusted R-Squared [2]: 0.262229
Joint F-Statistic [3]: 5.087499 Prob(>F), (4,42) degrees of freedom: 0.001958*
Joint Wald Statistic [4]: 12.241598 Prob(>chi-squared), (4) degrees of freedom: 0.015642*
Koenker (BP) Statistic [5]: 5.871813 Prob(>chi-squared), (4) degrees of freedom: 0.208928
Jarque-Bera Statistic [6]: 29.920914 Prob(>chi-squared), (2) degrees of freedom: 0.000000*

316
Table A.20
TRI/OLS Regression Results for Norfolk Model I
Variable Coefficient StdError t-Statistic Probability Robust_SE Robust_t Robust_Pr VIF [1]
Intercept 0.277537 0.151938 1.826649 0.071486 0.209672 1.323674 0.189385 --------
PCT_MIN 0.001613 0.001227 1.314416 0.192465 0.001654 0.975238 0.332376 1.609195
MED_INC_99 -0.000004 0.000003 -1.599135 0.113739 0.000003 -1.456273 0.149237 1.621949
PCT_MFG 0.005949 0.00996 0.597298 0.551996 0.011659 0.510255 0.611282 1.023673
POP_DENSITY -0.009586 0.002774 -3.456166 0.000885* 0.004327 -2.215448 0.029568* 1.089484

Number of Observations: 85 Number of Variables: 5
Degrees of Freedom: 80 Akaike's Information Criterion (AIC) [2]: 28.892601
Multiple R-Squared [2]: 0.179636 Adjusted R-Squared [2]: 0.138618
Joint F-Statistic [3]: 4.379428 Prob(>F), (4,80) degrees of freedom: 0.002974*
Joint Wald Statistic [4]: 5.719022 Prob(>chi-squared), (4) degrees of freedom: 0.221138
Koenker (BP) Statistic [5]: 12.119192 Prob(>chi-squared), (4) degrees of freedom: 0.016487*
Jarque-Bera Statistic [6]: 1008.893643 Prob(>chi-squared), (2) degrees of freedom: 0.000000*

Table A.21
TRI/OLS Regression Results for Salt Lake City Model II
Variable Coefficient StdError t-Statistic Probability Robust_SE Robust_t Robust_Pr VIF [1]
Intercept 1.198388 0.736586 1.626949 0.109573 1.169567 1.024643 0.310098 --------
PCT_MIN -0.002951 0.021922 -0.1346 0.893428 0.025536 -0.115548 0.908439 3.416342
MED_INC_99 -0.000014 0.000012 -1.096213 0.277849 0.00001 -1.392927 0.169352 1.642692
PCT_MFG -0.009456 0.066741 -0.141685 0.887856 0.132995 -0.071102 0.943578 2.549057
POP_DENSITY -0.000238 0.000456 -0.52209 0.603744 0.000144 -1.655908 0.103544 1.037205

Number of Observations: 59 Number of Variables: 5
Degrees of Freedom: 54 Akaike's Information Criterion (AIC) [2]: 228.257344
Multiple R-Squared [2]: 0.034042 Adjusted R-Squared [2]: -0.03751
Joint F-Statistic [3]: 0.475768 Prob(>F), (4,54) degrees of freedom: 0.753313
Joint Wald Statistic [4]: 6.158864 Prob(>chi-squared), (4) degrees of freedom: 0.187594
Koenker (BP) Statistic [5]: 4.933971 Prob(>chi-squared), (4) degrees of freedom: 0.29414
Jarque-Bera Statistic [6]: 2258.414498 Prob(>chi-squared), (2) degrees of freedom: 0.000000*

Tables A.22-A.26 present the data for the geographically weighted regressions on
LQG data for five of the six largest cities. The San Jose GWR was aborted due to an error
in the model.

317
Table A.22
LQG/GWR Results for Austin
Bandwidth : 20602.74241
ResidualSquares : 24.92115883
EffectiveNumber : 12.5130404
Sigma : 0.381213758
AICc : 178.313032
R2 : 0.104395855
R2Adjusted : 0.044268095

Table A.23
LQG/GWR Results for Boston
Bandwidth : 292233.2452
ResidualSquares : 67.34780848
EffectiveNumber : 7.005971408
Sigma : 0.670077059
AICc : 329.6454069
R2 : 0.086118761
R2Adjusted : 0.049525674

Table A.24
LQG/GWR Results for Philadelphia
Bandwidth : 393267.6229
ResidualSquares : 44.49737196
EffectiveNumber : 7.010856486
Sigma : 0.344935333
AICc : 279.4819404
R2 : 0.092036273
R2Adjusted : 0.07744323

Table A.25
LQG/GWR Results for San Diego
Bandwidth : 20602.74241
ResidualSquares : 24.92115883
EffectiveNumber : 12.5130404
Sigma : 0.381213758
AICc : 178.313032
R2 : 0.104395855
R2Adjusted : 0.044268095

318
Table A.26
LQG/GWR Results for Seattle
Bandwidth : 6428.915821
ResidualSquares : 116.7950223
EffectiveNumber : 25.14643766
Sigma : 1.076134477
AICc : 396.0203657
R2 : 0.559975696
R2Adjusted : 0.454624738

The following tables (A.27-A.32) present the data for the most significant OLS
regressions on LQG data for the six medium-sized cities.

Table A.27
LQG/OLS Regression Results for Dayton Model II
Variable Coefficient StdError t-Statistic Probability Robust_SE Robust_t Robust_Pr VIF [1]
Intercept 1.309749 0.617277 2.121817 0.037729* 0.552798 2.369309 0.020846* --------
PCT_MIN -0.002302 0.002621 -0.878416 0.383 0.002749 -0.837513 0.405417 1.22636
PCT_MFG 0.047551 0.02011 2.364574 0.021092* 0.020159 2.358733 0.021400* 1.157018
LOG_DENSITY -0.108227 0.054836 -1.973675 0.05274 0.039619 -2.73172 0.008132* 1.04312
SQ_INCOME -0.008505 0.003238 -2.626846 0.010770* 0.003505 -2.426904 0.018051* 1.399411

Number of Observations: 69 Number of Variables: 5
Degrees of Freedom: 64 Akaike's Information Criterion (AIC) [2]: 151.947746
Multiple R-Squared [2]: 0.185577 Adjusted R-Squared [2]: 0.134676
Joint F-Statistic [3]: 3.64582 Prob(>F), (4,64) degrees of freedom: 0.009738*
Joint Wald Statistic [4]: 10.481627 Prob(>chi-squared), (4) degrees of freedom: 0.033051*
Koenker (BP) Statistic [5]: 8.347361 Prob(>chi-squared), (4) degrees of freedom: 0.079651
Jarque-Bera Statistic [6]: 276.754144 Prob(>chi-squared), (2) degrees of freedom: 0.000000*

319
Table A.28
LQG/OLS Regression Results for Grand Rapids Model II
Variable Coefficient StdError t-Statistic Probability Robust_SE Robust_t Robust_Pr VIF [1]
Intercept 0.513708 0.563229 0.912077 0.366192 0.461868 1.11224 0.271461 --------
PCT_MIN 0.006479 0.00479 1.352668 0.182372 0.004425 1.46414 0.149544 2.49378
MED_INC_99-0.000002 0.00001 -0.21068 0.834012 0.000007 -0.284528 0.777207 1.780181
PCT_MFG 0.014374 0.013942 1.031046 0.30758 0.010867 1.322773 0.192052 1.714466
LOG_DENSITY -0.202954 0.087754 -2.312755 0.024977* 0.079633 -2.548607 0.013999* 1.075213

Number of Observations: 54 Number of Variables: 5
Degrees of Freedom: 49 Akaike's Information Criterion (AIC) [2]: 106.570786
Multiple R-Squared [2]: 0.212269 Adjusted R-Squared [2]: 0.147965
Joint F-Statistic [3]: 3.300999 Prob(>F), (4,49) degrees of freedom: 0.017935*
Joint Wald Statistic [4]: 9.699289 Prob(>chi-squared), (4) degrees of freedom: 0.045810*
Koenker (BP) Statistic [5]: 6.937014 Prob(>chi-squared), (4) degrees of freedom: 0.139254
Jarque-Bera Statistic [6]: 33.852508 Prob(>chi-squared), (2) degrees of freedom: 0.000000*

Table A.29
LQG/OLS Regression Results for Little Rock Model I
Variable Coefficient StdError t-Statistic Probability Robust_SE Robust_t Robust_Pr VIF [1]
Intercept -0.112283 0.540345 -0.207799 0.836392 0.767065 -0.14638 0.884323 --------
PCT_MIN -0.00147 0.004389 -0.334839 0.739414 0.004015 -0.365973 0.716223 2.341584
MED_INC_99 -0.000003 0.000008 -0.431619 0.668226 0.000008 -0.391991 0.697048 2.260059
PCT_MFG 0.091489 0.021441 4.266993 0.000110* 0.041917 2.182614 0.034707* 1.260743
POP_DENSITY -0.029865 0.014997 -1.99147 0.05296 0.013941 -2.142232 0.038019* 1.047709

Number of Observations: 47 Number of Variables: 5
Degrees of Freedom: 42 Akaike's Information Criterion (AIC) [2]: 90.723907
Multiple R-Squared [2]: 0.419886 Adjusted R-Squared [2]: 0.364637
Joint F-Statistic [3]: 7.599883 Prob(>F), (4,42) degrees of freedom: 0.000106*
Joint Wald Statistic [4]: 9.513952 Prob(>chi-squared), (4) degrees of freedom: 0.049461*
Koenker (BP) Statistic [5]: 18.18284 Prob(>chi-squared), (4) degrees of freedom: 0.001137*
Jarque-Bera Statistic [6]: 44.110334 Prob(>chi-squared), (2) degrees of freedom: 0.000000*

320
Table A.30
LQG/OLS Regression Results for Miami Model III
Variable Coefficient StdError t-Statistic Probability Robust_SE Robust_t Robust_Pr VIF [1]
Intercept -0.036101 0.68709 -0.052541 0.958237 0.063071 -0.572383 0.568823 --------
PCT_WHITE -0.008291 0.015699 -0.528111 0.599027 0.008089 -1.024983 0.308752 8.571817
PCT_BLACK 0.005931 0.009138 0.648983 0.518385 0.005601 1.058764 0.293195 15.488507
PCT_HISPAN 0.00425 0.008864 0.479521 0.633006 0.004702 0.90387 0.369032 14.461626
PCT_OTHER 0.374652 0.102052 3.671201 0.000461* 0.201185 1.862232 0.066597 1.570923
PCT_MFG -0.005286 0.009655 -0.547529 0.585688 0.007444 -0.710187 0.479848 1.079801
SQ_INCOME 0.002141 0.003377 0.634032 0.528042 0.001917 1.117053 0.267633 3.442868
DENSITY_LOG -0.162669 0.072454 -2.245135 0.027785* 0.125759 -1.293498 0.199919 1.28041

Number of Observations: 81 Number of Variables: 8
Degrees of Freedom: 73 Akaike's Information Criterion (AIC) [2]: 187.439141
Multiple R-Squared [2]: 0.264123 Adjusted R-Squared [2]: 0.193559
Joint F-Statistic [3]: 3.743052 Prob(>F), (7,73) degrees of freedom: 0.001619*
Joint Wald Statistic [4]: 5.502159 Prob(>chi-squared), (7) degrees of freedom: 0.598923
Koenker (BP) Statistic [5]: 13.148521 Prob(>chi-squared), (7) degrees of freedom: 0.068571
Jarque-Bera Statistic [6]: 7330.324722 Prob(>chi-squared), (2) degrees of freedom: 0.000000*

Table A.31
LQG/OLS Regression Results for Norfolk Model III
Variable Coefficient StdError t-Statistic Probability Robust_SE Robust_t Robust_Pr VIF [1]
Intercept -0.160185 0.399489 -0.400975 0.689555 0.172777 -0.927123 0.356754 --------
PCT_WHITE 0.006282 0.006289 0.998805 0.321013 0.004164 1.508647 0.135487 16.418947
PCT_BLACK 0.009591 0.005287 1.813939 0.073586 0.005242 1.829585 0.071186 13.497664
PCT_HISPAN 0.033674 0.027921 1.206062 0.231487 0.027317 1.232721 0.221431 1.467153
PCT_OTHER 0.006035 0.023101 0.261237 0.79461 0.010736 0.562101 0.575682 1.711511
MED_INC_99 -0.000003 0.000006 -0.483597 0.630049 0.000004 -0.760385 0.449342 2.982519
PCT_MFG 0.027419 0.016971 1.615624 0.110272 0.018718 1.464848 0.147039 1.193994
SQ_DENSITY -0.15892 0.039648 -4.008279 0.000144* 0.077892 -2.040263 0.044752* 1.427607

Number of Observations: 85 Number of Variables: 8
Degrees of Freedom: 77 Akaike's Information Criterion (AIC) [2]: 109.156339
Multiple R-Squared [2]: 0.226343 Adjusted R-Squared [2]: 0.156011
Joint F-Statistic [3]: 3.218194 Prob(>F), (7,77) degrees of freedom: 0.004801*
Joint Wald Statistic [4]: 6.108347 Prob(>chi-squared), (7) degrees of freedom: 0.527157
Koenker (BP) Statistic [5]: 10.30572 Prob(>chi-squared), (7) degrees of freedom: 0.1719
Jarque-Bera Statistic [6]: 5855.027725 Prob(>chi-squared), (2) degrees of freedom: 0.000000*

321
Table A.32
LQG/OLS Regression Results for Salt Lake City Model II
Variable Coefficient StdError t-Statistic Probability Robust_SE Robust_t Robust_Pr VIF [1]
Intercept 1.99757 0.477305 4.185102 0.000107* 0.545287 3.66334 0.000570* --------
PCT_MIN -0.014324 0.007748 -1.84867 0.069985 0.008764 -1.634411 0.107993 3.627652
PCT_MFG 0.045171 0.023233 1.944232 0.057086 0.036476 1.238383 0.220931 2.625602
SQ_INCOME -0.005337 0.002011 -2.654193 0.010422* 0.001779 -3.000001 0.004080* 1.774558
DENSITY_LOG -0.285161 0.041044 -6.947649 0.000000* 0.081843 -3.484233 0.000989* 1.025588

Number of Observations: 59 Number of Variables: 5
Degrees of Freedom: 54 Akaike's Information Criterion (AIC) [2]: 101.993592
Multiple R-Squared [2]: 0.497333 Adjusted R-Squared [2]: 0.460098
Joint F-Statistic [3]: 13.356743 Prob(>F), (4,54) degrees of freedom: 0.000000*
Joint Wald Statistic [4]: 16.411979 Prob(>chi-squared), (4) degrees of freedom: 0.002513*
Koenker (BP) Statistic [5]: 18.901529 Prob(>chi-squared), (4) degrees of freedom: 0.000822*
Jarque-Bera Statistic [6]: 41.042647 Prob(>chi-squared), (2) degrees of freedom: 0.000000*

Tables A.33-A.38 present the data for the geographically weighted regressions on
park data for five of the six largest cities.

Table A.33
Park/GWR Results for Austin
Bandwidth : 23274.00618
ResidualSquares : 211.2835018
EffectiveNumber : 9.608847824
Sigma : 1.100704063
AICc : 566.7672578
R2 : 0.06921339
R2Adjusted : 0.023264956

Table A.34
Park/GWR Results for Boston
Bandwidth : 11876.51514
ResidualSquares : 277.9173487
EffectiveNumber : 9.141320887
Sigma : 1.370990413
AICc : 553.8844661
R2 : 0.070451135
R2Adjusted : 0.019268778

322
Table A.35
Park/GWR Results for Philadelphia
Bandwidth : 14410.92275
ResidualSquares : 86.8357238
EffectiveNumber : 11.97868348
Sigma : 0.485091302
AICc : 539.6383343
R2 : 0.041753729
R2Adjusted : 0.013245125

Table A.36
Park/GWR Results for San Diego
Bandwidth : 25666.35592
ResidualSquares : 86.5895653
EffectiveNumber : 11.95679523
Sigma : 0.551159673
AICc : 498.7245423
R2 : 0.079098299
R2Adjusted : 0.043699696

Table A.37
Park/GWR Results for San Jose
Bandwidth : 491456.975
ResidualSquares : 141.8684175
EffectiveNumber : 6.007835911
Sigma : 0.846484209
AICc : 519.4099202
R2 : 0.014265764
R2Adjusted : -0.010666512

Table A.38
Park/GWR Results for Seattle
Bandwidth : 6796.885386
ResidualSquares : 397.7379136
EffectiveNumber : 21.38891998
Sigma : 1.949887938
AICc : 542.2080142
R2 : 0.386717692
R2Adjusted : 0.267187677

The following tables (A.39-A.41) present the results for the geographically
weighted regressions on the community garden data for three of the six largest cities.

323

Table A.39
Community Garden/GWR Results for Boston
Bandwidth : 292233.2452
ResidualSquares : 302.2627871
EffectiveNumber : 6.005667888
Sigma : 1.414855276
AICc : 563.1476374
R2 : 0.14185791
R2Adjusted : 0.11340933

Table A.40
Community Garden/GWR Results for Philadelphia
Bandwidth : 393267.6229
ResidualSquares : 324.7343498
EffectiveNumber : 6.009889461
Sigma : 0.930581057
AICc : 1034.661008
R2 : 0.055327271
R2Adjusted : 0.042706389

Table A.41
Community Garden/GWR Results for Seattle
Bandwidth : 320064.7386
ResidualSquares : 44.77937503
EffectiveNumber : 6.009164143
Sigma : 0.610892755
AICc : 242.1807185
R2 : 0.085987161
R2Adjusted : 0.047830578

Disamenity Rankings for Medium and Small Cities

The following tables present the detailed rankings for TRI sites and LQG sites for
the six medium cities and five smallest cities (recall that Santa Fe has no hazards). These
tables supplement Tables 7.4 and 7.5.
Table A.42 displays the summary information for TRI sites in medium-sized
cities. From most to least sites, the cities rank as follows: Grand Rapids, Salt Lake City,

324
Dayton, Miami, Little Rock, and Norfolk. Dayton has the best distribution of TRI sites,
and is quite favorable for minorities, but low-income populations are definitely burdened.
Although Salt Lake City has the largest TRI releases, Norfolk is the least equitable city,
with minority percentages of 97% for its biggest producing and mean center tracts and
98% for its hot spots. Norfolk also ranks the worst for low-income populations; the
median income of its biggest-producing tract is only $11,929. Sadly, Salt Lake City is
distinguished as the only medium-sized city with TRI cold spots, where white
percentages exceeded 75% in 16 hot-spot tracts and median income ranged from roughly
$85,000 to $700,000.

Table A.42
The Six Medium Cities and TRI Sites
Total
Pounds
Persons
Per Site
Biggest
Producer
Tract
Minority
%
Biggest
Producer
Tract
Income
Mean
Center
Minority
%
Mean
Center
Income
Hot Spots
Minority
%
Hot
Spots
Income
Miami 4 5 3 2 n/a n/a 6 6
Norfolk 2 6 1 1 1 4 1 1
Grand
Rapids
5 2 5 5 2 3 3 5
Little
Rock
6 4 2 3 1 2 2 2
Salt
Lake
City
1 1 4 6 3 1 4 3
Dayton 3 5 6 4 4 5 5 4

From most to least equitable in terms of LQGs, the medium cities rank as follows:
Miami, Grand Rapids, Dayton, Salt Lake City, Little Rock, and Norfolk (see Table A.43).
Dayton was the biggest producer with twice the number of tons as the next biggest
producer, Grand Rapids. The least equitable city was Grand Rapids, followed by Little
Rock. Minority percentages were quite high in both cities, as high as 97% minority for

325
the mean center tract in Little Rock. Poor populations also suffered in both cities. Salt
Lake City again registered cold spots in predominantly white and middle-income areas.
The most equitable city appears to be Norfolk, which is interesting because it ranks as the
least equitable for TRI sites.

Table A.43
The Six Medium Cities and LQGs
Total
Tons
Persons
Per LQG
Biggest
Producer
Tract
Minority %
Biggest
Producer
Tract Income
Mean
Center
Minority
%
Mean
Center
Income
Hot Spots
Minority %
Hot
Spots
Income
Miami 5 4 1 1 6 6 1 1
Norfolk 4 6 6 6 1 3 5 / 6 5 / 6
Grand
Rapids
2 2 2 3 2 5 3 3
Little
Rock
6 5 3 2 2 1 2 2
Salt
Lake
City
3 3 5 5 4 4 4 4
Dayton 1 1 4 4 5 2 5 / 6 5 / 6

Table A.44 displays the summary information for TRI sites in the smallest cities
in the study. Recall that Santa Fe has no hazards as I defined them. From most to fewest
TRI sites, the small cities rank as follows: Boulder, Decatur, Albany, Charleston, and
Flagstaff, though Flagstaff has only one facility. Decatur is the least equitable city, as it
ranks first in terms of total pounds, has the lowest income ($25,317) in its biggest-
producing tract, a mean center minority percentage of 58%, and a mean-center median
income of $18,333. Charleston emerged as the most equitable, although it only had two
facilities that produced the smallest release amount of all the cities. The one area where
Charleston is inequitable is in median income in its mean center tract. It is a dismal
$9,397, though Albany’s is not much better at $10,897.

326
Table A.44
The Five Smallest Cities and TRI Sites
Total
Pounds
Persons
Per Site
Biggest
Producer
Tract
Minority %
Biggest
Producer
Tract
Income
Mean
Center
Minority
%
Mean
Center
Income
Hot Spots
Minority
%
Hot
Spots
Income
Albany 3 4 1 2 2 2 3 / 4 3 / 4
Boulder 2 1 2 4 N/A N/A 1 2
Decatur 1 2 4 1 1 3 2 1
Charleston 5 3 3 5 3 1 3 / 4 3 / 4
Flagstaff 4 5 2 3 N/A N/A N/A N/A

Though the number of LQGs in the smallest cities is low, the cities rank from
most to least as follows: Albany, Boulder, Decatur and Charleston, and Flagstaff (see
Table A.45). Boulder’s 8 facilities managed 56,767.20 tons of hazardous waste for
reporting year 2005. Again, Decatur emerged as the least equitable city, ranking high on
both minority percentage (58%) and median income ($18,333) for its biggest-producing
tract and mean center tract, which were the same. Flagstaff had only one LQG that
produced only 38.06 tons in 2005.

Table A.45
The Five Smallest Cities and LQGs
Total
Tons
Persons
Per LQG
Biggest
Producer
Tract
Minority %
Biggest
Producer
Tract
Income
Mean
Center
Minority
%
Mean
Center
Income
Hot Spots
Minority
%
Hot
Spots
Income
Albany 2 1 2 3 4 4 2 1
Boulder 1 3 3 4 3 3 1 2
Decatur 3 4 1 1 1 1 3 / 4 3 / 4
Charleston 4 2 5 2 2 2 3 / 4 3 / 4
Flagstaff 5 5 4 5 N/A N/A N/A N/A

Amenity Rankings for Medium and Small Cities

The following tables present the detailed rankings parks and community gardens
for the medium cities and small cities. These tables supplement Tables 7.9 and 7.10 in

327
Chapter 7. As for the medium-sized cities (see Table A.46), Grand Rapids has the most
parks, followed by Salt Lake City, Little Rock, Miami, Dayton, and Norfolk. Recall that a
ranking of six indicates the most equitable where amenities are concerned. Though Salt
Lake City has the most parks, it is the least equitable overall, while Miami and Little
Rock tied for being most equitable. The persons per parks ratios were impressive for
Grand Rapids, Salt Lake City, and Little Rock, with all of them at or below 1 park for
approximately every 3,500 people.

Table A.46
The Six Medium Cities and Parks
Parks Persons Per
Park
Mean Center
Minority %
Mean Center
Income
Hot Spots
Minority %
Hot Spots
Income
Miami 3 1 5 5 6 / 5 / 4 6 / 5 / 4
Norfolk 1 2 6 3 6 / 5 / 4 6 / 5 / 4
Grand
Rapids
6 6 1 1 3 2 / 3
Little Rock 4 4 4 2 6 / 5 / 4 6 / 5 / 4
Salt Lake
City
5 5 2 4 1 1
Dayton 2 3 3 6 2 2 / 3

The medium cities rank in terms of most-to-least equitable for community
gardens as: Dayton, Miami, Little Rock, Salt Lake City and Grand Rapids, and Norfolk
(see Table A.47). Dayton ranks as the most equitable, with 1 garden for every 9,232
persons, no hot spots, and a mean center located in a tract that is 89% minority with a
median income of $19,144. Norfolk ranks as the least equitable; the minority percentage
in its mean center tract is a disappointing 35% with a median income of $44,716.

328

Table A.47
The Six Medium Cities and Community Gardens
Gardens Persons Per
Garden
Mean Center
Minority %
Mean Center
Income
Hot Spots
Minority %
Hot Spots
Income
Miami 6 5 5 2 6 / 5 / 4 / 3 6 / 5 / 4 / 3
Norfolk 1 1 1 1 1 1
Grand
Rapids
2 / 3 2 2 4 2 2
Little Rock 4 4 4 3 6 / 5 / 4 / 3 6 / 5 / 4 / 3
Salt Lake
City
2 / 3 3 3 5 6 / 5 / 4 / 3 6 / 5 / 4 / 3
Dayton 5 6 6 6 6 / 5 / 4 / 3 6 / 5 / 4 / 3

Concerning the smallest cities (see Table A.48), Albany tops the small cities in
terms of number of parks, followed by Boulder, Decatur, Santa Fe, Flagstaff, and
Charleston. The least equitable city was Charleston, with the fewest parks and the largest
ratio of persons to parks. As well, parks clustered in tracts that were 74% white, with a
median income of $33,431.

Table A.48
The Six Smallest Cities and Parks
Parks Persons Per
Park
Mean
Center
Minority
%
Mean Center
Income
Hot Spots Minority
%
Hot Spots
Income
Albany 4 2 3 5 6 / 5 / 4 / 3 / 2 6 / 5 / 4 / 3 / 2
Boulder 6 6 1 2 6 / 5 / 4 / 3 / 2 6 / 5 / 4 / 3 / 2
Decatur 5 5 4 3 6 / 5 / 4 / 3 / 2 6 / 5 / 4 / 3 / 2
Santa Fe 3 3 6 4 6 / 5 / 4 / 3 / 2 6 / 5 / 4 / 3 / 2
Charleston 1 1 5 6 1 1
Flagstaff 2 4 2 1 6 / 5 / 4 / 3 / 2 6 / 5 / 4 / 3 / 2

Ranking from most to least community gardens among the smaller cities are:
Albany, Boulder, Decatur, Santa Fe, Flagstaff, and Charleston (see Table A.49). Albany’s
20 community gardens give it a ratio of one garden for every 4,783—second among all
cities in the study. Decatur ranks second in terms of equity, but Santa Fe does the best job

329
of representing minorities in its mean center tract, where the minority percentage is 74%.
Flagstaff has the worst minority percentage in its mean center tract.

Table A.49
The Six Smallest Cities and Community Gardens
Gardens Persons Per Garden Mean Center Minority % Mean Center Income
Albany 6 6 4 4
Boulder 5 5 2 1
Decatur 4 2 5 1
Santa Fe 3 4 6 3
Charleston 1 1 3 5
Flagstaff 2 3 1 2

Abstract (if available)

Abstract In 1982, the predominantly African American residents of Warren County, NC, protested for six weeks against the siting of a landfill to contain illegally contaminated dirt. Though unsuccessful, those protests sparked the Environmental Justice Movement, a movement dedicated to reducing the exposure of poor and minority populations to environmental hazards. Since then, only one Executive Order has been issued, no federal legislation has passed, and judicial efforts have been mixed. Moreover, local governments have little guidance or financial support to address environmental inequity.

Linked assets

University of Southern California Dissertations and Theses

Conceptually similar

PDF

The social making of authoritarian environmentalism: protest-litigation nexus and policy changes in China

PDF

Environmental justice: geospatial impacts of hazardous materials spills

PDF

Processes, effects, and the implementation of market-based environmental policy: southern California's experiences with emissions trading

PDF

Environmental justice in real estate, public services, and policy

PDF

Assessing the governance of the independent regulatory agencies in China

PDF

The role of social media for community-based organizations focused on environmental justice in southeast Los Angeles

PDF

A study of Chinese environmental NGOs: policy advocacy, managerial networking, and leadership succession

PDF

A framework for evaluating urban policy and its impact on social determinants of health (SDoH)

Asset Metadata

Creator Yoder, Diane E. (author)

Core Title Cities on the environmental justice frontline: the intractability of hazards and the governability of amenities

School School of Policy, Planning, and Development

Degree Doctor of Philosophy

Degree Program Public Administration

Publication Date 12/02/2009

Defense Date 06/26/2009

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

Tag environmental equity,environmental justice,institutional analysis,OAI-PMH Harvest,social movements

Place Name USA (countries)

Language English

Contributor Electronically uploaded by the author (provenance)

Advisor Tang, Shui-Yan (committee chair), Schweitzer, Lisa (committee member), Sellers, Jefferey M. (committee member)

Creator Email diane.personal@gmail.com,yoder@usc.edu

Permanent Link (DOI) https://doi.org/10.25549/usctheses-m2778

Unique identifier UC1314739

Identifier etd-Yoder-3125 (filename),usctheses-m40 (legacy collection record id),usctheses-c127-284451 (legacy record id),usctheses-m2778 (legacy record id)

Legacy Identifier etd-Yoder-3125.pdf

Dmrecord 284451

Document Type Dissertation

Rights Yoder, Diane E.

Type texts

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

Repository Name Libraries, University of Southern California

Repository Location Los Angeles, California

Repository Email cisadmin@lib.usc.edu