Full text of DOI:10.1111/0033-0124.00333

Urban Environmental Justice Indices
John Harner, Kee Warner, John Pierce, and Tom Huber
University of Colorado at Colorado Springs
Environmental justice is the principle that environmental costs and amenities ought to be equitably distributed within
society. Due to the ethical, political, and public-health implications, and because many choices confront those re-
searching environmental justice, standardized measures are needed to inform public dialogue and policy. We develop
and test seven indices on three Colorado cities to measure the relationship between the distribution of environmental
hazards and minority and poverty-stricken populations, and recommend the Comparative Environmental Risk Index
as a preliminary, standardized measure for comparing urban areas. This index is particularly relevant to disadvantaged
communities, regional planning organizations, environmental-justice networks and scholars, and state and federal
agencies. Key Words: Colorado, environmental justice, GIS.
he Environmental Protection Agency (EPA) other results can be compared (Cutter 1995;
website (EPA 2001) defines environmental Mohai 1995; Cutter, Holm, and Lloyd 1996;
T
justice (EJ) as the “fair treatment for people of McMaster, Leitner, and Sheppard 1997). In
all races, cultures, and incomes, regarding the this article we develop seven EJ indices using
development of environmental laws, regula- geographic information systems (GIS). We
tions, and policies.” EJ policies seek to create evaluate these measures using data from three
environmental equity: the concept that all Colorado metropolitan statistical areas (MSAs)
people should bear a proportionate share of and propose a standardized measure that is a
environmental pollution and health risk and preliminary indicator of environmental justice.
enjoy equal access to environmental amenities.
EJ policies are intended to overcome environ-
mental racism caused by racial and economic
Developing Indices
advantages built into policy-making, enforce- Researchers and policymakers have long recog-
ment, and locating of waste disposal and nized the value of trying to measure EJ. Prior
polluting industries.
work has explored using various geographic
Executive Order 12898, passed in 1994, re- units of analysis, types of statistical tests, and
quires each federal agency to adopt the princi- risk indicators. The geographic units of analysis
ple of environmental justice in policy develop- used in previous research include states (Lester
ment. As a result, there is an increasing need and Allen 1999), counties (Stockwell et al. 1993;
for a single quantitative EJ measurement Cutter, Holm, and Lloyd 1996), zip codes
method to help federal and state policymakers (United Church of Christ 1987), census tracts
in their decision processes. To date, EJ has (Cutter, Holm, and Lloyd; Yandle and Burton
been measured in many different ways, often 1996), and census block groups (Cutter, Holm,
with contradictory results (Mohai 1996; Wein- and Lloyd; Chakraborty and Armstrong 1997;
berg 1998; Lester and Allen 1999; Williams Sheppard et al. 1999). Demographic variables
1999; Holifield 2001). Researchers have de- used to measure EJ include median family in-
bated such issues as the most appropriate scale, come (Yandle and Burton), nonwhite popula-
the spatial units of analysis, the selection of so- tion (Yandle and Burton), percent nonwhite
cioeconomic variables, statistical techniques, (Glickman 1994; Chakraborty and Armstrong),
the definition of facilities or physical features percent below poverty level (Cutter, Holm, and
that pose a threat, selection of at-risk popula- Lloyd; Chakraborty and Armstrong; Sheppard
tions, and demographic characteristics of areas et al.), African-American and Hispanic (Lester
at the time noxious facilities were sited. Stan- and Allen), and median household income and
dardized methodologies are needed in order to percent black (Cutter, Holm, and Lloyd).
compute benchmark measures against which Statistical tests assessing the magnitude of
The Professional Geographer, 54(3) 2002, pages 318–331 © Copyright 2002 by Association of American Geographers.
Initial submission, August 2000; revised submission, September 2001; final acceptance, November 2001.
Published by Blackwell Publishing, 350 Main Street, Malden, MA 02148, and 108 Cowley Road, Oxford, OX4 1JF, UK.

Urban Environmental Justice Indices
319
disparities in the distribution of environmental
Although many hazardous waste sites occur
hazards include chi-square and Cramers V in rural areas (Bullard 1996), this study is con-
(Yandle and Burton), multiple regression (Mohai fined to urban areas. Toxic sites tend to be cor-
and Bryant 1992), and t-tests (Cutter, Holm, related with low income and minority areas in
and Lloyd). The EPA (1994) has also developed urban areas (Cutter 1995), and most toxic re-
an EJ index in which categorical rankings for de- lease and transfer sites are near large popula-
gree of exposure (based on population density) tion centers (Stockwell et al. 1993). We look at
are multiplied by degree of vulnerability (based potential exposure to toxic sites only, but our
on minority and economically stressed rank- tests are flexible enough to expand the analysis
ings). The source of environmental threat con- to certain other environmentally sensitive is-
sidered has included waste disposal facilities sues. Provided an environmental threat or
(United Church of Christ; Yandle and Burton;), amenity has a known location, such as a flood-
transport, storage, and disposal (TSD) sites plain, a zone of urban blight, open space pre-
monitored under the Research Conservation serves, or transportation corridors, these fea-
and Recovery Act (RCRA) (Mohai and Bryant), tures could substitute for the toxic sites used in
the Toxic Release Inventory (TRI) (Stockwell et this study, thereby broadening the usual con-
al; Glickman; Scott and Cutter 1997; Sheppard ception of “environment” and opening new
et al), and TSD, TRI, and Superfund sites (Cut- possibilities for EJ research (Holifield 2001).
ter, Holm, and Lloyd). Research also has at- We measure outcome equity between MSAs, a
tempted to improve upon the problem of comparative measure for one point in time
under- or overestimating those at risk due to only. We do not attempt to determine the
discrete boundary changes between the units of causes of environmental inequity or policy dis-
analysis by employing buffers around toxic sites crimination, we do not perform longitudinal
(Glickman; Anderton 1996), a series of buffers case studies, and we do not delve into the socio-
(Mohai and Bryant; Glickman; Sheppard et al.), logical constructs of race, ethnicity, or commu-
a combination of buffers and plume analysis nity. The indices we develop are tools, and
(Chakraborty and Armstrong), and a cumula- should be useful for researchers to address the
tive distance decay function (Cutter, Hodgson, nuances of specific locales (Mohai and Bryant
and Dow 2001).
1992; McMaster, Leitner, and Sheppard 1997)
In addition to inconsistencies in measure- and explore the relationship of EJ to such top-
ment and analysis, researchers have conducted ics as political culture, sustainability, segrega-
longitudinal case studies to determine demo- tion, economic base, growth, and community.
graphic characteristics at the time of hazardous Our results will provide a benchmark for fur-
waste facility siting (Yandle and Burton 1996; ther scholarly research, a practical community
Boone and Modarres 1999), addressed prob- indicator of environmental justice, and an ini-
lems with the conceptualization of racism tial comparison measure between cities.
(Pulido 2001), explored the definition of “com-
Conclusions often vary among EJ studies be-
munity” (Williams 1999), and examined the cause the research questions differ. We address
treatment of EJ in local sustainability efforts four EJ dimensions of toxic siting that have
(Warner 2002). Research into EJ has also con- concerned scholars, officials, and affected pop-
fronted such issues as unequal enforcement of ulations: (1) the likelihood of exposure for
environmental laws, exclusionary decision- different groups, (2) demographic differences
making processes, and discriminatory zoning between at-risk and not-at-risk populations, (3)
(Bullard 1996). These and other research the relationship of concentration of toxic sites to
projects address different dimensions of EJ and social characteristics, and (4) the relationship of
its determinants, including procedural inequi- the toxicity of sites to social characteristics. The
ties, generational inequities, and outcome in- following indices address these dimensions:
equities (Cutter 1995).
The focus of this project is quite specific: to
develop an EJ index that is conceptually valid
and easily computed for any city. Our goal is to
develop a standard index that is relatively sim-
ple to calculate and interpret.
1. Comparative Environmental Risk Index
(CERI): Are racial minorities and low-
income people more likely to be exposed
to environmental hazards than is the rest
of the population?

Volume54,Number3,August2002
320
2a. Toxic Demographic Difference Index (TDDI): sites (when appropriate) to select block groups
Do the demographics for areas of the (buffer distances are discussed below) to best
city that are vulnerable to toxic hazards define at-risk and not-at-risk populations
differ significantly from those for other (Mohai 1995). Numerous people recognize the
areas of the urban region?
need to use buffers around toxic sites so as to
2b. Toxic Demographic Quotient Index (TDQI): avoid the boundary problem associated with
Are the proportions of racial minori- the units of analysis and to avoid undercount-
ties and low-income people in the at- ing those at risk (Glickman 1994; Mohai 1995;
risk areas greater than in not-at-risk Anderton 1996; Bullard 1996).
areas?
Where possible, we define toxic sites com-
3a. Toxic Concentration Equity Index (TCEI): prehensively as any location included in federal
Are the numbers of toxic sites more con- and state databases (Table 1). The need for
centrated in minority and low-income measures that treat toxic sites comprehensively,
areas?
rather than particular subsets of toxic sites, is
3b. Concentration Risk Comparison Index well known (Mohai 1995; McMaster, Leitner,
(CRCI): Are racial minorities and low- and Sheppard 1997). Furthermore, state and
income people more likely to live near regional databases are frequently superior to
areas of high toxic concentrations than the federal records (Anderton 1996). We therefore
rest of the population?
use a comprehensive list that includes state data
3c. Concentration Demographics Index (CDI): in addition to federal data. We use the Envi-
Do the demographic characteristics of ronmental Geographics™ data from VISTA
areas of the city with high toxic concen- Information Solutions, Inc., a GIS data vendor
trations differ significantly from other that provides continually updated and reliable
areas of the urban region?
locations of toxic sites from all known sources
4. Toxicity Equity Index (TEI): Do minority (VISTA 1998, 2000). These toxic sites are
and low-income areas contain more po- represented as both point and polygon (area)
tentially dangerous types of toxic sites?
features.
While it is often desirable to assign some
We use accepted GIS algorithms in our analy- level of potential threat to each site (Stockwell
sis, which introduce a higher degree of objec- et al. 1993; Glickman 1994; Cutter, Hodgson,
tivity (Glickman 1994) and are recognized to and Dow 2001), several factors preclude this
be the tools now needed to address current for six of our seven indices. Much of the data
problems in EJ analysis (Cutter, Holm, and concerning the exact volume and chemical
Clark 1996).
makeup of toxins in these sites are unavailable
Our units of analysis are census block for all of the datasets. Also, on-site toxins vary
groups. These are the smallest unit for which greatly on a daily basis for most sites, which
the Census Bureau reports the necessary socio- means the risk at a site is never stable. Having
economic information, ensuring the best pos- acknowledged this, we still want to incorporate
sible degree of representativeness in geo- some measure that addresses the discrepancies
graphic areas (Cutter, Holm, and Clark 1996; in the level of potential threat each type of site
Yandle and Burton 1996; McMaster, Leitner, poses. Even this proves difficult, however, be-
and Sheppard 1997; Sheppard et al. 1999). cause the levels of toxins between sites for even
With the GIS we generate buffers around toxic one type, such as RCRA large generators, vary
Table 1 Data Sources forToxic Sites
• National Priorities List (NPL) superfund sites
• All other superfund sites (Comprehensive Environmental Response
Compensation and Liability Act/CERCLIS)
• Research Conservation and Recovery Act (RCRA)
Corrective Actions sites (CORRACTS)
• RCRA small generators
• RCRA large generators
• RCRA Transport, Storage, and Disposal Facilities (TSDFs)
• Toxic Release Inventory sites (TRIs).
• State-level equivalents of the federal data sets
Sources: VISTA (1998) for Pueblo and El Paso counties; VISTA (2000) for Adams, Arapahoe, Denver, Douglas, and Jefferson
counties.

Urban Environmental Justice Indices
321
greatly. The only realistic ranking scheme we conservative side, because the comprehensive
can use for a comprehensive list of toxic sites nature of our toxic-site data includes many sites
concerning the level of potential threat these that probably never actually release contami-
datasets contain is that the Superfund sites are nants, such as the RCRA sites, and are thus
probably worse than the rest (Superfunds are the only potential threats.
only confirmed polluted sites in the database,
While we feel using two buffer distances is
while many other types of sites may use and the best method by which to address variations
store toxins but never release them, or may re- in the level of toxicity associated with different
lease toxins within allowable specifications). types of sites at this time, our last measure spe-
We account for this by generating larger buffer cifically addresses discrepancies in the level of
distances around Superfund National Priority toxicity—but in TRI sites only. Here we em-
List (NPL) and Comprehensive Environmen- ploy a measure for the total emissions from
tal Response Compensation and Liability In- these sites, weighted by a toxicity ranking for
formation System (CERCLIS) sites (one mile) the type of chemicals emitted. The drawback is
than the rest (one-half mile) in those tests that the sites are not comprehensive to all toxic
where buffers are used.
datasets, but the severity of risk is factored in
The choice of buffer distances is also a con- (see Table 2 for a summary of the strengths of
tested issue (Bowen 1999). Landfill impact on each test).
housing values range from 2 to 2.5 miles
Methods to select the geographical areas at
(Mohai 1995). Glickman (1994) claims the radius risk still cause considerable variation in the at-
of an area affected by a major chemical release risk population even when an acceptable buffer
often exceeds one mile. Zimmerman (1994) uses distance is determined. Buffers are only used to
a one-mile buffer around NPL sites, while select the underlying units of analysis, because
Chakraborty and Armstrong (1997) use both these are the units at which socioeconomic data
one-half- and one-mile buffers around TRI are reported. This again shows the value of
sites. Sheppard and colleagues (1999) use 100-, using the smallest possible unit of analysis
500-, and 1000-yard buffers around TRI sites. (block groups) in order to best approximate the
The EPA (1989) has determined “inner zones,” actual buffer radius used. Buffers can be used to
or zones of plausible exposure, for their Hazard select all block groups that intersect the buff-
Ranking System (HRS) for NPL sites: one mile ers, all that have their centroid within the
for air; four miles for surface water; and for buffer, or only those that are completely within
land, twelve feet for the depth to aquifer, one the buffer. Alternatively, the population for
mile for the distance to the nearest well, and each block group that intersects the at-risk
one-half mile for down-gradient runoff. These buffers can be estimated by using the propor-
EPA figures are only guides, because the HRS tion of the block group that lies within the
is a peer-reviewed process of NPL sites that buffer (buffer containment method—see
accounts for site-specific characteristics about Sheppard et al. 1999). Chakraborty and Arm-
volume and type of toxins present. Cutter and strong (1997) estimated the at-risk population
colleagues (2001) avoid buffers altogether by in this way, but found almost identical values
using a distance-decay function—a thorough between the centroid containment method and
yet computationally intense method. We chose the buffer containment approximation method.
one-half- and one-mile buffers to be on the Because we want to best approximate the buffer
Table 2 Strengths ofTests
Strength
Test 1
Test 2a
Test 2b
Test 3a
Test 3b
Test 3c
Test 4
Uses buffers to best define
those at-risk
Yes
Yes
Yes
Yes
Yes
Yes
No
Yes
Yes
Yes
Yes
Yes
No
No
Comprehensive list of toxic sites
Accounts for differences
between individual toxic sites
Accounts for concentrations of
toxic sites
Partiallya
Partiallya
Partiallya
No
No
No
Yes
No
No
No
No
Yes
Yes
Yes
a By using different buffer distances between Superfund sites and all others.

Volume54,Number3,August2002
322
radius and minimize computations, we select This normalized composite index controls for
only those block groups that have their cen- variations in the overall burden an MSA popu-
troid within the buffer, eliminating those with lation bears in terms of toxic sites. We do this
large spatial extent for which the majority of because some cities have relatively low overall
the area lies outside of the toxic site buffer.
environmental dangers, with few toxic sites per
The discussion thus far on use of buffers, capita, while others have a higher per-capita
buffer distances, measures of toxicity, and spa- number of toxic sites. Normalizing the indices
tial units of analysis requires a clarification allows for direct comparison of the EJ index
about the underlying assumptions in these tests between cities, where cities with greater levels
and the usefulness of the results. We recognize of environmental risk score higher.
that equating proximity (however defined) with
risk is a crude measure. Bowen (1999) recog-
nizes that proximity has become the standard
Case-Study Cities
proxy for risk, but attacks this assumption as far The three cities chosen for this analysis are
too simplistic. The complexities of each site— Pueblo, Colorado Springs, and Denver (Figure
issues such as toxin doses, exposure times, and 1). All are located along the front range of Col-
synergistic effects of simultaneous multiple orado, a high-growth corridor. Each city has
chemical exposure, as well as details on wind, unique characteristics that should be reflected
atmospheric, and hydrographic conditions— in the EJ indices. We chose these cities for their
must be fully analyzed before a definitive state- proximity and diversity. The purpose is not to
ment of risk can be made. Only this level of determine the specifics of EJ in these cities per
detail will define risk. It is not our intent to se, but to test between these cities in order to
provide the methodology needed to accurately correlate the best EJ index. Our familiarity
measure risk in any specific place, but to find a with these cities and ready access for field ob-
preliminary indicator that is easy to perform and servations allows ongoing “reality checks” for
compares relative conditions between places. our EJ calculations.
We do not specifically equate proximity with
Colorado Springs is the second-largest city
risk. Our indices reveal inequalities in potential in Colorado, with a 2000 MSA (El Paso
risk. Furthermore, our analysis should clarify County) population of 516,929. Founded by
confusion as to why EJ studies often generate General William Jackson Palmer in 1871, the
conflicting results, and it provides a necessary city was neither the typical Western boomtown
step toward standardization (Szasz and Meuser nor a transportation hub. Rather, it was explic-
1997). Researchers can use our measure as a itly founded to be a blue-blood resort town, a
starting point, then proceed into local histories clean getaway for elites from the East Coast.
and local conditions as needed.
Dubbed “Newport in the Rockies,” Colorado
We perform all tests for race, ethnicity, and Springs was initially a dry town with many mil-
poverty level from the 1990 Census (U.S. Bu- lionaires, a wide representation of churches,
reau of the Census 1990). Race is measured by and the moral values of its founder stamped on
using the nonwhite populations. Because “His- the social space. The Cripple Creek gold rush
panic” can include all racial categories, Hispan- of the 1890s brought more wealth to the city,
ics are treated separately from race.1 We also but also changed the ambiance with the intro-
use persons below poverty level for the low- duction of gold smelters and more heavy indus-
income index: some of our tests generate mean try. Today the city hosts a strong military pres-
values, so we cannot use median or mean Cen- ence (typically an important player for local
sus values, such as median household income, environmental concerns), plus a concentration
to generate our mean values. We therefore of high-tech industries, including silicon-chip
generate an EJ index measured against racial manufacturers. Minority percentages match
minorities, Hispanics, and the poor for each the national averages, but are low for an urban
MSA. We then average these indices and nor- area.
malize by the rate of toxic sites in the MSA
using the formula:
Pueblo was historically larger than neigh-
boring Colorado Springs because of its strate-
gic location on the Arkansas River, but today
(number of toxic sites/population) ꢀ 1000 (1) has only 141,472 people. Pueblo is a blue-collar

Urban Environmental Justice Indices
323
Figure 1 Case-study urban areas
and MSA boundaries in Colorado.
city with a large Hispanic population, many of encing rapid urban sprawl, which may have the
whom are migrants from neighboring New effect of leaving disadvantaged groups behind in
Mexico (with more recent Mexican immi- the potentially noxious industrial locations.
grants) who worked in the Colorado Fuel and
In order to compare results between cities,
Iron Company steel mill. Pueblo has a rich di- we normalize our composite indices by the
versity of other European ethnic groups, ances- rate of toxic sites. The toxic site rate is highest
tors of earlier immigrants who came to work in for Denver (1.21948 toxic sites per 1000
the mills. The steel mill, now called Rocky people), next highest in Colorado Springs
Mountain Steel, still operates today but em- (0.85891), and lowest in Pueblo (0.72480).
ploys far fewer workers than its 1920 peak of Denver, therefore, has the highest overall
over 6,500. General Palmer chose the site of number of toxic sites per capita, so its normal-
the mill to build rails for his Denver and Rio ized composite indices will be scaled higher
Grande Railroad because of the availability of than the other cities.
water, nearby coal, and limestone, but also to
ensure that the noxious industry was not lo-
Test 1: Comparative Environmental
Risk Index
cated in his beloved Colorado Springs, where
the high society lived.
Denver, the central place for a vast “empire”
of mountains and plains, is the dominant urban
player in the region. The MSA consists of five
counties (Denver, Arapahoe, Adams, Jefferson,
and Douglas) with a 2000 population of
2,109,282. The city has a large Hispanic popu-
lation and a diversity of other racial and ethnic
types. In addition, it has a large manufacturing
sector, which historically included ore smelters
and later nuclear arsenal plants, but is domi-
nated today by the telecommunications industry
and other high-tech offices. Denver is experi-
The first index measures whether minorities
and low-income people are more likely to be
exposed to environmental hazards than are the
rest of the population. This measure selects all
block groups that have their centroid within a
one-mile buffer around NPL/CERCLIS sites
and a one-half-mile buffer around all other
toxic sites to define the at-risk population.
To generate the race index, we sum the non-
whites in at-risk block groups and divide by the
total MSA nonwhites, then do the same for the
whites. The index is a quotient (a ratio of

Volume54,Number3,August2002
324
ratios) of the two, or percent of nonwhites at
risk divided by the percent of whites at risk. We
repeat this measure for Hispanics and persons
Test 2a:Toxic Demographic
Differences Index
below poverty level.2 To calculate an overall This index determines if there are statistically
CERI, we then average the three individual significant differences between nonwhite, His-
quotients and normalize by multiplying by the panic, and poor populations near toxic sites (at-
total MSA toxicity rate to account for the per risk) and away from toxic sites (not-at-risk). To
capita number of toxic sites.
do so, we first determine the at-risk and not-at-
The results show that in Pueblo and Denver, risk populations with the same buffer distances
the poor as a group are at greater toxic risk than and block group selection method as above,
nonwhites (Table 3). In Colorado Springs, then calculate the mean number of nonwhites,
nonwhites are at highest risk (but all 3 indices Hispanics, and persons below poverty level in
are similar). The highest risk for all groups and both populations. Since the two populations
all cities is on the poor in Denver, who are 33.8 are mutually exclusive, independence is main-
percent more likely to be at risk than are the tained. The t-score from the difference of
non-poor. In Colorado Springs, nonwhites are means between these populations indicates the
33 percent more likely to be at risk than are probability of this inequality occurring at ran-
whites. Pueblo has the lowest levels of inequal- dom, and one minus the significance level (1 ꢁ
ity for the three cities.
Overall, much higher percentages of all
p) is the index for each variable.3
Results show that differences in the popula-
three groups (nonwhite, poor, and Hispanic) tions for all three variables in each city are
are at risk in Denver than in the other two cit- highly significant (Table 4). On average, at-
ies. This higher overall risk in Denver is accen- risk areas have higher proportions of non-
tuated by the fact that Denver has a higher rate whites, Hispanics, and low-income people. In
of toxic sites than do the other two cities (ac- Denver, the percent poor in at-risk areas is
counted for in our normalization procedure). over three times that of not-at-risk areas, and
The normalized composite index ranks Denver differences between the percent nonwhite is
as most severe, Colorado Springs second, and nearly as great. Pueblo again shows the small-
Pueblo the least severe.
est differences between at-risk and not-at-risk
Table 3 Comparative Environmental Risk Indices (Test 1)
Pueblo
(at risk nonwhites/total MSA nonwhites)
(at risk whites/total MSA whites)
0.49261
0.43316
CERInw
CERIh
CERIp
ꢂ
ꢂ
ꢂ
ꢂ
ꢂ
ꢂ
ꢂ
ꢂ
ꢂ
1.137
1.157
1.229
(at risk Hispanic/total MSA Hispanic)
(at risk non-Hispanic/total MSA non-Hispanic)
0.48443
0.41874
(at risk poor/total MSA poor)
(at risk non-poor/total MSA non-poor)
0.52019
0.42300
Normalized composite CERI ꢂ (1.137 ꢃ 1.157 ꢃ 1.229)/3 ꢀ (0.73140) ꢂ 0.8555
Colorado Springs
(at-risk nonwhites/total MSA nonwhites)
(at-risk whites/total MSA whites)
0.64966
0.48804
CERInw
CERIh
CERIp
ꢂ
ꢂ
ꢂ
ꢂ
ꢂ
ꢂ
ꢂ
ꢂ
ꢂ
1.331
1.272
1.292
(at-risk Hispanic/total MSA Hispanic)
(at-risk non-Hispanic/total MSA non-Hispanic)
0.63478
0.49896
(at-risk poor/total MSA poor)
(at-risk non-poor/total MSA non-poor)
0.64093
0.49596
Normalized composite CERI ꢂ (1.331 ꢃ 1.272 ꢃ 1.292)/3 ꢀ (0.85891) ꢂ 1.1143
Denver
(at-risk nonwhites/total MSA nonwhites)
(at-risk whites/total MSA whites)
0.81796
0.62529
CERInw
CERIh
CERIp
ꢂ
ꢂ
ꢂ
ꢂ
ꢂ
ꢂ
ꢂ
ꢂ
ꢂ
1.308
1.266
1.338
(at-risk Hispanic/total MSA Hispanic)
(at-risk non-Hispanic/total MSA non-Hispanic)
0.79929
0.63120
(at-risk poor/total MSA poor)
(at-risk non-poor/total MSA non-poor)
0.84605
0.63227
Normalized composite CERI ꢂ (1.308 ꢃ 1.266 ꢃ 1.338)/3 ꢀ (1.21948) ꢂ 1.5902

Urban Environmental Justice Indices
325
Table 4 Toxic Demographic Differences Indices (Test 2a)
At Risk
Mean
Not-At-Risk
Mean
One-Tailed t
Probabilitya
Pueblo
Percent nonwhite
Percent Hispanic
Percent poor
0.179
0.420
0.253
0.121
0.303
0.176
t ꢂ 3.005c
t ꢂ 3.239b
t ꢂ 3.528b
p ꢂ 0.0015
p ꢂ 0.0007
p ꢂ 0.0003
TDDInw ꢂ (1 ꢁ p) ꢂ 0.9985
TDDIh ꢂ (1 ꢁ p) ꢂ 0.9993
TDDIp ꢂ (1 ꢁ p) ꢂ 0.9997
Normalized composite TDDI ꢂ (0.9985 ꢃ 0.9993 ꢃ 0.9997)/3 ꢀ (0.73140) ꢂ 0.7308
Colorado Springs
Percent nonwhite
Percent Hispanic
Percent poor
0.157
0.112
0.145
0.106
0.066
0.084
t ꢂ 3.761b
t ꢂ 5.719b
t ꢂ 5.527c
p ꢂ 0.0001
p ꢅ 0.0000
p ꢅ 0.0000
TDDInw ꢂ (1 ꢁ p) ꢂ 0.9999
TDDIh ꢂ (1 ꢁ p) ꢂ 1.0
TDDIp ꢂ (1 ꢁ p) ꢂ 1.0
Normalized composite TDDI ꢂ (0.9999 ꢃ 1.0 ꢃ 1.0)/3 ꢀ (0.85891) ꢂ 0.8589
Denver
Percent nonwhite
Percent Hispanic
Percent poor
0.196
0.164
0.138
0.073
0.077
0.045
t ꢂ 15.316c
t ꢂ 11.289c
t ꢂ 18.049c
p ꢅ 0.0000
p ꢅ 0.0000
p ꢅ 0.0000
TDDInw ꢂ (1 ꢁ p) ꢂ 1 .0
TDDIh ꢂ (1 ꢁ p) ꢂ 1.0
TDDIp ꢂ (1 ꢁ p) ꢂ 1.0
Normalized composite TDDI ꢂ (1.0 ꢃ 1.0 ꢃ 1.0)/3 ꢀ (1.21948) ꢂ 1.2195
a All probabilities are significant.
b Equal variance assumed.
cEqual variance not assumed.
populations; nonetheless, the differences are normalized by the MSA toxic site rates, Denver
significant. has the highest composite index of 3.0211, Col-
The normalized composite index is the aver- orado Springs the next, and Pueblo the lowest.
age of the three indices, again normalized by
the rate of toxic sites in the MSA. As with test 1,
Denver showed the highest inequality, Colo-
rado Springs second, and Pueblo third.
Test 3a:Toxic Concentration
Equity Index
This procedure generates an index indicating
the degree to which toxic sites are concentrated
within nonwhite, Hispanic, and low-income
areas. The previous tests found at-risk popula-
tions near toxic sites, but gave no indication as
to the number of toxic sites nearby. To generate
this index, we compare two distributions: the
percent of an MSA’s toxic sites in each block
group, and the percent of the MSA population
in each block group. The assumption is that eq-
uity occurs when a block group’s proportion of
the MSA’s toxic sites equals that block group’s
proportion of the MSA population. The ratio
of disadvantage, or the percentage of toxic sites
divided by the percentage of the population,
indicates the relative burden for each block
group. Numbers greater than 1.0 indicate a
disproportionate burden of toxic sites given
their share of the population, numbers less
than 1.0 have less than their fair share.
Test 2b:Toxic Demographics
Quotient Index
We can also measure inequality by comparing
the proportions of racial minorities and poor in
at-risk areas to their proportions in not-at-risk
areas. The index is the quotient between the at-
risk percentages and the not-at-risk percentages.
We use the same selection method to define the
at-risk population as in the two previous tests.
As expected, the minority and poor propor-
tions of the population in at-risk areas are
higher than their proportions in not-at-risk
areas for each city (Table 5). The highest ineq-
uity shows that the proportion of poor people
in the at-risk population in Denver is nearly
three times higher than the proportion of poor
people in the not-at-risk population. The Den-
ver at-risk areas also have over two times the
proportions of minorities of the not-at-risk
areas. Colorado Springs indices show that at-
risk areas have over one and a half times higher
proportions of minorities and poor than not-
at-risk areas. Pueblo again shows the least
inequality. When the individual indices are
A Gini coefficient (G) tells the degree of in-
equality between the two distributions (percent
of toxic sites and percent of the MSA population):
G ꢂ .5 ꢀ ꢄꢀ(percent of toxic sites) ꢁ
(percent of total population)ꢀ (2)

Volume54,Number3,August2002
326
Table 5 Toxic Demographic Quotient Indices (Test 2b)
Pueblo
(at-risk nonwhites/total MSA at-risk)
(not-at-risk nonwhites/total MSA not-at-risk)
0.16964
0.13852
TDQInw
TDQIh
TDQIp
ꢂ
ꢂ
ꢂ
ꢂ
ꢂ
ꢂ
ꢂ
ꢂ
ꢂ
1.2247
1.1852
1.3655
(at-risk Hispanic/total MSA at-risk)
(not-at-risk Hispanic/total MSA not-at-risk)
0.39141
0.33026
(at-risk poor/total MSA at-risk)
(not-at-risk poor/total MSA not-at-risk)
0.23247
0.16999
Normalized composite TDQI ꢂ (1.2247 ꢃ 1.1852 ꢃ 1.3655)/3 ꢀ (0.73140) ꢂ 0.9204
Colorado Springs
(at-risk nonwhites/total MSA at-risk)
(not-at-risk nonwhites/total MSA not-at-risk)
0.17605
0.09897
TDQInw
ꢂ
ꢂ
ꢂ
ꢂ
ꢂ
ꢂ
ꢂ
ꢂ
ꢂ
1.7788
1.6671
1.7122
(at-risk Hispanic/total MSA at-risk)
(not-at-risk Hispanic/total MSA not-at-risk)
0.10473
0.06282
TDQIh
(at-risk poor/total MSA at-risk)
(not-at-risk poor/total MSA not-at-risk)
0.12499
0.07300
TDQIp
Normalized composite TDQI ꢂ (1.7788 ꢃ 1.6671 ꢃ 1.7122)/3 ꢀ (0.85891) ꢂ 1.4768
Denver
(at-risk nonwhites/total MSA at-risk)
(not-at-risk nonwhites/total MSA not-at-risk)
0.17862
0.07473
TDQInw
TDQIh
ꢂ
ꢂ
ꢂ
ꢂ
ꢂ
ꢂ
ꢂ
ꢂ
ꢂ
2.3902
2.1183
2.9237
(at-risk Hispanic/total MSA at-risk)
(not-at-risk Hispanic/total MSA not-at-risk)
0.15701
0.07412
(at-risk poor/total MSA at-risk)
(not-at-risk poor/total MSA not-at-risk)
0.12420
0.04248
TDQIp
Normalized composite TDQI ꢂ (2.3902 ꢃ 2.1183 ꢃ 2.9237)/3 ꢀ (1.21948) ꢂ 3.0211
Gini coefficients vary from 0 (complete concentrations—fully 75 percent of the toxic
equality) to 100 (complete inequality). The sites would need to be redistributed to achieve
calculated Gini coefficient shows that Pueblo equity (Table 6). However, this measure says
has the most unequal burden between toxic nothing about the population that is adversely
concentrations and the overall population affected by inequality. To generate an EJ index,
Table 6 Toxic Concentration Equity Index (Test 3a)
Pueblo
Percent nonwhite
Percent Hispanic
Percent poor
RD ꢂ 13.31 ꢁ 27.836 (%nw)
RD ꢂ 11.83 ꢁ 6.240 (%Hispanic)
RD ꢂ 2.35 ꢃ 14.060 (%poor)
R
R
R
2 ꢂ 0.038 F ꢂ 1.476
2 ꢂ 0.009 F ꢂ 0.345
2 ꢂ 0.142 F ꢂ 6.126
p ꢂ 0.232 TCEInw ꢂ (1 ꢁ p) ꢂ 0.768
p ꢂ 0.561 TCEIh ꢂ (1 ꢁ p) ꢂ 0.439
p ꢂ 0.018a TCEIp ꢂ (1 ꢁ p) ꢂ 0.982
Multiple regression
RD ꢂ 6.719 ꢃ 7.65 (%nw) ꢁ32.27 (%Hispanic) ꢃ 59.47 (%poor)
R
2 ꢂ 0.280
F ꢂ 4.535
p ꢂ 0.009
Normalized composite TCEIb ꢂ (1ꢁ 0.009) ꢀ (0.73140) ꢂ 0.7248
Colorado Springs
Percent nonwhite
Percent Hispanic
Percent poor
RD ꢂ 4.37 ꢁ 2.875 (%nw)
RD ꢂ 4.17 ꢁ 2.487 (%Hispanic)
RD ꢂ 1.695 ꢃ 15.240 (%poor)
R
R
R
2 ꢂ 0.005 F ꢂ 0.500
2 ꢂ 0.001 F ꢂ 0.135
p ꢂ 0.481 TCEInw ꢂ (1 ꢁ p) ꢂ 0.519
p ꢂ 0.714 TCEIh ꢂ (1 ꢁ p) ꢂ 0.286
2 ꢂ 0.102 F ꢂ 12.405 p ꢂ 0.001a TCEIp ꢂ (1 ꢁ p) ꢂ 0.999
Multiple regression
RD ꢂ 2.975 ꢁ 4.92 (%nw) ꢁ13.49 (%Hispanic) ꢃ 22.07 (%poor)
R
2 ꢂ 0.162
F ꢂ 6.920
p ꢂ 0.000
Normalized composite TCEIb ꢂ (1ꢁ 0.000) ꢀ (0.85891) ꢂ 0.8589
Denver
Percent nonwhite
Percent Hispanic
Percent poor
RD ꢂ 3.94 ꢃ 7.795 (%nw)
RD ꢂ 4.40 ꢁ 5.567 (%Hispanic)
RD ꢂ 4.50 ꢃ 5.420 (%poor)
R
R
R
2 ꢂ 0.009 F ꢂ 4.950
2 ꢂ 0.004 F ꢂ 2.432
2 ꢂ 0.002 F ꢂ 1.305
p ꢂ 0.026a TCEInw ꢂ (1 ꢁ p) ꢂ 0.974
p ꢂ 0.119 TCEIh ꢂ (1 ꢁ p) ꢂ 0.881
p ꢂ 0.254 TCEIp ꢂ (1 ꢁ p) ꢂ 0.746
Multiple regression
RD ꢂ 3.99 ꢃ 8.06 (%nw) ꢃ 2.58 (%Hispanic) ꢁ 3.52 (%poor)
R
2 ꢂ 0.010
F ꢂ 1.794
p ꢂ 0.147
Normalized composite TCEIb ꢂ (1ꢁ 0.147) ꢀ ( 1.21948) ꢂ 1.0402
Note: Pueblo: Gini ꢂ 74.76; Colorado Springs: Gini ꢂ 65.70; Denver: Gini ꢂ 69.35.
a Significant relationships.
b Normalized composite (TCEI): 1 ꢁ (p value of multiple regression) ꢀ (number of toxic sites)/(total population) ꢀ 1000.

Urban Environmental Justice Indices
327
we use simple linear regression to regress the posite index by multiplying one minus the mul-
ratio of disadvantage on the percent nonwhite, tiple regression probability times the MSA toxic
then on percent Hispanic, and finally on the site rate. Once again, Denver has the highest
percent poor for each block group. Each index composite index, Colorado Springs the second,
is one minus the probability associated with and Pueblo the lowest.
each regression equation (the linear relation-
ship between areas with disproportionate num-
bers of toxic sites and each demographic vari-
Tests 3b and 3c
able). We excluded all block groups that had Using the ratio of disadvantage from test 3a al-
zero toxic sites because the dependent variable lows a new method to define those areas at-risk.
(ratio of disadvantage) is highly skewed when We first select all block groups that bear more
all block groups are included (as many block than five times their fair burden (ratio of dis-
groups have no toxic sites). We are therefore advantage ꢆ5.0), then include all block groups
testing whether, for the population at risk, whose centroid is within one-half mile of this
higher percentage of poor, nonwhite, or His- selected set. This selection method defines at-
panic explain higher concentrations (burdens) risk areas as only those block groups near areas
of toxicity.
of high burden (toxic-site concentrations
Although Pueblo has an unequal distribution higher than population concentrations). Once
of toxic sites compared to population, only the this selection method is completed, we com-
percent poor is significantly related to this in- pare the demographics for those areas to the
equality, explaining 14.2 percent of the varia- remainder of the population with a compara-
tion in the burden ratio. Colorado Springs also tive-risk test similar to test 1 and a difference-
has a strong relationship between percent poor of-means test similar to Test 2a.
and toxic concentrations (R2 ꢂ 0.102), while the
Normalized composite indices for test 3b
percent nonwhite explains almost 1 percent of show the same trend in rankings, with Denver
the variation in toxic concentrations in Denver. the highest EJ index, Colorado Springs second,
To generate a composite index, we first run a and Pueblo third (Table 7). However, within
multiple regression with all three variables in the cities, likelihoods of living near concentra-
the equation, then calculate a normalized com- tions of toxic sites for the demographic groups
Table 7 Concentration Risk Comparison Indices (Test 3b)
Pueblo
(at risk nonwhites/total MSA nonwhites)
(at risk whites/total MSA whites)
0.39324
0.33575
CRCInw
CRCIh
CRCIp
ꢂ
ꢂ
ꢂ
ꢂ
ꢂ
ꢂ
ꢂ
ꢂ
ꢂ
1.171
1.317
1.367
(at risk Hispanic/total MSA Hispanic)
(at risk non-Hispanic/total MSA non-Hispanic)
0.40746
0.30950
(at risk poor/total MSA poor)
(at risk non-poor/total MSA non-poor)
0.43918
0.32118
Normalized composite CRCI ꢂ (1.171 ꢃ 1.317 ꢃ 1.367)/3 ꢀ (0.73140) ꢂ 0.9398
Colorado Springs
(at risk nonwhites/total MSA nonwhites)
(at risk whites/total MSA whites)
0.33354
0.24139
CRCInw
CRCIh
CRCIp
ꢂ
ꢂ
ꢂ
ꢂ
ꢂ
ꢂ
ꢂ
ꢂ
ꢂ
1.382
1.471
1.717
(at risk Hispanic/total MSA Hispanic)
(at risk non-Hispanic/total MSA non-Hispanic)
0.35965
0.24443
(at risk poor/total MSA poor)
(at risk non-poor/total MSA non-poor)
0.40734
0.23720
Normalized composite CRCI ꢂ (1.382 ꢃ 1.471 ꢃ 1.717)/3 ꢀ (0.85891) ꢂ 1.3084
Denver
(at risk nonwhites/total MSA nonwhites)
(at risk whites/total MSA whites)
0.77758
0.63704
CRCInw
CRCIh
CRCIp
ꢂ
ꢂ
ꢂ
ꢂ
ꢂ
ꢂ
ꢂ
ꢂ
ꢂ
1.221
1.243
1.299
(at risk Hispanic/total MSA Hispanic)
(at risk non-Hispanic/total MSA non-Hispanic)
0.79200
0.63722
(at risk poor/total MSA poor)
(at risk non-poor/total MSA non-poor)
0.82976
0.63877
Normalized composite CRCI ꢂ (1.221 ꢃ 1.243 ꢃ 1.299)/3 ꢀ (1.21948) ꢂ 1.5296

Volume54,Number3,August2002
328
are greater than those seen in test 1. For exam- [(emissions) ꢀ (toxicity value)] for each TRI
ple, in Colorado Springs, the poor are 72 per- site in the three MSAs, then regress its log
cent more likely to live near toxic-site concen- transformation (because the range of values, a
trations than are the non-poor; Hispanics are unitless measure, is very large) on the three
47 percent more likely to do so, and nonwhites demographic variables. Pueblo is excluded from
are 38 percent more likely to do so.
this test because only two TRI sites were iden-
Like test 2a, test 3c shows highly significant tified with the EPA toxicity index. Colorado
differences between the at-risk populations of Springs is included, but only has a sample size
nonwhites, Hispanics, and poor and the not-at- of eleven. No significant results occurred, so no
risk populations. On average, there are much linear relationship exists between the toxicity
higher percentages for each of these groups near index for TRI emissions and the demographic
toxic site concentrations (Table 8). Again, Denver variables (Table 9). Denver again has a higher
has the highest level of inequality, Colorado composite EJ index than Colorado Springs.
Springs the second, and Pueblo the lowest.
Test Comparisons and Conclusions
Test 4:Toxicity Equity
While the indices we generated indicate vary-
For the final test, we want to better account for ing degrees of environmental injustice, the end
the degree of severity, or actual threat, between results are remarkably similar. All tests show
toxic sites, rather than to treat all toxic sites that Denver has the highest levels of injustice,
within a certain database as equal. We there- while Pueblo has the lowest. The similarities
fore employ a measure of total emission re- between indices are encouraging, showing that
leases for sites in the TRI database. The emis- contradictory results between EJ measures are
sion releases are then weighted by the level of probably not as common as some claim, al-
toxicity for each chemical released, as devel- though investigators could substitute other
oped by the Office of Pollution Prevention and variables or at-risk selection methods that may
Toxics in the Risk Screening Environmental Indi- alter results.
cators CD (EPA 1999). This measure discrimi-
When comparing the similarities between
nates between toxic sites, rather than treating indices, test 1 shows up as most similar to the
them all the same, but only includes TRI sites others (Table 10). Test 1 and test 3b employ the
(as opposed to a comprehensive list, in the pre- same statistical test, but with a different
vious tests).
method to select those at-risk (as is also true
For this method, we query the EPA index between tests 2a and 3c). It is not surprising
Table 8 Concentration Demographics Indices (Test 3c)
At Risk
Mean
Not-At-Risk
Mean
One-Tailed t
Probabilitya
Pueblo
Percent nonwhite
Percent Hispanic
Percent poor
0.168
0.434
0.270
0.133
0.307
0.174
t ꢂ 1.740b
t ꢂ 3.444b
t ꢂ 4.303b
p ꢂ 0.0419
p ꢂ 0.0004
p ꢂ 0.00001
CDInw ꢂ (1 ꢁ p) ꢂ 0.9581
CDIh ꢂ (1 ꢁ p) ꢂ 0.9996
CDIp ꢂ (1 ꢁ p) ꢂ 0.9999
Normalized composite CDI ꢂ (0.9581 ꢃ 0.9996 ꢃ 0.9999)/3 ꢀ (0.73140) ꢂ 0.7211
Colorado Springs
Percent nonwhite
Percent Hispanic
Percent poor
0.174
0.125
0.170
0.113
0.073
0.090
t ꢂ 3.837c
t ꢂ 5.391c
t ꢂ 6.099c
p ꢂ 0.00009
p ꢅ 0.0000
p ꢅ 0.0000
CDInw ꢂ (1 ꢁ p) ꢂ 0.9999
CDIh ꢂ (1 ꢁ p) ꢂ 1.000
CDIp ꢂ (1 ꢁ p) ꢂ 1.000
Normalized composite CDI ꢂ (0.9999 ꢃ 1.0 ꢃ 1.0)/3 ꢀ (0.85891) ꢂ 0.8589
Denver
Percent nonwhite
Percent Hispanic
Percent poor
0.186
0.162
0.135
0.098
0.078
0.051
t ꢂ 8.904c
t ꢂ 11.330c
t ꢂ 16.460c
p ꢅ 0.000
p ꢅ 0.000
p ꢅ 0.000
CDInw ꢂ (1 ꢁ p) ꢂ 1.000
CDIh ꢂ (1 ꢁ p) ꢂ 1.000
CDIp ꢂ (1 ꢁ p) ꢂ 1.000
Normalized composite CDI ꢂ (1.0 ꢃ 1.0 ꢃ 1.0)/3 ꢀ (1.21948) ꢂ 1.2195
a
All probabilities are significant.
bEqual variance assumed.
c Equal variance not assumed.

Urban Environmental Justice Indices
329
Table 9 Toxicity Equity Index (Test 4)
Pueblo
N/A
Colorado Springs
Percent nonwhite
Percent Hispanic
Percent poor
logtox ꢂ 8.57 ꢁ 6.240(%nw)
logtox ꢂ 7.50 ꢁ 1.056(%Hispanic)
logtox ꢂ 6.86 ꢃ 4.493(%poor)
R
R
R
2 ꢂ 0.075 F ꢂ 0.725 p ꢂ 0.417 TEInw ꢂ (1 ꢁ p) ꢂ 0.583
2 ꢂ 0.001 F ꢂ 0.005 p ꢂ 0.944 TEIh ꢂ (1 ꢁ p) ꢂ 0.056
2 ꢂ 0.067 F ꢂ 0.648 p ꢂ 0.442 TEIp ꢂ (1 ꢁ p) ꢂ 0.558
Multiple regression
logtox ꢂ 8.95 ꢁ 6.38 (%nw) ꢁ22.44 (%Hispanic) ꢃ 13.36 (%poor)
R
2 ꢂ 0.288
F ꢂ 0.943
p ꢂ 0.470
Normalized composite TEIa ꢂ (1 ꢁ 0.470) ꢀ (0.85891) ꢂ 0.4552
Denver
Percent nonwhite
Percent Hispanic
Percent poor
logtox ꢂ 6.03 ꢃ 0.505(%nw)
logtox ꢂ 6.27 ꢁ 4.200(%Hispanic)
logtox ꢂ 5.58 ꢃ 2.892(%poor)
R
R
R
2 ꢂ 0.006 F ꢂ 0.327 p ꢂ 0.570 TEInw ꢂ (1 ꢁ p) ꢂ 0.430
2 ꢂ 0.002 F ꢂ 0.097 p ꢂ 0.757 TEIh ꢂ (1 ꢁ p) ꢂ 0.243
2 ꢂ 0.047 F ꢂ 2.639 p ꢂ 0.110 TEIp ꢂ (1 ꢁ p) ꢂ 0.890
Multiple regression
logtox ꢂ 5.76 ꢁ 0.64 (%nw) ꢁ2.09 (%Hispanic) ꢃ 5.09 (%poor)
R
2 ꢂ 0.085
F ꢂ 1.575
p ꢂ 0.207
Normalized composite TEIa ꢂ (1 ꢁ 0.207) ꢀ (1.21948) ꢂ 0.9670
a Normalized Composite (TEI) ꢂ 1ꢁ (p value of multiple regression) ꢀ (number of toxic sites)/(total population) ꢀ 1000.
then that these indices are highly correlated. very low probability values (for instance, p ꢅ
Test 1 is also highly correlated with tests 2a and .0000000000 for all demographic groups in test
3c. The index most unlike the others was test 3c for Denver); hence, the indices equal 1.0,
3a, where we regressed the ratio of disadvan- and the normalized composite index is reduced
tage on the demographic variables.
The two tests that calculate quotients, tests 1
to simply the toxicity rate for that MSA.
The goal of keeping the index simple disfa-
and 2b, produce numbers that are most mean- vors tests 3a, 3b, and 3c because they require
ingful. They are also simple to calculate and more calculations for little to no extra benefit.
easily interpretable. Test 1 calculations show The measures of concentration used in those
the percentage of each demographic group that tests are simply a refinement on tests 1 and 2a,
is at risk in an urban area, and how much more a different method to determine the at-risk
likely that group is to live in at-risk areas than is groups. The normalized composite indices cal-
the remainder of the population. Test 2b shows culated using multiple regression (tests 3a and
the proportion of the at-risk population that is 4) are also somewhat flawed. Since some, if not
comprised of each demographic group, the all three, of the demographic variables are not
same for not-at-risk areas, and the ratio (in- linearly related to the dependent variable in
equality) between these proportions.
these tests, several actually have a negative co-
The remaining tests are less easy to inter- efficient in the multiple regression equation.
pret, or the methods to produce the indices This occurs even though the mean value for all
have shortcomings. Tests 2a and 3c only deter- demographic variables in the at-risk group is
mine whether a statistical difference occurs be- significantly higher and the likelihood of being
tween populations, a conclusion that is not as exposed is greater.
useful as the results from tests 1 or 2b. The dif-
Test 4 is the only measure that accounts for
ference of means for tests 2a and 3c also have toxicity levels, a point many researchers claim
Table 10 Correlations between Indicesa
Test 1
Test 2a
Test 2b
Test 3a
Test 3b
Test 3c
Test 4
Test 1
1.00
0.931c
1.00
0.831c
0.669b
1.00
0.353
0.346
0.695b
0.711b
0.248
0.198
0.353
1.00
0.113
0.195
0.491
0.694b
0.640b
0.346
0.113
1.00
0.950c
0.995c
0.695b
0.195
0.670b
1.00
0.638
0.513
0.711b
0.491
0.179
0.513
1.00
Test 2a
Test 2b
Test 3a
Test 3b
Test 3c
Test 4
0.931c
0.831c
0.248
0.694b
0.950c
0.638
0.669b
0.198
0.640b
0.995c
0.513
0.670b
0.179
0.513
a n ꢂ 12 for all tests except test 4, where n ꢂ 8.
b Significant at the 0.05 level (2-tailed).
c Significant at the 0.01 level (2-tailed).

Volume54,Number3,August2002
330
only two indices, one for minorities and one for
poor, but we did not find an easy way to combine
race and Hispanic variables at the block-group
level.
is necessary. Relying on TRI data for this mea-
sure excludes a comprehensive set of toxic sites,
so this test is only applicable to larger urban
areas. While we agree that the level of toxicity
is an important variable that could be used in
defining an EJ index, at this time adequate data
are not available. Should more thorough mea-
sures of site toxicity become available in the fu-
ture, indices could be developed using regres-
sion (as we attempted), or by ranking sites into
toxicity categories and comparing the demo-
graphics between them. Certainly using level
of toxicity is an avenue that should be explored
more thoroughly in the future, when it can be
better measured.
Given the ease of calculation, the interpret-
ability of the numbers, and the high correlation
to other measures, we conclude that test 1 is the
best measure, and the normalized composite is
the best candidate for a preliminary standard-
ized EJ indicator. Test 2b is also a valuable quo-
tient, but it measures a different aspect of EJ.
Where test 1 captures comparative risk, test 2b
reveals the relative burden a group bears. We
feel the latter is less useful than test 1’s informa-
tion about the likelihood that each demo-
graphic group will be located in an at-risk area.
While we recommend the adoption of our
CERI as a standardized indicator, investigators
may have a specific research question that re-
quires some other index. Our results can be
used to clarify the strengths and relationships
between other EJ indices and the standardized
measure we propose. Note again that this mea-
sure does not adequately measure risk, but is a
preliminary indicator that can be used to com-
pare places. By substituting, for toxic sites,
natural hazards that have a known location,
transportation corridors, or environmental ame-
nities such as parks or open space, the same in-
dex can be expanded to measure environmental
justice in other dimensions. It is our hope that
this index will have a direct impact on commu-
nity groups and government agencies by mak-
ing a contribution towards building sustainable
communities in poor and minority neighbor-
hoods in U.S. cities and providing benchmarks
for detailed, site-specific analysis. ꢀ
2 Poverty level is calculated annually by the Census
Bureau based on a ratio of income to costs of an eco-
nomic food plan. Number of family members is in-
cluded in the calculation to determine the total
number of persons below poverty level.
3 We subtract p from 1.0 so that this index “reads” in
a comparable form—i.e., a higher number means
more inequality.
Literature Cited
Anderton, Douglas L. 1996. Methodological issues
in the spatiotemporal analysis of environmental
equity. Social Science Quarterly 77 (3): 508–15.
Boone, Christopher G., and Ali Modarres. 1999.
Creating a toxic neighborhood in Los Angeles
County: A historical examination of environmen-
tal inequity. Urban Affairs Review 35 (2): 163–87.
Bowen, William M. 1999. Comments on “‘Every
Breath You Take . . .’: The demographics of toxic
air releases in Southern California.” Economic De-
velopment Quarterly 13 (2): 124–34.
Bullard, Robert D. 1996. Environmental justice: It’s
more than waste facility siting. Social Science Quar-
terly 77 (3): 493–99.
Chakraborty, Jayajit, and Marc P. Armstrong. 1997.
Exploring the use of buffer analysis for the identi-
fication of impacted areas in environmental equity
assessment. Cartography and Geographic Informa-
tion Systems 24 (3): 145–57.
Cutter, Susan L. 1995. Race, class, and environmen-
tal justice. Progress in Human Geography 19 (1):
111–22.
Cutter, Susan L., Michael E. Hodgson, and Kirstin
Dow. 2001. Subsidized inequities: The spatial pat-
terning of environmental risks and federally as-
sisted housing. Urban Geography 22 (1): 29–53.
Cutter, Susan L., Danika Holm, and Lloyd Clark.
1996. The role of scale in monitoring environ-
mental justice. Risk Analysis 16 (4): 517–26.
Environmental Protection Agency (EPA). 1989. EPA
risk screening guide. Vol. 1, The process. U.S. EPA 5
60/2–89–002, Pesticides and Toxic Substances.
Washington, DC: Federal Printing Office.
———. 1994. Computer-assisted environmental justice
methodology. Dallas: Office of Planning and Analy-
sis, U.S. EPA Region 6.
———. 1999. Risk-screening environmental indicators.
CD-ROM, version 1.0. Washington, DC: Office
of Pollution Prevention and Toxics.
———. 2001. http://www.epa.gov/swerosps/ej/
index.html (last accessed 25 March 2002).
Glickman, Theodore S. 1994. Measuring environ-
mental equity with geographic information sys-
tems. Renewable Resources Journal Autumn:17–21.
Notes
1 EJ analyses typically measure inequity for minori-
ties and/or the poor. We initially wanted to develop

Urban Environmental Justice Indices
331
Holifield, Ryan. 2001. Defining environmental jus-
tice and environmental racism. Urban Geography
22 (1): 78–90.
Lester, James P., and David. W. Allen. 1999. Envi-
ronmental justice in the U.S.: Myths and realities.
Paper read at the 1999 Western Political Science
Association Annual Meeting, Seattle, WA, 25–27
March.
McMaster, Robert B., Helga Leitner, and Eric Shep-
pard. 1997. GIS-based environmental equity and
risk assessment: Methodological problems and
prospects. Cartography and Geographic Information
Systems 24 (3): 172–89.
Mohai, Paul. 1995. The demographics of dumping
revisited: Examining the impact of alternate meth-
odologies in environmental justice research. Vir-
ginia Environmental Law Journal 13 (4): 615–53.
———. 1996. Environmental justice or analytic jus-
tice? Reexamining historical hazardous waste
landfill siting patterns in metropolitan Texas. So-
cial Science Quarterly 77 (3): 500–7.
Mohai, Paul, and Bunyan Bryant. 1992. Race and the in-
cidence of environmental hazards. Boulder: Westview.
Pulido, Laura. 2000. Rethinking environmental rac-
ism: White privilege and urban development in
southern California. Annals of the Association of
American Geographers 90 (1): 12–40.
Scott, Michael S., and Susan L. Cutter. 1997. Using
relative risk indicators to disclose toxic hazard in-
formation in communities. Cartography and Geo-
graphic Information Systems 24 (3): 158–71.
Sheppard, Eric, Helga Leitner, Robert B. McMaster,
and Hongguo Tian. 1999. GIS-based measures of
environmental equity: Exploring their sensitivity
and significance. Journal of Exposure Analysis and
Environmental Epidemiology 9:18–28.
Stockwell, John R., Jerome W. Sorensen, James W.
Eckert, Jr., and Edward M. Carreras. 1993. The
U.S. EPA geographic information system for
mapping environmental releases of Toxic Chemi-
cal Release Inventory (TRI) chemicals. Risk Anal-
ysis 13 (2): 155–64.
Szasz, Andrew, and Michael Meuser. 1997. Environ-
mental inequalities: Literature review and propos-
als for new directions in research and theory. Current
Sociology 45 (3): 99–120.
———. 2000. Environmental geographics data for GIS,
Adams, Arapahoe, Denver, Douglas, and Jefferson
Counties, Colorado. San Diego: VISTA Information
Solutions, Inc.
Warner, Kee. 2002. Linking local sustainability initi-
atives with environmental justice. Local Environ-
ment 7 (1): 35–47.
Weinberg, Adam S. 1998. The environmental justice
debate: New agendas for a third generation of
research. Society and Natural Resources 11:605–14.
Williams, Robert W. 1999. The contested terrain of
environmental justice research: Community as
unit of analysis. The Social Science Journal 36 (2):
313–28.
Yandle, Tracy, and Dudley Burton. 1996. Reexamin-
ing environmental justice: A statistical analysis of
historical hazardous waste landfill siting patterns
in metropolitan Texas. Social Science Quarterly 77
(3): 477–92.
Zimmerman, R. 1994. Issues of classification in envi-
ronmental equity: How we manage is how we
measure. Fordham Urban Law Journal 29 (3): 633–
69.
JOHN HARNER is an Assistant Professor in the
Department of Geography and Environmental
Studies at the University of Colorado at Colorado
Springs, Colorado Springs, CO 80918. E-mail:
jharner@uccs.edu. His research interests are in urban
landscapes, GIS, and U.S.-Mexico interactions.
KEE WARNER is an Associate Professor in the De-
partment of Sociology at the University of Colorado
at Colorado Springs, Colorado Springs, CO 80918.
E-mail: kwarner@uccs.edu. His research interests
are in urban political economy, environmental jus-
tice, and sustainable development.
JOHN PIERCE is Vice Chancellor for Academic
Affairs Emeritus and Professor Emeritus of the Po-
litical Science Department at the University of
Colorado at Colorado Springs, Colorado Springs, CO
80918. E-mail: jpierce@uccs.edu. His research inter-
ests focus on the impact of social capital on the qual-
ity of urban public life.
United Church of Christ. 1987. Toxic wastes and race
in the United States. New York: United Church of
Christ, Commission for Racial Justice.
U.S. Bureau of the Census. 1990. Census of population
and housing, STF-3, census block groups. Washing-
ton, DC: Department of Commerce.
VISTA. 1998. Environmental geographics data for GIS,
Pueblo and El Paso Counties, Colorado. San Diego:
VISTA Information Solutions, Inc.
TOM HUBER is a Professor in the Department of
Geography and Environmental Studies and a Presi-
dential Teaching Scholar at the University of Colo-
rado at Colorado Springs, Colorado Springs, CO
80918. E-mail: thuber@uccs.edu. His research inter-
ests are in remote sensing, Colorado issues, and envi-
ronmental problems.