You are on page 1of 33

Knrivonment and Planning A 200l, volume 33, pages 955 –

973

DO¦:10.1068/a33137

A GIS – environmental justice analysis of particulate air


pollution in Hamilton, Canada

Michael Jerrett
School of Geography and Geology, and McMaster Institute of Environment and
Health, McMaster University, l280 Main Street, West Hamilton, Ontario, Canada L8S
4kl;
e-mail: jerrettm@mcmaster.ca
Richard T Burnett
Health Canada, 200 Environmental Health Centre, Health Canada, †unney’s Pasture, Ottawa,
Canada klA 0L2; e-mail: rick burnett@hc-sc.gc.ca
Pavlos Kanaroglou, John Eyles, Norm Finkelstein, Chris Giovis
School of Geography and Geology, and McMaster Institute of Environment and
Health, McMaster University, l280 Main Street, West Hamilton, Ontario, Canada L8S
4kl;
e.m ail: pavlos@mcmaster.ca, eyles@mcmaster.ca, finkeln@mcmaster.ca,
giovisc@mcmaster.ca
Jeffrey R Brook
Air Quality Processes Research Division, Meteorological Service of Canada, 4905 Dufferin
Street,
†oronto, Ontario, Canada M3H 5†4; e-mail: Jeff.Brook@ec.gc.ca
Received ll September 2000; in revised form l8 January 200l

Abxtract. The authors address two research questions: (1) Are populations with lower socioeconomic
status, compared with people of higher socioeconomic status, more likely to be exposed to higher
levels of particulate air pollution in Hamilton, Ontario, Canada? (2) How sensitive is the association
between levels of particulate air pollution and socioeconomic status to specification of exposure
estimates or statistical models? Total suspended particulate (TSP) data from the twenty-three monitor-
ing stations in Hamilton (1985 – 94) were interpolated with a universal kriging procedure to develop an
estimate of likely pollution values across the city based on annual geometric means and extreme events.
Comparing the highest with the lowest exposure zones, the interpolated surfaces showed more than a
twofold increase in TSP concentrations and more than a twentyfold difference in the probability of
exposure to extreme events. Exposure estimates were related to socioeconomic and demographic data
from census tract areas by using ordinary least squares and simultaneous autoregressive (SAR) models.
Control for spatial autocorrelation in the SAR models allowed for tests of how robust specific
socioeconomic variables were for predicting pollution exposure. Dwelling values were significantly
and negatively associated with pollution exposure, a result robust to the method of statistical analysis.
Low income and unemployment were also significant predictors of exposure, although results varied
depending on the method of analysis. Relatively minor changes in the statistical models altered the
significant variables. This result emphasizes the value of geographical information systems (GIS)
and spatial statistical techniques in modelling exposure. The result also shows the importance of
taking spatial autocorrelation into account in future justice – health studies.

Introduction
Environmental justice has emerged as an important dimension of
environmental and public health policy in North America, especially in the United
States. In l994 President Clinton issued an executive order mandating federal
agencies to consider the environmental justice consequences of their decisions
(that is, whether poor or minority groups bear disproportionately high exposure to
environmental contaminants and potential health risks), and the US
Environmental Protection Agency (EPA) subsequently established an office to
conduct justice analyses (Bowen et al, l995; Glickman, l994). With the
exception of an early paper (Handy, l977), interest in environmental justice in
Knrivonment and Planning A 200l, volume 33, pages 955 –
Canada
973 arose only recently (Jerrett et al, l997). Govern- ment-sponsored
initiatives suggest policymaking and public interest groups may have a growing
interest in this issue (Jerrett, l999).
956 M Jerrett, R † Burnett, P kanaroglou, J Eyles, and
coworkers

Exposure assessment is often a challenging aspect of environmental justice


research, and the literature to date has seen progressive advances toward removing
uncertainties associated with exposure misclassification (Sexton and Adgate,
l999). Problems with exposure assessment and subsequent uncertainty about
statistical results of justice studies have hindered policymakers in their attempts to
formulate an appropriate response to what many perceive as an important public
health issue (Institute of Medicine, l999).
In this paper we use geographic information systems (GIS) and related
geostatis- tical techniques in combination with fixed-site air-pollution monitors to
improve the exposure estimates for justice and health analysis. We address two
central research questions focused on supplying policy guidance and resolving
methodological issues. First, are populations with low socioeconomic status more
likely to be exposed to high levels of particular air pollution in Hamilton, Ontario,
Canada? Second, how sensitive is the association between levels of particulate air
pollution and socioeconomic status to specification of exposure estimates or
statistical models? †o address these questions, we pursue three objectives. First, we
review recent literature to frame the justice issue with reference to the problem
of particulate air pollution in the City of Hamilton. Second, with use of GIS and
ten years of data from twenty-three monitoring stations on total suspended
particulate (†SP) concentrations, we identify areas in Hamilton most likely to
exceed the Ontario Ministry of the Environment (MOE) air-quality objectives for
this pollutant. †hird, we combine the pollution-exposure estimates with
socioeconomic status and demographic data in multiple ordinary and spatial regression
models to assess whether populations with lower socioeconomic status or
recent immigration status are exposed to higher levels of air pollution. We
conclude with a discussion of the implications of the results for future
environmental justice research and government air-pollution monitoring
programs.
Enrironmental juxtice ax an interpretire frame
Environmental justice can be viewed as an interpretive frame or conceptual
construct for understanding environmental problems (Perrolle, l993). Broadly defined,
the frame includes the concepts of environmental equity, equality, and racism. It is
simultane- ously defined from the bottom up— as grass-roots organizations
discover a pattern to their grievances— and from the top down— as leading or
‘peak’ environmental organi- zations convey the justice idea to these groups
(Capek, l993). It incorporates many preexisting ideas such as economic
externalities (Baumol and Oates, l988), unequal political power among different
groups (Hamilton, l993), and the ability of powerful groups to shape and use
institutions affecting land use to their benefit (Bullard, l990; Pulido, 2000). As seen in
figure l, environmental justice influences both the geograph- ical location of
polluting facilities (Hamilton, l993; McMaster et al, l997) and the segregation of
racial and socioeconomic groups over space (Fong, l994). Factors such as
(political) power, externalities (market failures), and land-use institutions coalesce
to determine the spatial distribution of pollution and other hazards in relation to
specific demographic and socioeconomic groups (for example, the poor). An
unequal or inequitable distribution of pollution is also important in assessing
whether such exposure may result in health effects when other known
determinants of health are taken into account (Nyergis et al, l997). In addition,
health effects from exposure to environmental pollutants can be identified. †he
discovery of health effects may result in policy action aimed at promoting a more
equitable sharing of the distribution of pollution and other environmental
956 M Jerrett, R † Burnett, P kanaroglou, J Eyles, and
coworkers

hazards or at least to be more diligent in the health surveillance of high-risk


populations. A complementary approach is depicted at the bottom of figure l, for a
potential health risk (for example, air pollution), representing
A GIS – environmental justice analysis of particulate air 957
pollution

Environmental
justice as a
conceptual
construct

Market Political Institutional


Research failures inequality structures
context among
(externalities) for land-use
groups decisions

Geographical Spatial
segregation of
siting of socioeconomic
and racial
groups
polluting
facilities

Spatial inequality
Research focus in pollution
exposure

Other health
determinants:
●genetic predisposition
Related research ●diet
●lifestyle
●occupation
●social position

Health effects of pollution


Complementary research

Figure 1. †he environmental justice framework.


958 M Jerrett, R † Burnett, P kanaroglou, J Eyles, and
coworkers

an attempt to establish a link between the spatial and temporal dimensions of


health effects and groups who live in polluted areas. If health risks are found,
there is reason to question factors that have influenced the segregation of specific
income and racial groups, the location of polluting facilities, and the enforcement
of environ- mental protection laws. In this paper we focus on a single component of
figure l, that is, establishing whether inequity (defined as inequality in exposure)
exists in the distribu- tion of particulate air pollution in Hamilton.
Erolution in expoxure axxexxmentx in enrironmental juxtice rexearch
Many studies of the spatial distribution of negative externalities have been
undertaken (for a recent review, see McMaster et al, l997). Yet the results remain
equivocal, with some researchers finding no association or a negative one between
the presence of low- income populations or racial minorities and the presence of
polluting or potentially hazardous facilities (Anderton et al, l994; Bowen et al, l995;
Jerrett et al, l997). †hese equivocal findings probably arise because of the complex
methodological problems that confront researchers interested in environmental
justice (for reviews, see Baumol and Oates, l988; Mohai, l995). †he methods used
to estimate potential exposure in vulnerable, poor, and minority populations
represent a major challenge to current research (Sexton and Agate, l999). Most
studies have used proximity-based assessments where individuals in a given area
(usually a census tract, postal code, or county) are assumed to receive the same
level of exposure based on the presence or absence of polluting facilities
(Greenberg, l993). Some researchers have buffered facilities with concentric
impact zones and have estimated daytime and nighttime exposures to assess the
potential impact on the findings (Glickman, l994; Sheppard et al, l999). Others
have used toxicity indices in combination with EPA’s †oxic Release Index (†RI)
to weigh the potential harm caused by the pollutants to proximate populations
(Bowen et al, l995). Earlier studies used simple interpolation techniques to establish
the spatial dimensions of air pollution in relation to vulnerable population groups,
although these estimates were based on few sampling points (Earickson and Bullick,
l988). Chakraborty and Armstrong (l997) used plume dispersion modelling to
assess the potential for exposure based on different wind directions and other
assumptions about local con- ditions. †he found larger proportions of nonwhites
and individuals below the poverty line resided within buffers areas determined by
plume modelling than within concen- tric or adjacency buffers. †his finding
suggests an accurate assessment of the spatial dimensions of the physical
pollution phenomenon can influence both the exposure estimates in and the
results of justice studies.
†he studies cited above have done much to improve exposure assessments
and illuminate the empirical dimensions of environmental justice. Yet none of
the pre- vious research has used monitored data, modelled with an optimal
interpolation estimator, to assess the equity characteristics of potential exposure.
Where data are available, advances in GIS technologies and associated modelling
programs offer the potential of improving the exposure estimates used to link
environmental pollution to socioeconomic status or health outcomes.
Enrironmental juxtice and air pollution in Hamilton
Hamilton, Ontario, provides an excellent study site for three reasons. First, the city
has a monitoring network capable of supporting optimal interpolation. Until
funding cut- backs by the provincial government in l996, Hamilton possessed
one of the most comprehensive intraurban air-pollution monitoring networks
in North America. Second, Hamilton has a major industrial area that houses
Canada’s two largest steel- makers, resulting in zones with high levels of pollution
958 M Jerrett, R † Burnett, P kanaroglou, J Eyles, and
coworkers

in the northeast and, to a lesser extent, in the central western parts of the city. †his
spatial concentration of potential
A GIS – environmental justice analysis of particulate air 959
pollution

impacts makes equity analyses more tenable than they are in places with many
impact zones. †hird, Hamilton experiences serious ambient-air-pollution problems
that time- series studies suggest are associated with an estimated 90 to 32l premature
deaths per year (HAQI, l997). †hese time-series studies involved the regression
of mortality counts onto pollution values from fixed-site monitors, and,
because of the many methodological challenges, the mortality estimates from
these models must be viewed with caution. keeping this scientific uncertainty
about health effects in mind, the mone- tary cost of illness and death associated
with particulate air pollution in the region is in the range of Can$537 million per
year.(l)
Earlier work on the spatial distribution of air pollution in Hamilton suggested
the presence of inequalities in air-pollution exposure, with the poor being exposed
to higher levels of ambient pollution (Handy, l977). †he question of whether the
reported inequities resulted from actual correlations or from inaccurately measured
pollution exposure, based on isotropic proximity, remains unanswered. Given the
significant air- pollution problem that exists in the city (HAQI, l997), answering this
question could provide important guidance to policymakers.

Data and methodx


†his analysis relies on a GIS environmental health database for Hamilton. All
data were cleaned and processed into a relational GIS database by using
ArcInfo 7.2 and ArcView 3.2 software (ESRI Inc., Redlands, CA). Statistical
analyses were performed with use of the S-Plus 2000 (Mathsoft Inc.,
Cambridge,MA)
‡ spatial statistics module, GS 2.l (Gamma Design Software,
Plainwell, MI), SPSS l0 (SPSS Inc., Chicago, IL), and Minitab l2 (Minitab Inc.,
University Park, PA) software packages. Pollution estimates were based on ten
years (l985 – 94) of particulate air-pollution data from twenty-three monitoring
stations in the Hamilton area operated by MOE (Dobroff, l988; l992; l993; l994;
l996; personal communications, l998 – 99). [For a site map of Hamilton and the
†SP monitoring network, see figure 2 (over)]. Socioeconomic and demographic
data were drawn from the l99l Census of Canada (Statistics Canada, l99l).
High-exposure zones were based on the current MOE provincial ambient-air-
—3 —
quality objective for †SP, namely, a 60 µg m annual geometric mean, and l20 µg m
per 24- hour criterion. Suspended particulates are air-borne particles usually less
3

than 50 µm in aerodynamic diameter. Particles less than l0 µm in size (PMl0)


penetrate the human upper respiratory tract, and those less than 2.5 µm in size
(PM2.5) can penetrate deep into the alveoli of the lungs. Both of these smaller
types of particles have been associated in time-series studies with increased
incidence of adverse health outcomes and mortality (Dockery and Pope, l997). A
recent study in Hamilton found suspended particulates were associated with about
95% of the variation in PMl0 levels (kim and Jerrett, 2000). Each microgram
increase in ambient †SP concentration was associated with a 0.3l – 0.39 µg increase
in PMl0 levels. †he difference depends on the location of the monitoring site within
the city. †he overall mean ratio of PMl0 to †SP is 0.37, which is low compared
with other Canadian cities where the mean ratio is 0.44 (Brook et al, l997). †he
MOE (Dobroff, l997) found similar results, with a PMl0 /†SP ratio of 0.30 for the
Hamilton beach strip downwind of the industrial core, and 0.40 for the rest of the
city (Dobroff, l994; l997). †hus, despite some intraregional variation, readings of
†SPs provide a reasonable proxy for the more damaging smaller particles.
†he MOE monitors †SP on a six-day North American synoptic cycle,
provid- ing about sixty-one data points per year. Areas known to experience
†SP levels
A GIS – environmental justice analysis of particulate air 959
pollution

(l)
Cost estimates were derived using a program developed for the Ontario Medical
Assocation, available at www.oma.org/phea¦th/smogmain.htm
96
0

M
Jer
re
tt,
R

Bu
rn
et
t,
P
ka
na
ro
†SP sampling stations gl
2 0 2 4
ou
km
,J
Ey
Figure 2. Hamilton site map and total suspended particulate (†SP) sampling station network
le
s,
A GIS – environmental justice analysis of particulate air 96l
pollution

above the MOE objective— referred to as nonattainment areas— exhibit the


following characteristics: (l) high population density; (2) proximity to either the
major indus- trial area of the city (in north central and northeast Hamilton) or
the peripheral industrial area(s) in the west end of the central city; and (3)
experience regular lake- induced advection inversions that produce a
fumigation plume between the late winter and the early summer (Dobroff,
l992; l996; Farhang, l983; Pengelly et al, l984; Rouse and Burghardt, l987).
†hus there is potential for high exposures to affect large residential populations for
a significant portion of the year.
†wo methods of exposure assessment were employed. One method focused on
chronic average exposure and the other on the probability of extreme events. In
the first method, we used ten years of annual geometric mean concentration data
published by the MOE. We then calculated a ten-year geometric mean from the
annual geometric means. Although there is some debate over the merits of using
geometric means for measuring ambient air pollution (see Parkhurst, l998), this
remains the accepted practice for dealing with pollution concentrations that are
highly variable with a low average and large range between near-zero values and
maxima during episodic events (Colls, l997). More importantly for this analysis, use
of the geographic mean allows for a direct linkage to the MOE air-quality objective. A
comparison of the arithmetic mean and the geometric mean for the ten years of
—3
data showed the geometric mean to be about 0.5 µg m lower on average,
based on the twenty-three sites. †his method produces a ten-year average †SP
concentration and allows for a conservative estimate of those areas likely to exceed
—3
the MOE’s criterion of 60 µg m annual exposure. †he resulting pollution surface
proxies chronic exposure to air pollution above the suggested government
objectives.
For our second exposure assessment we estimated the probability of extreme events.
†his involved identifying all records with a value equal to or greater than the
—3
MOE’s daily objective of l20 µg m . For each station, we then calculated the
proportion of the total monitoring days in which the station reading exceeded the
MOE objective.
†hese values represent the probability of each station reading exceeding the
MOE objective. †he resulting surfaces can be interpreted as the probability of a
given area exceeding the 24-hour regulatory objective and thus estimates the
probability of exposure to extreme events.
Both estimates relied on universal kriging to generate the pollution surfaces
that formed the basis of our exposure estimates. kriging is a geostatistical
technique that estimates values of the variable of interest for points that are
between sampling sites. It does this by means of spatial interpolation and by
taking account of spatial dependence. Spatial dependence is the property of
observations over space to exhibit functional similarity as a result of proximity
(for reviews of kriging, see Bailey and Gatrell, l995; Burrough and McDonnell,
l998). kriging models are considered optimal interpolators because they supply the
best linear unbiased estimate of the value of the variable at any point in the
coverage (Burrough and McDonnell, l998). †he universal method of kriging was
chosen, as opposed to the ordinary method, because of prior evidence suggesting
the existence of a spatial trend in the intraurban distribution of
†SP in Hamilton (Farhang, l983; Pengelly et al, l984). Universal kriging is designed
to take account of the trend in variables over the interpolated ‡area. GS 2.l
geostatistical software was used to generate the interpolation estimates. †he
resulting maps were displayed in ArcView 3.2.
†he kriging procedure produces an optimal interpolation surface, but we
A GIS – environmental justice analysis of particulate air 96l
pollution

must emphasize that the monitoring network, although better than most, still
has areas with sparse data. †his is particularly the case in the southern areas
of the city, and estimates derived from these sparse data will be subject to
larger standard errors.
962 M Jerrett, R † Burnett, P kanaroglou, J Eyles, and
coworkers

†he pollution surfaces generated from these procedures were overlayed on the
centroids of census tracts (C†s) from the l986 Census of Canada (N = 90). Owing
to missing data and lack of pollution monitors, five C†s were excluded, leaving
eighty five for analysis. †he l986 boundaries were used with l99l Census data
to allow for future research comparing the temporal aspects of equity and air
pollution in Hamilton. †o perform the overlay, the pollution surfaces were
classified into one-unit isobands (either micrograms or percentages). Each
isoband was converted into a separate polygon for a polygon-on-point overlay
operation. †his overlay combined pollution and socioeconomic data to address the
question of whether populations with lower incomes and other socioeconomic
characteristics are more likely to live in high-pollution zones than are other
population groups.
Some of the C† units had large areas with no residential population. In these
cases, the geographic (or geometric) centroids of the C† units may not accurately
represent the centre of the residential population. †o address this problem, we
calculated a popula- tion-weighted centroid by using a point coverage of the
housing stock. In particular, we overlayed the housing stock on each C†, created a
new point coverage for each C†, and then calculated the median value of the
universal transverse Mercator (U†M) easting and northing for the housing-stock
point coverage, measured in metres, to arrive at a new centroid based on the
residential population of each C†. In the downtown areas, this correction made
little difference, but in some areas near the industrial core and in the southern
suburbs the corrections were fairly large (around l.4 km). Even with this refinement
in the pollution estimate, we are using point estimates to represent larger areas of
residential population. †hus, some exposure misclassification will be intro- duced
as a result of treating an areal feature as a point. Mismatch in the spatial scale of
exposure and the units of analysis often plagues exposure estimates in
environmental justice research (Cutter et al, l996).
We used multiple ordinary least squares (OLS) and simultaneous autoregressive
(SAR) models to assess whether populations with lower socioeconomic status have
a greater likelihood of living in areas with high levels of air pollution than have
pop- ulations with higher socioeconomic status. Here, regression models are used
not to infer causality in the relationships but to test for associations between
pollution exposure and relevant socioeconomic variables while controlling for the
confounding effect of other variables known to affect pollution exposure such as
manufacturing employment (for a general discussion of interpreting regression models
where causality is not inferred, see Cook and Weisberg, l999). As discussed by Hill
(l965), causality in health and environmental research is a complex concept that
is rarely, if ever, addressed through one study. We emphasize that the dependent
variable used repre- sents cuvvent exporuve because our kriging models estimate
potential ambient concen- trations that have been generated by industrial and
transportation processes or exacerbated by meteorological processes that have
little to do with the socioeconomic and demographic characteristics of the C†s. In
other words, we are not positing that poverty or other socioeconomic variables
cause pollution. Instead, in the context of the conceptual model presented in figure
l, we suggest that the combination of land-use decisions, industrial location, and
uncontrolled external costs over time may lead certain demographic groups to
live in areas of higher exposure than other groups. Studies that investigate these
longer term causal mechanisms are usually called ‘process’ equity or justice studies.
†hose that test for the possibility of environmental inequity in the present are
referred to as ‘outcome’ studies (see Cutter, l995; Pijawka et al, l998). In this paper
962 M Jerrett, R † Burnett, P kanaroglou, J Eyles, and
coworkers

we investigate outcomes, and we have selected the regression model because it allows
for a more thorough assessment of potential relationships, better control for
A GIS – environmental justice analysis of particulate air 963
pollution

confounders, and diagnostics for statistical problems resulting from collinearity and
spatial autocorrelation.
Most of the predictor variables used have been found to be significant in
earlier justice studies (for example, Anderton et al, l994; Bowen et al, l995;
Chakraborty and Armstrong, l997; Jerrett et al, l997; McMaster et al, l997). We
included a previously untested education variable to assess the hypothesis that,
even after controlling for economic inequity and insecurity, educational status would
exert an independent health risk. †his is based on the expectation that persons of
lower educational status may misperceive the health risks from pollution more than
highly educated individuals. Literature on the health effects of education suggests
persons with low education are more likely to engage in risky behaviour such as
smoking (Birch et al, 2000), and it is possible that risky behaviour carries over
into environmental exposure. We also included an unemployment variable to
proxy for economic insecurity and stability. Numerous health studies have shown
that unemployment tends to be associated with unhealthy lifestyle behaviours such
as excessive consumption of alcohol (for example, see Catalano et al, l993) and
subsequent adverse health outcomes. Recent studies have suggested that
socioeconomically disadvantaged groups face a ‘double jeopardy’ because they
already experience increased health risks from such things as unemploy- ment or
poor nutrition, and they live in areas of high exposure to pollution or other
environmental risks (Institute of Medicine, l999). Variables such as
unemployment therefore appear to be important to include in justice studies.
Specifications and descriptive statistics for the variables used in the regression
analyses are given in table l. Based on past research, we would expect to see
exposure increase as median household income and educational levels decrease
and as the unemployment and low-income levels increase. †he manufacturing
employment variable controls for the potential association between work in the
industrial core and the increased likelihood of a person living near the industrial
plants. Previous studies have found positive associations between the level of
manufacturing employment and level of environ- mental exposure (Anderton et al,
l994; Jerrett et al, l997). Failure to control for this
Table 1. Variable specification and descriptive statistics (source: Statistics Canada, l99l).

Mean Stand Minimu Maximu Ran


ard m m ge
deviati
on
TSP exposure, Y1 50.3 10.9 31.0 77.0 46.0
TSP probability, Y2 6.4 5.0 1.0 22.0 21.0
Median income, X1 36 083 10 037 12 771 60 840 48
069
Low income, X2 22.4 11.1 5.9 53.1 47.1
Dwelling value, X3 154 235 32 563 91 680 258 389 166
709
Unemployment rate, X4 11.4 4.8 4.7 27.7 22.9
Manufacturing, X5 23.2 5.8 11.3 36.8 25.6
Education, X6 36.2 8.8 14.2 55.1 40.9
Immigrants, X7 2.7 3.7 0.0 23.9 23.9
Note„ TSP is total suspended
—3
particulate; Y1 , a variable representing exposure to suspended
particulates (in µg m ), based on a ten-year mean of the geometric means published by the
Ontario Ministry of Environment (MOE, 1985 − 94); —3
Y2 , a variable representing the proportion
of total measurement days that exceed 120 µg m over a 24-hour average (1985 − 94); 1 X,a
vari- able based on median household income; X2 , a variable based on Statistics Canada’s
‘low-income’ classification; X3 , a variable based on dwelling value; X4 , a variable based on the
A GIS – environmental justice analysis of particulate air 963
pollution

percentage of unemployment; X5 , a variable based on the percentage of manufacturing


occupation; X6 , a variable based on the percentage of persons with no high-school education;
X7 , a variable based on the percentage of recent immigrants (past three years).
964 M Jerrett, R † Burnett, P kanaroglou, J Eyles, and
coworkers

variable could confound relationships between pollution exposure and


socioeconomic status. In l99l, the Census of Canada did not include explicit race
variables. †hus we have used a variable on recent immigration to give some
level of comparability with past US research that has focused on race. A large
proportion of recent immigrants to Canada are from visible minorities (Avery,
2000; kazemipur and Halli, 2000). We would therefore hypothesize a positive
relationship between the proportion of recent immigrants and pollution
exposure.
Variables were analysed and transformed to their closest approximation of
the normal or Gaussian distribution before the models were run. Model
selection pro- ceeded in five stages. First, we ran each variable in bivariate
models to assess the potential strength and shape of the relationships.
Second, we ran a model with all the significant variables together. †hird, we ran
substantively selected subsets (that is, one with variables representing
socioeconomic status, one with occupation status, and one with immigration
status). Fourth, we combined the dwelling value variable, which appeared to exert
the most consistent effect on exposure, with each of the other significant
variables in multivariate models. Last, we ran the selected models with
untransformed independent variables, and, if the results remained similar, we used
the models with the raw variables to ease interpretation. Standard diagnostic
tests were performed to ensure our models conformed reasonably with the
underlying assumptions of the regression analysis.
Based on the results of diagnostic tests for spatial autocorrelation in the
residuals, we also employed SAR models with a first-order nearest-neighbour
connectivity matrix to test the associations. †he SAR models simultaneously
estimate an autocorrelation parameter to derive an error term that takes
autocorrelation into account and thus meets the independent, identically
distributed assumptions about the error terms. As Odland (l988) discusses, this
formulation of the model provides a powerful framework for analyzing spatial
relationships.

Rexultx
Figure 3 show the pollution isopleths generated from the probability of extreme
events, from a regional perspective. As shown, some areas of the city near the
industrial core exceed MOE objectives more than 25% of the time, whereas
those in the distant western part of the city exceed the objective less than 5%
of the time. Regionally, most of the areas of nonattainment are in the north
central parts of the city, close to the industrial core. Figure 4 (see over) shows
the annual average pollution zone obtained with the dwelling value variable
used in the regression models. Figure 4 suggests a strong relationship
between dwelling value and †SP pollution.
†able 2 (see over) shows the zero-order correlation coefficients for all the
variables. From this table we see that dwelling value and low income have the
strongest unad- justed correlations with pollution exposure. But some of the
independent variables have high intercorrelations— in particular, low income
with unemployment and median income (v > 0.8). †his suggests that it will be
difficult to disentangle the independent effects of the collinear variables.
Results of the regression modelling are shown in table 3 (see over). †hese
results show that dwelling value, unemployment, and low income exert significant
effects on pollution exposure. All variables take the expected sign. †hese results
are consistent across both exposure estimates but vary with the model
964 M Jerrett, R † Burnett, P kanaroglou, J Eyles, and
coworkers

specification. In OLS models, dwelling value and low income are significant, but in
the SAR models the low-income variable is replaced by the unemployment variable.
Significant positive autocorrela- tion in the residuals of the OLS models was
found with Moran’s I test (see table 3).
A
GI
S

en
vir
on
m
en
tal
ju
sti
ce
an
al
ysi
s
of
pa
rti
cu

†SP exceedance time (%)


†SP sampling station

2 0 2 4 km

Figure—3. Percentage of time of sampling period (l985 – 94) in excess of 24-hour total suspended particulate (†SP) objective (l20 96
µg m ).
3

5
96
6

M
Jer
re
tt,
R

Bu
rn
et
t,
†SP level (µg m—3 ) P
Dwelling value
9l680 – l30 980 l30 98l – ka
l78 460 l78 46l– 258 389
2 0 2 4 km na
ro
gl
Figure 4. Dwelling values (X3 , see table l) and total suspended particulate (†SP) levels, l985 – ou
94. ,J
Ey
le
s,
Table 2. Zero-order correlation coefficients, with p values (in parentheses). A
GI
TSP TSP Media Low Dwellin Unemployme Manufacturin Educati Immigrant
S
exposur probabilit n incom g nt rate, g on s
e, y, incom e, value, X5 X6 X7 –
e, en
Y1 Y2 X1 X2 X3 X4 vir
on
TSP exposure, Y1 1.000 0.866 —0.537 0.569 0.621 0.541 0.160 0.476 0.030
(0.000) (0.000) (0.000) (0.000) (0.000) (0.144) (0.000) (0.784) m
TSP probability, Y2 1.000 —0.492 0.561 0.536 0.540 0.030 0.349 0.084 en
(0.000) (0.000) (0.000) (0.000) (0.783) (0.001) (0.445) tal
Median income, X1 1.000 —0.836 0.653 —0.588 —0.041 —0.621 0.419
ju
(0.000) (0.000) —(0.000) (0.707) (0.000) (0.000)
Low income, X2 1.000 —0.569 0.733 0.062 0.523 0.578 sti
(0.000) (0.000) (0.574) (0.000) (0.000) ce
Dwelling value, X3 1.000 —0.607 —0.399 —0.645 0.219 an
(0.000) —(0.000) (0.000) (0.044)
al
Unemployment rate, X4 1.000 0.298 0.618 0.508
(0.006) (0.000) (0.000) ysi
Manufacturing, X5 1.000 0.535 0.119 s
(0.000) (0.280) of
Education, X6 1.000 0.195
(0.073) pa
Immigrants, X7 1.000 rti
Note„ TSP is total suspended particulate; for a full description of the variables, see table 1. cu

96
7
968 M Jerrett, R † Burnett, P kanaroglou, J Eyles, and
coworkers

Table 3. Comparison of regression results from ordinary least squares (OLS) and simultaneous
autoregressive (SAR) models: þ regression coefficients.

Variable OLS SAR OLS SAR TSP


TSP model TSP model exceed exceed
—3 —3
model — model —
3
60 µg m 60 µg m 120 µg m 120 µg m
3

(—4.424)**
—0.439 (—2.614)*
—0.227 (—3.048)**
—0.105
Dwelling value, X3 —
0.191
(—
Low income, X2 0.3 0.379 2.408)*
19
(3.215)* (3.594)**
*
Unemployment rate, X4 a 0.3 0.185
72
(4.734)* (2.383)*
*
Goodness of fitb 0.4 0.28 0.370 0.09
2
b
Adjusted R -statistic. 41
c
p-values are given in parentheses.
Note„ TSP is total suspended particulate; t-statistics are given in parentheses; for a full
description of the variables, see table l.

Positive autocorrelation increases the probability of false positive findings in


standard significance tests (Odland, l988); therefore, the results of the SAR models
are probably more reliable. †he overall fit for the SAR models is not as high as
for the OLS model as measured by the adjusted coefficient of determination,
which suggests that some of the effects observed in the OLS model were a result
of autocorrelation. A likelihood ratio test was applied to the correlation
parameter, q, and this showed it was signifi- cant in the SAR model (p c 0.05).
†his added further evidence to suggest a need to control autocorrelation in the
model. Use of the more conservative likelihood ratio test on the individual
coefficients suggested that only the dwelling value variable signifi- cantly
improved the overall model prediction. When the unemployment variable was
removed from the model, however, the coefficient for the dwelling value
variable increased by more than 20%. Given the results of the t-test and the
previous expect- ation that unemployment represents social status, the
unemployment variable was left in the model. †his finding of a weaker
association with unemployment than with dwelling value combined with a robust
association between the dwelling value variable and the ambient-air-pollution
variable in all models suggests that the dwelling value variable exerts a more
consistent effect than the other variables.
†he coefficient for the dwelling value variable implies that each standard
deviation increase (Can$32 563) in dwelling value is associated with a 0.227
—3
standard deviation, or 2.48 µg m reduction — in †SP× exposure (=0.227 10.93 µg
—3 —
m — 2.48 µg m
3

). Each standard deviation increase × =
in dwelling value reduces the predicted
——3
probability of exceeding l20 µg m by 0.96% ( 0.191 5.03 0.96%). Each

standard deviation × increase
= in the unemployment rate is associated with an
— —3 —3
increase of 4.07 µg m in †SP exposure ( 0.372 10.93 4.07 µg m ). Each
× =
standard deviation increase in unemployment corresponds to a 0.93% increase
in the probability of experiencing extreme events (0.185 5.03 0.93). Although
968 M Jerrett, R † Burnett, P kanaroglou, J Eyles, and
coworkers

interpretation of the coefficients provides a useful guide to the relative


relationships, these estimates must be viewed with caution because of the
possibility of error propagation, which is discussed in more detail below.
A GIS – environmental justice analysis of particulate air 969
pollution

Dixcuxxion and concluxionx


Groups with lower socioeconomic status are exposed to higher levels of
ambient particulate air pollution in Hamilton than are groups with higher
socioeconomic status. Our finding of inequity agrees with earlier research on
Hamilton (Handy, l977) and with much of the US research. †he similarity
between the Canadian and US findings supports the idea that similar societal
processes are operating, with disadvantaged groups who live in low-cost housing
being subjected to higher potential exposure (for an analysis of how the housing
market contributes to environmental injustice, see Been, l994). One exception to
this similarity appears to be with the recent immigration variable. †his variable
was insignificant in all models. Although this is an imperfect measure of race, it
suggests that environmental injustice in Canada may not have the same racial
dimensions as it does in the United States. †he l996 Census of Canada contains
the first race variables, and this appears to be a useful area for future research, as
the race variable will allow for a direct comparison between the two countries.
Dwelling values associate significantly and negatively with pollution exposure.
†his result is robust to the method of analysis. Low income and unemployment
were also significant predictors of exposure, although these results varied depending
on whether the model took spatial autocorrelation into account. †hus, on the
second research question, there is a difference in significant variables among
models. Although there is consistency between the results of chronic and extreme
exposure estimates, the model specification conditions the results. In both SAR
models, which control for spatial autocorrelation, the results are consistent, with
the dwelling-value and unem- ployment variables exerting significant effects on
the exposure variable. OLS models produce results that are consistent between
chronic and extreme exposure estimates, but they differ from the SAR model
results. Much of the current literature on justice has presumed inequity in
exposure exists, based on models that may or may not have adequately controlled
for potentially confounding methodological problems such as spatial
autocorrelation. With a few exceptions, much of the current research has used
proximity-based exposure estimates or modelled estimates based on assumptions
about dispersion conditions that may not hold for all times of the year. †he
relative consistency of the exposure estimates, after the effect of autocorrelation
was taken into account, emphasizes the importance of basing exposure on
measured ambient conditions.
†hese findings have implications for current government policies and programs
aimed at monitoring air pollution. If few cities possess a network capable of
supporting kriging or other types of sophisticated interpolation, then how useful
are such analyses if they cannot be replicated elsewhere? †his question can be
answered in two ways. First, if there are only a few places where such an
analysis can be replicated, then this approach is of limited value. Second, this
analysis suggests that more accurate exposure assessments are available through
GIS and spatial statistics software now widely available. Given the potential
importance of exposure assessment for justice and health studies, governments
may wish to rethink their monitoring strategies. †he trend has been toward fewer
sites with greater ability to detect changes in fine-particle and gas levels over time.
†he expense of these facilities limits the budget available for spatially
disaggregated monitoring. Yet, the twentyfold difference in the probability of
extreme events and the twofold increase in ambient †SP support the idea that
intraregional exposure estimates may be an important source of information for
justice and health studies. Both spatially and temporally disaggregated monitoring
are needed to supply defensible exposure estimates.
A GIS – environmental justice analysis of particulate air 969
pollution

Although we were able to develop reasonable exposure estimates for Hamilton


with straightforward methods, interpolation models will tend to break down in
settings
970 M Jerrett, R † Burnett, P kanaroglou, J Eyles, and
coworkers

with more complex air-emission patterns. Industrial cities such as Hamilton, with a
few dominant sources of pollution, are declining in number, and in many other
places transportation and related nonpoint sources such as shopping malls are
becoming relatively important (Anderson et al, l996). Assessment of exposure in
these complex settings will require more sophisticated models that take into
account multiple sources through detailed emission inventories and
transportation models, monitored concentrations, meteorological conditions, and
the time – activity patterns of residents. Models for accomplishing this type of
integrated assessment have been developed (Novak et al, l995),(2) but are costly
in terms of data collection and computing resources. For urban areas with more
complex emission patterns, future justice and health studies may have to employ
estimates for integrated models.
Another concern that tempers our results arises from the potential for error
propagation. †he pollution surfaces used to proxy for chronic and extreme
event exposures are subject to estimation error. When this error combines with
sampling error from the census and the errors induced by using the centroid
points of residential populations, the propagation of these errors may influence the
results. In particular, the elasticities derived to show the likely change in pollution
exposure for changes in significant variables such as dwelling values are best
treated as general guides on the magnitude of effects rather than as precise
estimates of unit changes.
One final limitation in the exposure estimates deserves mention.
Specifically, evidence is mounting on the predominance of indoor exposure
(Paustenbach, 2000). A link between ambient sources and indoor pollutants
exists (Wallace, l996), but indoor ventilation systems and activities such as
cooking and cigarette smoking can have a significant impact on pollution exposure
(Ott and Roberts, l998). Our analysis of variables associated with ambient
exposure provides only partial evidence on total personal exposure. †his focus
on outdoor exposure and other methodological limitations temper our findings,
yet we can still make some conclusions about the importance of different
variables to ambient †SP exposure.
On the empirical issue of which variables measure potential exposure, it
appears that, in the absence of other information, dwelling value provides a
more reliable indication of potential exposure. Dwelling value is sometimes
seen as a proxy for permanent or lifetime income (Jerrett et al, l997), because
people with lower incomes or with unstable incomes over time are unlikely to
purchase homes that they cannot pay for over the long term. †he reliability and
importance of dwelling value also suggest this variable may contribute to
population health studies that attempt to assess the social and environmental
determinants of health. †hese studies have tended to focus more on deprivation
variables or on more transitory variables such as poverty.
†he unemployment variable raises the spectre of ‘triple jeopardy’ for
socially deprived groups. As noted earlier, the US Institute of Medicine (l999)
suggested that poor and minority groups may experience ‘double jeopardy’, in which
the population suffers both higher burdens of sociobehavioural health risks and
experiences greater exposure to pollution. Impoverished and minority groups
already tend to suffer from poor nutrition, higher exposure to pollutants in the
workplace, and the psychosocial stress of knowing others have more material
wealth and more control over their lives. And, as shown in many studies and
confirmed here, empirical evidence shows these groups often experience higher
exposure to environmental pollutants. A larger liter- ature suggests that both the
unemployment experience and the ‘anticipation phase’ prior to unemployment
970 M Jerrett, R † Burnett, P kanaroglou, J Eyles, and
coworkers

are associated with more risky lifestyle behaviours such as smoking and excess
alcohol consumption (for example, see Catalano et al, l993; Ferrie
(2)
Program information for the M-3 air pollution modelling sysem is available from the
US Environmental Protection Agency at www.epa.gov/asmdner¦/mode¦s3/index.htm¦
A GIS – environmental justice analysis of particulate air 97l
pollution

et al, l995). An interesting question relates to whether this ‘double jeopardy’ interacts
to increase health risks more than the simple additive effect of two or more of
these risk factors combined. †he idea that disadvantaged groups may display
heterogeneity of effect or suffer greater illness from particular types of exposure
has emerged in the literature on smoking (Birch et al, 2000). Heterogeneity in
ambient environmental exposure may also exist. In a recent study on air
pollution and mortality, krewski et al (2000) found significant effect modification
via an education variable. Persons with the lowest education experienced the
largest health effects from sulfate pollution.
†his could create a ‘triple jeopardy’. In other words, disadvantaged groups face:
first, increased risks from social and behavioural determinants of health; second,
higher risks from high ambient pollution exposure; and, third, an effect
modification that makes exposure to ambient pollutants exert disproportionately
large health effects on them compared with advantaged groups.
†he implications of this hypothesis for government policy are considerable. If
the ‘triple jeopardy’ hypothesis survives further empirical testing, it suggests that the
largest health benefits would accrue not from simply reducing ambient
exposures to air pollutants but also from reducing pollution in areas where it is
worst and where social deprivation is largest. Certainly, this suggests a challenging
and incomplete agenda, and future research should investigate the possibility of
‘triple jeopardy’ arising from environmental injustice.
Acknowledgementx. We acknowledge funding from the †oxic Substances Research Initiative,
a granting program administered jointly by Health Canada and Environment Canada.
Referencex
Anderson W P, kanaroglou P S, Buliung R N, l996, ‘‘Simulating automobile emissions in an
integrated urban model’’ Tvanrpovtation Rereavch Recovd 71 7l – 80
Anderton D L, Aderson A B, Oakes J M, Fraser M R, l994, ‘‘Environmental equity: the
demographics of dumping’’ Demogvaphy 31 229 – 248
Avery D, 2000, ‘‘Peopling Canada: immigration defines the nation’’ The Bearev 86 28 – 37
Bailey † C, Gatrell A, l995 Intevactire Spatial Data Analyrir (Longman Scientific and †echnical,
Harlow, Essex)
Baumol W J, Oates W E, l988 The Theovy of Knrivonmental Policy 2nd edition (Cambridge
University Press, Cambridge)
Been V, l994, ‘‘Unpopular neighbors: are dumps and landfills sited equitably?’’ Rerouvcer 115
l6 – l9
Birch S, Jerrett M, Eyles J, 2000, ‘‘Heterogeneity in the determinants of health and illness: the
example of socioeconomic status and smoking’’ Social Science and Medicine 51 307 – 3l7
Bowen W M, Salling M J, Haynes k E, Cyran E J, l995, ‘‘†oward environmental justice:
spatial equity in Ohio and Clevaldn’’ Annalr of the Arrociation of Amevican Geogvaphevr 85
64l – 663
Brook J R, Dann † F, Burnett R †, l997, ‘‘†he relationship among †SP, PMl0, PM2.5,
and inorganic constituents of atmospheric particulate matter at multiple Canadian
locations’’ Jouvnal of the Aiv and Warte Management Arrociation 47 2 – l9
Bullard R D, l990 Dumping in Dixie: Race, Clarr, and Knrivonmental Quality (Westview Press,
Boulder, CO)
Burrough P A, McDonnell R A, l998 Pvincipler of Geogvaphical Infovmation Syrtemr (Oxford
University Press, New York)
Capek S M, l993, ‘‘†he ‘environmental justice’ frame: a conceptual discussion and an
application’’
Social Pvoblemr 46 5– 24
Catalano R, Dooley D, Wilson G, Hough R, l993, ‘‘Job loss and alcohol abuse: a test using
data from the epidemiologic catchment area project’’ Jouvnal of Health and Social
Behariouv 34 2l5 – 225
Chakraborty J, Armstrong M P, l997, ‘‘Exploring the use of buffer analysis for the identification
of impacted areas in environmental equity assessment’’ Cavtogvaphy and Geogvaphic
Infovmation Syrtemr 24 l45 – l58
A GIS – environmental justice analysis of particulate air 97l
pollution

Colls J, l997 Aiv Pollution: An Intvoduction (E & FN Spon, London)


972 M Jerrett, R † Burnett, P kanaroglou, J Eyles, and
coworkers

Cook R D, Weisberg S, l999 Applied Regverrion Including Computing and Gvaphicr (John Wiley,
New York)
Cutter S L, l995, ‘‘Race, class and environmental justice’’ Pvogverr in Human Geogvaphy 19 lll –
l22 Cutter S L, Holm D, Clark L, l996, ‘‘†he role of geographic scale in monitoring
environmental
justice’’ Rirk Analyrir 16 5l7 – 526
Dobroff F, l988 l986 Hamilton Aiv Quality (Queen’s Printer for Ontario, †oronto)
Dobroff F, l992 l990 Aiv Quality Data Summavy, Regional Municipality of Hamilton –
Wentwovth (Queen’s Printer for Ontario, †oronto)
Dobroff F, l993 l99l – l992 Aiv Quality Data Summavy, Regional Municipality of Hamilton –
Wentwovth (Queen’s Printer for Ontario, †oronto)
Dobroff F, l994 l993 Aiv Quality Data Summavy, Regional Municipality of Hamilton –
Wentwovth (Queen’s Printer for Ontario, †oronto)
Dobroff F, l996 l994 Aiv Quality Data Summavy, Regional Municipality of Hamilton –
Wentwovth (Queen’s Printer for Ontario, †oronto)
Dockery D W, Pope C A, l997, ‘‘Outdoor air l: particulates’’, in Topicr in Knrivonmental
Kpidemiology Eds k Steeland, D A Savitz (Oxford University Press, New York) pp ll9 – l66
Earickson R J, Bullick H I, l988, ‘‘†he areal association of urban air pollutants and residential
characteristics: Louisville and Detroit’’ Applied Geogvaphy 8 5– 23
Farhang A C, l983, ‘‘†he effects of wind on air quality in Hamilton, Ontario’’, report AQRB-
83- 00l-M, Atmospheric Environment Service, Environment Canada, 4905 Dufferin Street,
Downsview, †oronto
Ferrie J E, Shipley M J, Marmot M G, Stansfeld S, Smith G D, l995, ‘‘Health effects of
anticipation of job change and non-employment: longitudinal data from the Whitehall
study’’ Bvitirh Medical Jouvnal 311 l264 – l269
Fong E, l994,‘‘Residential proximity among racial groups in the US and Canadian
neighborhoods’’
Uvban Affaivr Quavtevly 36 285 – 297
Glickman † S, l994, ‘‘Measuring environmental equity with geographical information systems’’
Rerouvcer 116 2– 6
Greenberg M, l993, ‘‘Proving environmental inequity in siting locally unwanted land uses’’
Rirk: Irruer in Health and Safety 4 235 – 252
Hamilton J †, l993, ‘‘Politics and social costs: estimating the impact of collective action
on hazardous waste facilities’’ Rand Jouvnal of Kconomicr 24 l0l – l25
Handy F, l977, ‘‘Income and air pollution in Hamilton Ontario’’ Altevnatirer 6 l8 – 24
HAQI, l997, ‘‘Ambient air quality and effects on the environment in Hamilton –
Wentworth’’,
final report of the Environment Work Group, Hamilton – Wentworth Air Quality
Initiative, City Hall, 7l Main Street West, 5th Floor, Hamilton, Ontario L89 4YF
Hill A B, l965, ‘‘†he environment and disease: association or causation’’ Pvoceedingr of the
Royal Arrociation of Medicine 58 295 – 300
Institute of Medicine, l999 Towavd Knrivonmental Jurtice: Rereavch Kducation and Health Policy
Needr (National Academy Press, Washington, DC
Jerrett M, l999, ‘‘Air pollution, environmental equity and health: †SRI research program
(†oxic Substances Research Initiative l999 – 2002)’’ The Arrociation of Amevican
Geogvaphevr Newrlettev 34(9) 28
Jerrett M, Eyles J, Cole D, Reader S, l997, ‘‘Environmental equity in Canada: an empirical
investigation into the income distribution of pollution in Ontario’’ Knrivonment and Planning A
29 l777 – l800
kazemipur A, Halli S S, 2000, ‘‘†he colour of poverty: a study of the poverty of
ethnic and immigrant groups in Canada’’ Intevnational Migvation 38 69 – 89
kim D, Jerrett M, 2000,‘‘Spatiotemporal and predictive modelling of ambient PMl0
concentrations from †SP monitoring data’’, paper presented at 2nd International
Health Geographics Conference, Chevy Chase, MD, hosted by the Johns Hopkins
University School of Public Health; available as diskette from Johns Hopkins
krewski D, Burnett R, Goldberg M, Hoover k, Siemiatycki J, Jerrett M, Abrahamowich M,
White W, 2000, ‘‘Reanalysis of the Harvard six-cities study and the American Cancer
Society study of air pollution and mortality, phase II: sensitivity analysis’’, Health Effects
Institute, Cambridge, MA
McMaster R, Leitner H, Sheppard E, l997, ‘‘GIS-based environmental equity and risk
assessment: methodological problems and prospects’’ Cavtogvaphy and Geogvaphic
972 M Jerrett, R † Burnett, P kanaroglou, J Eyles, and
coworkers

Infovmation Syrtemr 24 l72 – l89


A GIS – environmental justice analysis of particulate air 973
pollution

Mohai P, l995, ‘‘†he demographics of dumping revisited: examining the impact of alternate
methodologies in environmental justice research’’ Vivginia Knrivonmental Law Jouvnal
14 6l5 – 653
Novak J H, Dennis R L, Byun D W, Pleim J E, Galluppi k J, Coats C J, Chall S, Vouk M A,
l995, ‘‘EPA third generation air quality modeling system, model-3, volume l: concept’’,
EPA/600/R95/084, National Exposure Research Laboratory, Research †riangle Park, NC
Nyergis †, Robkin M, Moore † J, l997, ‘‘Geographic information sysems for risk
evaluation: perspectives on applications to environmental health’’ Cavtogvaphy
and Geogvaphic Infovmation Syrtemr 24 l23 – l44
Odland J, l988 Spatial Autocovvelation (Sage, Beverly Hills, CA)
Ott W R, Roberts J W, l998, ‘‘Everyday exposure to toxic pollutants’’ Scientific Amevican
(February) 72 – 9l
Parkhurst D F, l998, ‘‘Arithmetic versus geometric means for environmental concentration data’’
Knrivonmental Science and Technology l February, pp 92A – 98A
Paustenbach D J, 2000, ‘‘†he practice of exposure assessment: a state-of-the-art review’’ Jouvnal
of Toxicology and Knrivonmental Health Pavt B 3 l79– 29l
Pengelly L D, kerigan A †, Goldsmith C H, Inman E M, l984, ‘‘†he Hamilton study:
distribution of factors confounding the relationships between air quality and
respiratory health’’ Jouvnal of the Aiv Pollution Contvol Arrociation 34 l039 – l043
Perrolle J A, l993, ‘‘Comments from the special issue editor: the emerging dialogue
on environmental justice’’ Social Pvoblemr 46 l– 4
Pijawka D J, Blair J, Guhathakurta S, Lebiednik S, Ashur S, l998, ‘‘Environmental equity
in central cities: socioeconomic dimensions and planning strategies’’ Jouvnal of
Planning Kducation and Rereavch 18 ll3 – l23
Pulido L, 2000, ‘‘Rethinking environmental racism: white privilege and urban development in
Southern California’’ Annalr of the Arrociation of Amevican Geogvaphevr 96 l2– 40
Rouse W R, Burghardt A F, l987, ‘‘†he natural environment: climate, weather, and
society’’, in Steel City: Hamilton and Region Eds M J Dear, J J Drake, L G Reeds
(University of
†oronto Press, †oronto) page 308
Sexton k, Adgate J L, l999, ‘‘Looking at environmental justice from an environmental
health perspective’’ Jouvnal of Kxporuve Analyrir and Knrivonmental Kpidemiology 9 3 – 8
Sheppard E, Leitner H, McMaster R B, †ian H, l999, ‘‘GIS-based measures of environmental
equity: exploring their sensitivity and significance’’ Jouvnal of Kxporuve Analyrir and
Knrivonmental Kpidemiology 9 l8 – 28
Statistics Canada, l99l Data Documentation fov the l99l Cenrur Pvofile Sevier— Pavt A ov B
(Supply and Services Canada, Ottawa)
Wallace L, l996, ‘‘Indoor particles: a review’’ Jouvnal of the Aiv and Warte Management
Arrociation
46 98 – l26
© 200l a Pion publication printed in Great Britain

You might also like