Revisiting environmental inequity in Southern California: Does environmental risk increase in ethnically homogeneous or mixed communities?

Abstract

This study revisits the concept of environmental inequity in Southern California using the California Environmental Protection Agency’s most recent data and spatial models. Empirical studies in the late 1990s documented the existence of environmental inequity among disadvantaged populations in the area, and we still found evidence of environmental inequity. However, our findings were more nuanced and subtler than previous results. The risk of being exposed to pollutants (e.g. ozone, PM2.5 and others) increases with a corresponding increase in Hispanic or Asian populations in a census tract. The risk of living near adverse environmental conditions (e.g. hazardous facilities, ground water threats and more) was less clear according to minority status. As the percentage of Hispanics in a census tract increases, the environmental risk increases only to a point, and then decreases. This finding suggests that, at present, some Hispanic communities enjoy better environmental conditions than do ethnically mixed communities, but the risk of being exposed to pollutants still increases with an increase in the percentage of Hispanics in a census tract. If policy needs to be developed and updated accordingly to reflect changing environments, this new evidence directs urban environmental inequity research to pay attention to ethnically mixed communities as well.

Keywords

CalEnviroScreen tool environmental inequity spatial regression

The US Environmental Protection Agency (EPA) states that environmental justice (EJ) for all communities and persons across the nation will be achieved when everyone enjoys ‘the same degree of protection from environmental and health hazards and equal access to the decision-making process to have a healthy environment in which to live, learn and work’ (EPA, 2018). Over the past three decades, EJ research has become an integral aspect of achieving this goal, because the findings of the research inform us of the status and progress of EJ in society. Currently, evidence shows the distinct presence of environmental inequity (EI) in the United States by race/ethnicity and economic class (Ringquist, 2005). However, it also is well known that the empirical results of EI studies are often sensitive to the spatial units of analysis, measurements and empirical model specifications (Chun et al., 2012; Mohai et al., 2009). In particular, analytical questions – such as those pertaining to measures of environmental risks and spatial dependence – remain in EJ analyses (Chakraborty et al., 2011; Ringquist, 2004). Acknowledging these limitations, we revisited EI in Southern California using data that the California EPA (2017) released recently. In this article, we use the term EJ to refer to a desired state of society or the field of research (regardless of analytical approaches), while we use EI to refer to the empirical reality that different social groups bear environmental risks differentially (focused more on distributional outcomes).

Southern California has been a fertile ground for scholarly work in the EJ literature because of the historical and ongoing struggle with pollution problems, as well as minority and low-income communities’ increased concern about pollution and health risks. The American Lung Association (2013) ranked the Los Angeles–Long Beach–Riverside area fourth among US cities with the most persistent air pollution in the country. The current estimate of Los Angeles County’s population is approximately 10 million, which makes it the most populous US county (US Census Bureau, 2010). The majority of the population is White (approximately 50%), while African Americans comprise approximately 8% and 14% are Asians. Hispanics and Latinos of any race constitute 48% of the population. Previous studies of Southern California EJ in the late 1990s and early 2000s presented consistent evidence that minority residents tend to disproportionately bear environmental risks (Boer et al., 1997; Boone and Modarres, 1999). This article examines whether EI still remains in Southern California by using the most recent data and an analytical technique that can address spatial dependence. Therefore, our research questions are as follows: 1) Does EI persist in the study area? 2) Are the empirical findings consistent when environmental risks are measured and tested with traditional environmental risk measures vs. composite scores? 3) Do the empirical findings remain when spatial dependence is considered?

We found that the risk of being exposed to pollutants increases with an increase in Hispanic or Asian population within a given census tract. This result was consistent and robust regardless of analytical decisions about measurement and model choice. However, a link between the risk of living in adverse environmental conditions and race/ethnicity was less clear. When examined using an ordinary least squares (OLS) model, the risk increases as the percentage of Hispanics increases in a census tract. However, no association was found in a spatial model based on an assumption of linearity. Instead, the spatial model supported a curvilinear association between risk and Hispanics. As the percentage of Hispanics in a census tract increases, the environmental risk increases to a point, and then decreases. Further, we found that Hispanics’ segregation was associated with this environmental risk. Our study suggested that, at present, some Hispanic communities may enjoy better environmental conditions – living near fewer hazardous facilities and having fewer ground water threats, among others – than do ethnically mixed communities, but the risk of exposure to such pollutants as ozone, PM2.5, toxic releases and others increases as the percentage of Hispanics increases in a census tract. Thus, the answer to the question of who bears the environmental risk in Southern California depends upon the types of environmental risk and communities under scrutiny.

Environmental justice analysis

Recent EJ studies have expanded their scope to include environmental goods, such as green spaces (e.g. Boone et al., 2009), as well as a broader range of issues like climate justice or the human relation with the non-human world (Schlosberg, 2013). This is a constructive development in the field. At the same time, there still are issues of inequity in the distribution of environmental risks that require scholarly attention to further advance EJ in society.

From the onset of the field, EJ analysis has focused on testing the relationship between the choice of sites for environmentally hazardous facilities and the disproportionate environmental burdens exerted upon disadvantaged communities. Current evidence of EI based on race in the United States is quite well-established (Ringquist, 2005). However, studies that explain why race-based EI exists are both more complicated and less conclusive. In theory, for example, the racial discrimination thesis suggests that communities with a high percentage of minorities are often disproportionately exposed to greater environmental risks because decision makers are assumed to be racially motivated in their decisions about the locations of environmental hazards (Bullard, 1996). In an empirical sense, this thesis predicts that the greater the proportion of racial or ethnic minorities in a spatial unit, the higher the environmental risk because of decision makers’ site choices for environmental disamenities (e.g. hazardous waste treatment, storage and disposal facilities (TSDs) or toxic release inventory facilities (TRIs)).

The lack of a resistance thesis assumes that decision makers consider their communities’ potential to act collectively in order to resist the presence of environmental hazards (Hamilton, 1995). Communities with highly disadvantaged populations are less likely to organise and address problems that require collective action, so they are more likely to bear an environmental burden. Conversely, communities populated by wealthy residents are less likely to be exposed to environmental hazards because they are more willing to pay for environmental quality or have the capital, resources or political power at their disposal in order to protect their communities. Furthermore, income and race are highly correlated, such that minority communities often are economically disadvantaged. Thus, the thesis hypothesises that the higher the average household income, the lower the risk of exposure to environmental harm. Morello-Frosch et al. (2001) examined a slightly modified hypothesis in the Los Angeles area, with income as a quadratic variable. Extremely poor areas may be exposed to fewer environmental risks because of low levels of economic activity, and areas with wealthier residents also tend to have fewer environmental risks because of their political ability to resist pollution-generating activities. The increase in median household income may increase environmental risks, but only to a certain point. Thereafter, the increase in median household income may reduce environmental risks. In addition to income, previous studies have tested other wealth and community stability variables that are shown to be related to the potential to resist environmental hazards, such as median home value and home ownership rate.

The segregation thesis assumes that ‘residential segregation helps produce environmental racial inequality’ (Downey, 2005: 978). In Downey’s (2005)simple segregation model, the resources available to different social groups limit residential choices so that minorities are more likely to live in environmentally hazardous neighbourhoods. In an industrial and Latino city in the Los Angeles area, the coincidence of Latino and toxic sites was the result of zoning decisions in the 1920s and 1930s (Boone and Modarres, 1999). Segregation can also be the outcome of residential similarity preferences or other institutional practices in the EJ context (Downey, 2005). Some EJ analyses have tested the association between segregation and EI using a Dissimilarity Index ([DI], e.g. Downey, 2007; Lopez, 2002; Morello-Frosch and Jesdale, 2006). Those studies found a tenuous positive and significant association between segregation and EI using data from the 1990s, primarily in US metropolitan areas (e.g. Lopez, 2002). While segregation in census tracts can also be operationalised using the DI (Wong, 1993), we are unaware of such research in EJ studies to date.

These theses attempt to tease out the various complexities of EI according to specific perspectives. In efforts to answer the ‘why’ question based on the theories, several theoretical and analytical issues have been identified and discussed, such as income or race/ethnicity as the predictors of EI (Downey, 1998), as well as a temporal order in the decisions about the location of environmental hazards and residents (Campbell et al., 2010). Because such theoretical issues have gained significant attention elsewhere (Downey and Hawkins, 2008; Mohai et al., 2009; Pastor et al., 2001), in this article we focus on discussing two analytical issues: measurement and spatial dependence.

Two analytical issues and opportunities in the California EPA data

In the early stages of the field, a substantial body of EJ analyses relied on spatial coincidence analyses in order to assess EI. This approach compares demographic characteristics in predefined geographical units that contain hazardous sites or facilities (e.g. TSDs or TRIs) with those geographical units without them. However, despite its popular use, this approach has been criticised because of its limitations, such as not making a distinction between one vs. many hazard sources in a unit, ignoring different risk levels associated with different types of hazards in a unit or not addressing the boundary effects of hazards (Chakraborty et al., 2011).

EJ research has moved slowly towards the examination of ‘the distribution of pollution emission, pollutant levels and sources of pollution other than hazardous-waste facilities’, which is superior to facility location as a proxy for risk (Ringquist, 2004: 263). Rather than relying on the presence of – or proximity to – hazardous facilities, other studies have used data on the quantities of toxic chemicals released from TRIs (total tons of TRI emissions, e.g. Morello-Frosch et al., 2001). For example, Downey and Hawkins (2008) used pollution data from the EPA’s Risk Screening Environmental Indicators (RSEI) project. RSEI models the ‘toxicity-weighted concentration of air pollutants released from every facility listed in the Toxics Release Inventory program’ (Downey and Hawkins, 2008: 764). Nonetheless, this modification falls short in terms of accurately determining the spatial extent of toxic exposure (Chakraborty et al., 2011). Emissions and pollution levels in an area are not direct indicators of actual exposure to individuals, but are perceived as more relevant with respect to assessing environmental risks.

TRIs and TSDs have been the foci of the measurement of environmental risks in many EJ studies for a considerable period of time. However, TRIs or TSDs are not the only sources of potentially hazardous chemicals. According to Lopez (2002), TRIs contribute a minuscule proportion of the total amount of toxins in the air. Today, cars, buses and small non-stationary sources of chemicals contribute to a much larger proportion of hazardous chemical releases. This is also the case in California, which suggests that focusing on stationary sources alone can distort any estimates of the distribution of environmental burdens (Pastor et al., 2005). A cumulative exposure measure would be even better, but actual exposure and pollution levels are already sufficiently difficult to measure and collect (Campbell et al., 2015). The California EPA (2017: 10) acknowledges that, currently, ‘no data are available statewide that provide direct information on exposures’. Instead, their tool and the new data selected combine multiple environmental risk indicators in order to assess pollution burdens in Californian communities. We took the opportunity that these data offer us in order to update current knowledge in the area.

Regardless of the way that environmental risks are constructed and measured, EJ analyses also involve certain issues that must be confronted when analysing spatial data, because previous studies have tested the predictions of these theories at the aggregate level. The ideal in testing the theses would have been to use data at the individual or organisational level, as well as their decision-making practices, but such data are not readily available. This data challenge has been dealt with by measuring environmental risks in some spatial units; however, this has also introduced different analytical challenges. For example, Ringquist (2004) pointed out that EJ research needs to test various spatial definitions of communities in order to address any aggregation bias. In this respect, the EJ results’ sensitivity to spatial scales (e.g. census tract, zip code or county) has been well-discussed in the literature (Baden et al., 2007), and evidence of EI at one level does not imply the same result at another (Lopez, 2002).

Recently, others have also referred to the issue of spatial dependence in EJ analyses (Chakraborty et al., 2011; Chun et al., 2012). This refers to the phenomenon that observations from nearby locations are more similar than one would expect if they were random. For example, Figure 1 shows that pollution impact scores (see data section for the scores) are not spatially independent. In general, high scores in the exposure score map (i.e. the risk of being exposed to pollutants) were concentrated primarily on the east side of Los Angeles, but areas with high environmental effect scores (i.e. the risk of living near adverse environmental conditions) were found primarily on the central and south-east sides. This visual observation and Moran’s I results¹ from the data require us to attend to spatial dependence with respect to examining EI in the research area. Spatial dependence is a problem in classical statistical analyses – such as regression – that assume independence of observations and can lead to a bias in coefficient and variance estimations that subsequently makes a statistical decision unreliable (Cliff and Ord, 1981). However, very few EJ analyses have paid attention to this issue (e.g. Chun et al., 2012; Pastor et al., 2005). In our examination of EI in Southern California, we seriously considered the implications of these two analytical decisions for our knowledge about the important matter of EJ.

Figure 1.

LA county maps of pollution impact scores.

Methods

This section discusses our analytical approach, data and variables. Specifically, it discusses a spatial regression model specification, and then describes the California EPA’s environment screening dataset. Environmental risk measures and potential explanatory variables follow. The unit of analysis is census tracts.

Spatial autocorrelation is frequently observed when analysing aggregate data – such as the CalEnviroScreen 3.0 – because observations among spatial neighbours are often more similar than expected randomly (Kissling and Carl, 2008). In this situation, the OLS approach produces biased and inconsistent estimates (Chakraborty et al., 2011). In SAR models, the neighbouring relation is captured in an n-by-n matrix of spatial weights (W), with elements ( $w_{ij})$ representing a measure of the connection between locations i and j. In the SAR model, the usual OLS model is complemented by a term $(ρ Wu)$ that represents the spatial structure ( $ρ W)$ in the spatially dependent error term ( $u$ ).

y = X β + u, where u = ρ Wu + ε

where $ρ$ is the spatial autoregressive coefficient, W is the spatial weight matrix, $β$ represents the slopes associated with correlates in the original predictor matrix X and $ε$ represents the spatially independent errors.

Data

As part of the implementation of the California EPA’s 2004 Environmental Justice Action Plan, the agency developed the California communities’ environmental health screening tool (CalEnviroScreen: California EPA, 2017). The tool was designed to help the agency carry out its EJ mission in a manner that ensures the fair treatment of all Californians, including those in minority and low-income populations (California EPA, 2017). The first version of the dataset was released in April 2013, the second updated version in October 2014 and the most recent one in January 2017 (CalEnviroScreen 3.0). Over the years, the datasets have been updated and improved in order to incorporate multiple public reviews and comments (California EPA, 2017). CalEnviroScreen 3.0 uses census tracts as the geographic scale, because these may allow pollution burdens and vulnerabilities in communities to be screened more accurately than do other scales, such as those based on zip code and county (California EPA, 2017).

Environmental risk measures

Our study tested four environmental risk measures, the first two of which have been examined in the previous EJ literature. Pollution data from TRIs has been used in EJ studies using RSEI data (e.g. Downey and Hawkins, 2008). The CalEnviroScreen 3.0 dataset includes an indicator that captures ‘toxicity-weighted concentrations of modeled chemical releases to air from facility emissions and off-site incineration, averaged over 2011 to 2013’ (California EPA, 2017: 53) – TOXIC in tables. Second, the location of hazardous waste sites in communities has long been an EJ concern in California. For example, Boer and his colleagues (1997: 793) found that in LA in the 1990s, ‘race and ethnicity were significantly associated with TSDF location, even when the percentages of African American and Latino are evaluated as separate groupings’. The data also include ‘permitted hazardous waste facilities and hazardous waste generators within each census tract’ (California EPA, 2017: 52) – HAZWAST in tables.

The next set of environmental risk measures includes composite scores that show the relative pollution impact in communities. The exposures score (EXPOSURE) is calculated based on the seven exposures indicators that ‘people may be exposed to a pollutant if they come [in]to direct contact with it, by breathing contaminated air’ (California EPA, 2017: 10). Some pollution sources produce emissions and discharges that are accompanied by environmental concentrations to which individuals and populations are exposed. Such sources are: 1) ozone concentrations in the air; 2) PM 2.5 concentrations in air; 3) diesel particulate matter emissions; 4) drinking water contaminants; 5) pesticide use; 6) toxic releases from facilities; and 7) traffic density in census tracts. Thus, it includes both stationary and non-stationary pollution sources.

The environmental effects score (ENVEFF) is calculated in order to measure adverse environmental conditions using the five indicators that include ‘environmental degradation, ecological effects and threats to the environment and communities’ (California EPA, 2017: 10). The following five indicators were selected by the California EPA in order to assess adverse environmental conditions that are caused by pollutants in census tracts: 1) toxic cleanup sites; 2) ground water threats from leaking underground storage sites and cleanups; 3) hazardous waste facilities and generators; 4) impaired water bodies; and 5) solid waste sites and facilities. The effects of these pollution sources may be either immediate or delayed; however, living in an environmentally degraded community can adversely affect the health of individuals and populations (California EPA, 2017).

To create each composite score (consistent with the CalEnviroScreen 3.0), a percentile of each indicator for each census tract was assigned first, based on the rank order of the value obtained from original data sources. Then, the average percentile of the indicators selected for each composite score was scaled, ranging from 0 (lowest) to 10 (highest). The higher the score, the higher the relative effect of pollution in a community. Readers must refer to the report about the CalEnviroScreeen 3.0 data, measures and scoring methods (California EPA, 2017).

Table 1 shows that the California EPA (2017) collected each indicator value at different times and derived them from various data sources. Table 2 presents the descriptive statistics of the environmental risk variables mentioned above (i.e. single indicators: TOXIC and HAZWAST; composite scores: EXPOSURE and ENVEFF).

Table 1.

The twelve indicators in CalEnviroScreen 3.0, released in January 2017.

	Indicators	Description	Data sources	Median	Mean	Std. Dev.	Range
1	Ozone	Mean of summer months (May–October) of the daily maximum 8-hour ozone concentration (ppm), averaged over 2012 to 2014	California Air Resources Board (CARB)	0.05	0.05	0.01	0.04–0.07
2	PM2.5	Annual mean concentration of PM2.5 (average of quarterly means, μg/m³), 2012 to 2014	CARB	12.05	11.51	1.30	5.34–12.89
3	Diesel matter	Spatial distribution of gridded diesel PM emissions from on-road and non-road sources for a 2012 summer day in July (kg/day)	CARB; San Diego Association of Governments	22.77	24.47	17.76	0.15–208.40
4	Drinking water contaminants	Drinking water contaminant index for selected contaminants	Drinking water systems geographic reporting tool	618.0	552.7	197.3	0.0–1114.1
5	Pesticide use	Total pounds of selected active pesticide ingredients (filtered for hazard and volatility) used in production-agriculture per square mile, averaged over 2012 to 2014	California Department of Pesticide Regulation	0.0	6.37	101.35	0.0–3381.61
6	Toxic releases from facilities*	Toxicity-weighted concentrations of modelled chemical releases to air from facility emissions and off-site incineration, averaged over 2011 to 2013	Risk Screening Environmental Indicators EPA, Toxic Release Inventory programme, Mexico Registry of Emissions and Pollutant Transfer	4009	5064	7147	3–140,569
7	Traffic density	Sum of traffic volumes adjusted by road segment length (vehicle kilometres per hour) divided by total road length (kilometres) within 150 metres of the census tract boundary (2013)	California Environmental Health Tracking Program, California Department of Public Health	1017	1289	856	0–5,104
8	Clean–up sites	Sum of weighted sites within each census tract (data downloaded December 2016)	EnviroStor Cleanup Sites Database, Department of Toxic Substances Control (DTSC)	3.95	9.75	17.00	0.00–268.1
9	Groundwater threats	Sum of weighted scores for sites within each census tract (data downloaded December 2016)	GeoTracker Database, State Water Resources Control Board (SWRCB)	5.5	12.9	21.0	0.0–335.9
10	Hazardous waste generators and facilities*	Sum of weighted permitted hazardous waste facilities and hazardous waste generators within each census tract (permitted hazardous waste facilities was downloaded December 2016, hazardous waste data is from 2012–2014)	EnviroStor Hazardous Waste Facilities Database; Hazardous Waste Tracking System, DTSC	0.10	0.64	1.89	0.0–28.7
11	Impaired water bodies	Summed number of pollutants across all water bodies designated as impaired within the area (2012)	303(d) List of Impaired Water Bodies, SWRCB	0.0	2.98	4.41	0.0–32.0
12	Solid waste sites and facilities	Sum of weighted solid waste sites and facilities (as of December 2016)	Solid Waste Information System; Closed, Illegal, and Abandoned Disposal Sites Program, California Department of Resources Recycling and Recovery, CalRecycle; Hazardous Waste Tracking System, DTSC	0.50	2.33	4.89	0.0–97.8

Notes: The exposures score includes 1–7, and the environmental effects score includes 8–12. Indicators noted with * are traditional EJ measures tested in previous studies. See the original report by California EPA for details of each indicator (OEHHA, 2018).

Table 2.

Descriptive statistics.

	Mean	Std. Dev.	Min	Max	Description
Environmental risk
TOXIC	5064	7147	3	140,569	Toxic concentrations released from facilities
HAZWAST	0.64	1.89	0	28.7	Hazardous waste generators and facilities
EXPOSURE	6.01	1.52	0	10	Composite score of seven exposures indicators
ENVEFF	3.58	2.23	0	10	Composite score of five environmental effects indicators
Socio-economic status
HISP	0.47	0.29	0	0.99	% of Hispanic residents
BLAK	0.08	0.13	0	0.90	% of African American residents
ASIAN	0.13	0.15	0	0.87	% of Asian residents
HH_INC	59,599	30,292	0	242,935	Median household income (US$)
H_VAL	494,284	218,115	0	1,000,000	Median home value (US$)
H_OWN	0.47	0.26	0	1.00	% of residents who own their home
DI	0.03	0.02	0	0.11	Dissimilarity Index
DIST_HW	2.29	5.01	0	42.42	Average distance to the nearest highway from the centroid of a census tract (miles)
POPDEN	131	109	0	944	Population density (1000 persons/sq. mile)/100

Notes: N = 2341 census tracts in Los Angeles (LA) County; within the LA County boundary, some census tracts include mountains and areas without households, so that some variables have a minimum value of zero.

EJ correlates

The CalEnvScreen 3.0 dataset also includes 2010 census data on the racial and ethnic compositions of communities throughout the state at the census tract level. We included the percentages of Hispanics (HISP), African Americans (BLACK) and Asians (ASIAN) in the census tracts. Non-Hispanic whites were the reference group. For wealth and community stability variables, we included median household income (HH_INC), median home value (H_VAL) and home ownership rate (H_OWN). Data for wealth and community stability variables were collected separately from the 2010 census. We calculated the Dissimilarity Index (DI) of each census tract based on Hispanics vs. non-Hispanics in census block groups nested within each census tract (see also, Wong, 2003). The DI value increases when either of the ethnic groups in a census tract is greater than the other. DI values are higher in the areas with a high proportion of Hispanics, but they are also higher in areas with a high proportion of non-Hispanics. Hispanics were used in both the minority status and segregation variables, but DI values represent a different concept from the proportion of Hispanics. The correlation between the percentage of Hispanics and the DI was 0.30 (significant, but not extremely high). We also included distance to the nearest highways from a centroid of a census tract (DIST_HW). The further from the nearest highway, the lower the environmental risk. We included population density (POPDEN), considering the different population sizes and land areas of the census tracts. The higher the population density, the lower the environmental risk.

Results

This section presents the results first of OLS, and then of the SAR model. The OLS results showed consistent findings of EI in the area. When spatial dependence was considered, the OLS findings did not remain, except for one environmental risk measure.

OLS models showed EI by minority and economic status

Table 3 summarises the OLS results using the two traditional environmental risk measures (M1: TOXIC, and M2: HAZWAST), as well as the two pollution impact scores (M3: EXPOSURES, and M4: ENVEFF). The M1 results present the coefficient of each variable in predicting the pollution level of toxic concentrations that hazardous facilities release into the air, such as TRIs. As previous studies in the area have found (see also, Boer et al., 1997), the increase of racial and ethnic minorities in census tracts was significantly and positively associated with higher pollution levels. An increase in median household income reduced the risk. Community stability variables were not associated statistically significantly with increasing toxic concentrations. The results also showed that the higher the ethnic segregation in a census tract, the higher the environmental risk. An increase in distance to the nearest highway reduced the risk, as did an increase in population density. The M1 results suggest race- and ethnicity-based EI, rather than EI by economic status, in Southern California.

Table 3.

OLS results of environmental inequity.

	Traditional environmental risk measures						Relative pollution impact (composite scores)
	M1: TOXIC(yj)			M2: HAZWAST(yj)			M3: EXPOSURE			M4: ENVEFF
	Estimate	Std. Error	p-value	Estimate	Std. Error	p-value	Estimate	Std. Error	p-value	Estimate	Std. Error	p-value
HISP	21.382	1.281	***	0.036	0.014	**	0.863	0.127	***	1.044	0.230	***
BLACK	17.056	2.050	***	0.006	0.022		−0.083	0.203		0.191	0.368
ASIAN	32.624	1.784	***	0.005	0.019		1.621	0.177	***	0.248	0.320
HH_INC	−3.146	1.138	**	0.043	0.012	***	0.676	0.113	***	0.561	0.204	**
HH_INC^2	0.166	0.097		−0.004	0.001	***	−0.056	0.010	***	−0.050	0.017	**
H_VAL	−0.099	0.122		−0.004	0.001	**	−0.026	0.012	*	−0.069	0.022	**
H_OWN	2.337	1.583		−0.060	0.017	***	−0.775	0.157	***	−1.941	0.284	***
DI	65.404	11.484	***	0.327	0.121	**	−0.934	1.140		10.178	2.062	***
DIST_HW	−1.137	0.049	***	−0.004	0.001	***	−0.165	0.005	***	−0.119	0.009	***
POPDEN	−0.009	0.003	**	0.000	0.000	***	−0.001	0.000	**	−0.010	0.001	***
Intercept	35.635	2.128	***	0.247	0.022	***	5.896	0.211	***	5.817	0.382	***
Adj. R²	0.381			0.117			0.475			0.200

Notes: HH_INC and H_VAL are log transformed. (yj) refers to the Y-J procedure to transform the original value of the measures to improve normality (Yeo and Johnson, 2000). ***p < 0.001, **p < 0.01, *p < 0.05.

In M2, the percentage of Hispanics in a census tract was associated positively and significantly with the presence of hazardous waste generators and facilities. However, the percentages of African Americans and Asians were not associated significantly with the risk. As median household income increased, the risk increased, but the results did not show a clear inverted U distribution (cf. Morello-Frosch et al., 2001). Instead, the association became less steep as median household income increased, and the environmental risk decreased as home value and home ownership rates increased, while an increase in ethnic segregation increased the risk. Therefore, rather than race and ethnicity, economic status seems to be most strongly associated with the environmental risk of hazardous waste generators and facilities in M2.

The results in M3 and M4, based on composite scores, were somewhat consistent with the findings from the single indicator models (i.e. M1 and M2). The effect of pollution increased with an increase in the percentage of Hispanics, but the percentage of African Americans was not associated with the risk scores in M3 and M4. The risk of being exposed to an increasing effect of pollution increased significantly with the percentage of Asians in M3, but there was no association between the risk of living in adverse environmental conditions and the percentage of Asians in M4. In the examination of the economic status variable, the pollution effect increased as median household income increased in M3 and M4, but not exactly as an inverted U distribution. An increase in median home value was associated negatively with the risk of being exposed to increasing levels of pollution, or of living in increasingly adverse environmental conditions. An increase in home ownership rate reduced the environmental risk in M3 and M4, and an increase in ethnic segregation increased the risk of living in adverse environmental conditions in M4. Distance to the nearest highway and population density were associated negatively with environmental risk measures in both M3 and M4. In summary, the results from the composite scores models were quite consistent with those from the single indicator models: Hispanics were observed consistently to experience EI, regardless of the way that the OLS models measured environmental risks.

Spatial models showed less consistent results of EI by minority status

Table 4 summarises the estimated coefficients from the SAR models. All SAR models showed positive and highly significant spatial autoregressive coefficients ( $ρ)$ . This indicates that there is a strong spatial autocorrelation in the error terms in the OLS results, which is problematic.

Table 4.

SAR results of environmental inequity.

	Traditional environmental risk measures						Relative pollution impact (composite scores)
	M5: TOXIC(yj)			M6: HAZWAST(yj)			M7: EXPOSURE			M8: ENVEFF
	Estimate	Std. Error	p-value	Estimate	Std. Error	p-value	Estimate	Std. Error	p-value	Estimate	Std. Error	p-value
HISP	−1.163	0.655		0.007	0.018		0.833	0.134	***	−0.042	0.275
BLACK	−1.838	1.146		0.028	0.031		0.308	0.236		0.423	0.484
ASIAN	−0.905	0.927		0.003	0.026		0.760	0.193	***	−0.712	0.399
HH_INC	−0.036	0.298		0.025	0.009	**	0.133	0.064	*	0.381	0.135	**
HH_INC^2	−0.013	0.026		−0.002	0.001	**	−0.011	0.006		−0.028	0.012	*
H_VAL	−0.046	0.030		−0.001	0.001		−0.004	0.007		−0.019	0.014
H_OWN	0.796	0.454		−0.002	0.014		0.024	0.097		−0.513	0.204	*
DI	4.292	3.417		0.301	0.103	**	0.299	0.735		8.436	1.544	***
DIST_HW	−0.315	0.114	**	−0.004	0.002	**	−0.218	0.016	***	−0.107	0.028	***
POPDEN	−0.001	0.001		0.000	0.000	***	−0.001	0.000	***	−0.006	0.000	***
Intercept	38.013	1.992	***	0.127	0.019	***	5.797	0.170	***	3.659	0.318	***
$ρ$	0.971			0.813			0.894			0.859
AIC	11,950			−4820			4539			7946
Log likelihood	−5962			2423			−2251			−3957

Notes: HH_INC and H_VAL are also log transformed. (yj) refers to the Y-J procedure to transform the original value of the measures to improve normality (Yeo and Johnson, 2000). ***p < 0.001, **p < 0.01, *p < 0.05.

Most interestingly, minority status variables were not significantly associated with either of the single indicator models (M5 or M6). Also, socio-economic variables – such as minority status, economic status or ethnic segregation – were not associated with toxic concentrations in the air from hazardous facilities in the SAR model (M5). Similar to the OLS models, the risk of living in a community with an increasing number of hazardous waste generators and facilities was associated negatively with an increase in the median household income in a census tract (M6), as was an increase in the distance to the nearest highway. The ethnic segregation measure was positively associated with an increased number of hazardous waste generators and facilities.

In the analyses of composite scores, EI was found only in the exposures model (M7), in which the coefficient estimates for the percentages of Hispanics or Asians were positive and statistically significant. This implies that there was an increased risk of being exposed to increasing effects of pollution as the percentages of Hispanic or Asian populations increased. However, neither the percentage of African Americans nor the ethnic segregation measure was associated significantly with the risk measure. In M7, other EJ correlates were controlled and spatial dependence was addressed, but the analysis still demonstrated EI by minority status.

The environmental effects model (M8) showed no EI. The percentages of race and ethnicity variables were not significantly associated with an increase in the risk score. An increase in median household income in a census tract was significantly associated with the risk of living in severe adverse environmental conditions. An increase in home ownership rate was statistically significant and negatively associated with the risk. Ethnic segregation was significantly and positively associated with the risk score. The distance to the nearest highway and population density were also significantly and negatively associated with the risk measures in M7 and M8.

The finding that minority status variables were not associated with the environmental risk measure in the environmental effects SAR model (M8), but that segregation was, suggests that the association between environmental risk and minority status may not be linear, as both traditional EJ studies and our study assume. Because EJ research has rarely tested the possibility of a curvilinear relation between environmental risk and minority status, this implication is interesting, and Table 5 provides some insight into this finding. When the percentage of Hispanics was tested as a quadratic variable, we found that the assumption of a curvilinear relation may be correct. In Figure 2, we adjusted the coefficients of the EI correlates mentioned above in the intercept, and predicted the relationship between the environmental effects score and the percentage of Hispanics within the census tracts. The association was a widely inverted U-shaped distribution that showed a one-point difference in the score between the highest value (3.5 in a census tract with approximately 40%∼50% of Hispanics) and the lowest value (2.5 in a census tract with 100% of Hispanics). The slope became negatively steeper as the percentage of Hispanics increased. The percentage of Hispanics associated with the highest environmental risk score was between 40% and 50% of those in a census tract, which points to communities being significantly ethnically mixed.

Table 5.

SAR result of the environmental effects (ENVEFF) measure with the percentage of Hispanics as a quadratic variable.

	Estimate	Std. Error	p-value
HISP	3.004	0.754	***
HISP^2	−3.383	0.776	***
BLACK	−0.056	0.495
ASIAN	−1.025	0.404	*
HH_INC	0.311	0.135	*
HH_INC^2	−0.023	0.012
H_VAL	−0.021	0.014
H_OWN	−0.533	0.203	**
DI	12.452	1.789	***
DIST_HW	−0.106	0.029	***
POPDEN	−0.006	0.000	***
Intercept	3.423	0.324	***
$ρ$	0.864
AIC	7924
Log likelihood	−3948

Notes: HH_INC and H_VAL are log transformed. ***p < 0.001, **p < 0.01, *p < 0.05.

Figure 2.

SAR Relationship between the Environmental Effects Scores and the Percentage of Hispanics in a Census Tract.

Discussion

We discuss three key findings and their implications for the literature. First, previous studies in the late 1990s reported differential experience of environmental risks by minorities in Southern California (Boer et al., 1997; Boone and Modarres, 1999; Pastor et al., 2001; Morello-Frosch et al., 2001; Sadd et al., 1999). Our results showed that there remains evidence of EI in Los Angeles County even in the most recent data, but the finding was most apparent in the OLS models that do not account for spatial autocorrelation, and in which the measurement decision of environmental risks did not influence the findings of EI. However, when spatial dependence was considered, EI was found only in the model that measured the risk of being exposed to pollutants. This environmental risk increased as the percentages of Hispanics or Asians in a census tract increased. Therefore, it may be safe to say that EI still persists in Southern California for Hispanics, and sometimes for Asians. The racial/ethnic minority groups appeared to bear a disproportionate pollution burden, particularly with respect to exposure to pollutants, and this finding was quite robust in both the OLS and SAR models. At the same time, the results lead us to question previous findings in the area, particularly those based on a single indicator focused on TRIs or TSDs that did not account for spatial dependence.

Second, the most unsettling finding is that, with respect to the risk of living in adverse environmental conditions, EI was observed in the OLS model (Table 3: M4) but not in the SAR model (Table 4: M8). Both models are based on the linearity assumption between risk and minority status. However, Figure 2 suggests that there is an association between them, but that it is curvilinear. As the percentage of Hispanics in a census tract increases, the environmental risk increases to a point and then decreases. In our prediction based on Figure 2, ethnically mixed communities (with approximately 40%∼50% of Hispanics in the census tracts) experience a higher risk of living near adverse environmental conditions in Southern California than do predominantly Hispanic communities. The right tail of the prediction curve was steeper than was the left, as seen in Figure 2.

From the most recent data, can it be said that ethnically homogeneous or mixed communities experience increased environmental risks? The answer depends on the way that the research defines and measures environmental risks: exposure to pollutants or living near adverse environmental conditions. EI based on the former is quite robust in the most recent data and model choice in Southern California, but EI based on the latter is less clear. The answer to the latter measure depended on the analytical decisions in terms of measurement, spatial dependence and even the linearity assumption, to which EI studies have rarely paid attention. Given the results from Table 3: M4, Table 4: M8 and Table 5, we concluded that it is safe to assume that there is EI even in the risk measure of living near adverse environmental conditions. However, the association between the risk and the percentage of Hispanics in the community seemed to differ, in that it is the more ethnically mixed communities that experience adverse environmental conditions today.

The third key finding from our study is that the risk of living near adverse environmental conditions was associated with Hispanic segregation in Table 4: M8. The greater the ethnic segregation (Hispanic or Non-Hispanic), the greater the environmental risk of living near adverse environmental conditions. Downey (2005) found evidence of a relation between segregation and EI at the census tract level in the Detroit metropolitan area not by using a Dissimilarity Index (DI), as we did, but rather by measuring the changes in neighbourhoods and exposure. In Detroit, environmental racial inequality was primarily the product of residential segregation, which interestingly reduced African Americans’ proximity to manufacturing facility pollution (Downey, 2005). Several studies have also shown a tenuous association between segregation and EI at the metropolitan level using the DI (Lopez, 2002; Morello-Frosch and Jesdale, 2006; Downey, 2007). In our study, ethnic segregation in Los Angeles County was independently associated with the risk of living in adverse environmental conditions. However, this result neither speaks to a causal relation nor provides insight about the underlying mechanisms that lead to segregation in the county.

A DI for segregation has been used at the metropolitan level in urban studies. Its use at the census tract level may be unfamiliar, although the possibility has been previously proposed and discussed (Wong, 1993). This approach certainly differs from the measures employed in previous research at the census tract level (i.e. Downey, 2005). Other studies have used a DI to measure and test segregation at the metropolitan area level (Lopez, 2002; Morello-Frosch and Jesdale, 2006; Downey, 2007). We found no theoretical rationale to use the DI only in metropolitan-area-level analyses. Instead, geographic studies have developed a framework that can link segregation measures at multiple geographical scales, and our research was built on such work (Wong, 1993, 2003). Others can also point out that we could have constructed the measure based on different racial/ethnic groups. In our side analysis, we constructed and tested the DI measure based on Whites vs. Non-Whites. The results of EI by minority status remained the same as those in Table 4: M7 and M8. The racial segregation was not associated with the risk of being exposed to pollutants, whereas it was negatively associated with the risk of living near adverse environmental conditions. Here, we reported ethnic segregation based on previous studies that examined EI in a largely industrial and Latino city such as the City of Commerce (Boone and Modarres, 1999) and observed EI by Latinos in the Los Angeles area (e.g. Boer et al., 1997; Pastor et al., 2005).

This study focused on Los Angeles County because of its significance in the EJ literature, as well as the most recent data that provide opportunities to examine the implications of enduring analytical issues for our knowledge. However, the underlying data used in this study can be prone to error, given how environmental risk data have often been collected. The California EPA data may not be free from some known problems such as the self-reporting of TRI data (Marchi and Hamilton, 2006) and inequitable institutional processes in the identification and remediation of environmental hazards (Dull and Wernstedt, 2010). Additionally, a geographic scale potentially has an impact on EJ analysis results. Census tracts and census block groups have been widely utilised (e.g. Cutter et al., 1996), but a smaller spatial-unit-level analysis (e.g. neighbourhood) can be preferred in certain applications (e.g. Maantay, 2002). An empirical phenomenon is not likely to be homogeneous in each spatial unit, which becomes more obvious with large spatial units. Some studies present approaches to identify an appropriate geographic scale (e.g. Fisher et al., 2006), but there is no clear consensus on the appropriateness of geographic scale, especially with different research questions and aims. One related potential issue here is data availability. Often, socio-economic data are not available at a finer geographic scale than census block groups; our analysis is also restricted to the census tract level, because race/ethnic group data are not available at a finer geographic scale to allow for the calculation of the Dissimilarity Index values. Lastly, the trade-off of focusing on Los Angeles County is that we were unable to draw conclusions about the role of our findings in larger contexts. We also included and tested variables in this study because of theoretical rationales and previous empirical EJ studies. However, the results of association analyses do not imply causation, and modelling individual-level processes at the aggregate level can lead to the ecological fallacy.

The robust findings of EI in Hispanic or Asian populations – especially in the composite exposures score – are quite troubling from the point of view of environmental policy. If policy needs to be developed and updated accordingly in order to reflect changing environments, this new evidence directs urban EI research to pay close attention to ethnically mixed communities in order to alleviate their pollution burdens as well. On the other hand, it is surprising that it may actually be the ethnically homogeneous communities that experience decreasing adverse environmental conditions today. This finding provides a compelling reason to revisit EI with different environmental risk measures and analytical approaches. Future research is needed in order to know how ethnically homogeneous communities achieved the state of decreasing environmental risk (e.g. the role of social cohesion, technologies in organising for collective action). At the same time, it suggests that addressing EI may be like pushing a balloon. In at least one conceptualisation of environmental risk (i.e. the risk of living near adverse environmental conditions), equity in an ethnic community seemed to improve, but now is causing trouble in another community type (i.e. mixed communities). The issues of social justice and the distress that indications of environmental injustice bring to some groups are sufficiently significant that EJ researchers should consider the implications of environmental risk measurements and spatial dependence in the design of their empirical studies.

Footnotes

Acknowledgements

We appreciate comments on our article from Heather E. Campbell and Spiro Maroulis. We also benefited from our presentation at the 2017 Urban Affairs Association conference. We remain responsible for all errors.

Declaration of conflicting interests

The author(s) declared no potential conflicts of interest with respect to the research, authorship, and/or publication of this article.

Funding

The author(s) disclosed receipt of the following financial support for the research, authorship, and/or publication of this article: This work was supported by the Ministry of Education of the Republic of Korea and the National Research Foundation of Korea (NRF-2016S1A3A2924956).

Notes

References

American Lung Association (2013) State of the Air 2015. Available at: http://www.stateoftheair.org/2015/assets/ALA_State_of_the_Air_2015.pdf (accessed 18 September 2018).

Baden

Noonan

Turaga

RMR

(2007) Scales of justice: Is there a geographic bias in environmental equity analysis? Journal of Environmental Planning and Management 50(2): 163–185.

Boer

Pastor

Sadd

et al . (1997) Is there environmental racism? The demographics of hazardous waste in Los Angeles County. Social Science Quarterly 78(4): 793–810.

Boone

Modarres

(1999) Creating a toxic neighborhood in Los Angeles County: A historical examination of environmental inequity. Urban Affairs Review 35(2): 163–187.

Boone

Buckley

Grove

et al . (2009) Parks and people: An Environmental Justice inquiry in Baltimore, Maryland. Annals of the Association of American Geographers 99(4): 767–787.

Bullard

(1996) Environmental justice: It’s more than waste facility siting. Social Science Quarterly 77: 493–499.

California EPA (2017) California Communities Environmental Health Screening Tool. Version 3.0.

Campbell

Kim

Eckerd

(2015) Rethinking Environmental Justice in Sustainable Cities: Insights from Agent-Based Modeling. New York: Routledge.

Campbell

Peck

Tschudi

(2010) Justice for all? A cross-time analysis of toxics release inventory facility location. Review of Policy Research 27(1): 1–25.

10.

Chakraborty

Maantay

Brender

(2011) Disproportionate proximity to environmental health hazards: Methods, models, and measurement. American Journal of Public Health 101(S1): S27–S36.

11.

Chun

Kim

Campbell

(2012) Using Bayesian methods to control for spatial autocorrelation in environmental justice research: An illustration using toxics release inventory data for a Sunbelt county. Journal of Urban Affairs 34(4): 419–439.

12.

Cliff

Ord

(1981) Spatial Processes: Models & Applications. London: Pion.

13.

Cutter

Holm

Clark

(1996) The role of geographic scale in monitoring environmental justice. Risk Analysis 16(4): 517–526.

14.

Downey

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

15.

Downey

(2005) The unintended significance of race: Environmental racial inequality in Detroit. Social Forces 83(3): 971–1008.

16.

Downey

(2007) U.S. metropolitan-area variation in environmental inequality outcomes. Urban Studies 44(5–6): 953–977.

17.

Downey

Hawkins

(2008) Race, income, and environmental inequity in the United States. Sociological Perspectives 51(4): 759–781.

18.

Dull

Wernstedt

(2010) Land recycling, community revitalization, and distributive politics: An analysis of EPA brownfields program support. Policy Studies Journal 38(1): 119–141.

19.

Fisher

Kelly

Romm

(2006) Scales of environmental justice: Combining GIS and spatial analysis for air toxics in West Oakland, California. Health & Place 12(4): 701–714.

20.

Hamilton

(1995) Testing for environmental racism: Prejudice, profits, political power? Journal of Policy Analysis and Management 14(1): 107–132.

21.

Kissling

Carl

(2008) Spatial autocorrelation and the selection of simultaneous autoregressive models. Global Ecology and Biogeography 17: 59–71.

22.

Lopez

(2002) Segregation and black/white differences in exposure to air toxics in 1990. Environmental Health Perspectives 110: 289–295.

23.

Maantay

(2002) Mapping environmental injustices: Pitfalls and potential of geographic information systems in assessing environmental health and equity. Environmental Health Perspectives 110 (Suppl. 2): 161–171.

24.

Marchi

Hamilton

(2006) Assessing the accuracy of self-reported data: An evaluation of the Toxics Release Inventory. Journal of Risk and Uncertainty 32(1): 57–76.

25.

Mohai

Pellow

Roberts

(2009) Environmental justice. Annual Review of Environment and Resources 34: 405–430.

26.

Morello-Frosch

Jesdale

(2006) Separate and unequal: Residential segregation and estimated cancer risks associated with ambient air toxics in U.S. metropolitan areas. Environmental Health Perspectives 114(3): 386–393.

27.

Morello-Frosch

Pastor

Sadd

(2001) Environmental justice and Southern California’s ‘riskscape’: The distribution of air toxics exposures and health risks among diverse communities. Urban Affairs Review 36(4): 551–578.

28.

OEHHA (2018) CalEnviroScreen 3.0. Available at: https://oehha.ca.gov/calenviroscreen/report/calenviroscreen-30 (accessed 18 September 2018).

29.

Pastor

Morello-Frosch

Sadd

(2005) The air is always cleaner on the other side: Race, space, and ambient air toxics exposures in California. Journal of Urban Affairs 27(2): 127–148.

30.

Pastor

Sadd

Hipp

(2001) Which came first? Toxic facilities, minority move-in, and environmental justice. Journal of Urban Affairs 23(1): 1–21.

31.

Ringquist

(2004) Environmental justice. In: Durant

Fiorino

O’Leary

(eds) Environmental Governance Reconsidered: Challenges, Choices, and Opportunities. Cambridge, MA: MIT Press, pp. 255–287.

32.

Ringquist

(2005) Assessing evidence of environmental inequities: A meta-analysis. Journal of Policy Analysis and Management 24(2): 223–247.

33.

Sadd

Pastor

Boer

et al . (1999) ‘Every breath you take …’: The demographics of toxic air releases in Southern California. Economic Development Quarterly 13(2): 107–123.

34.

Schlosberg

(2013) Theorizing environmental justice: The expanding sphere of a discourse. Environmental Politics 22(1): 37–55.

35.

US Census Bureau (2010) American fact finder. Available at: http://factfinder.census.gov (accessed 18 September 2018).

36.

US Environmental Protection Agency (EPA) (2018) Environmental justice. Available at: https://www.epa.gov/environmentaljustice (accessed 18 September 2018).

37.

Wong

DWS

(1993) Spatial indices of segregation. Urban Studies 30(3): 559–572.

38.

Wong

DWS

(2003) Spatial decomposition of segregation indices: A framework toward measuring segregation at multiple levels. Geographical Analysis 35(3): 179–194.

39.

Yeo

Johnson

(2000) A new family of power transformations to improve normality or symmetry. Biometrika 87(4): 954–959.