Race,Ethnicity,and the Distribution of Environmental Justice in the United States

Abstract

With data from the U.S. General Social Survey linked with regional environmental quality data this article considers the individual demand for environmental quality as measured by stated preferences, relative to its supply, to determine if race/ethnicity matter for the demand and supply of environmental quality and environmental justice. Parameter estimates reveal that relative to Whites, the distribution of environmental quality is unjust for African Americans, Native Americans, Asian Indians, Southeast Asians and other race. This suggests that as race/ethnicity conditions the distribution of environmental justice, there remains an unmet demand for environmental quality in the United States among non-Whites, As such, environmental policy interventions that acknowledge the intersection of race/ethnicity, civil rights, and environmental—such as the Environmental Protection Agency equity action plan―are well-motivated.

Keywords

supply/demand environmental quality environmental justice racial/ethnic inequality

INTRODUCTION

At least since 1983, the pioneering scholarship of Robert Bullard has produced insight on how race adversely conditions environmental quality.^1,2,3,4 Across a wide variety of environmental quality measures, relative to Americans in general, the communities where Black and other non-White Americans reside are lower in environmental quality,^5,6,7,8,9 racial disparities in environmental quality raise questions of whether there is “environmental justice” that is independent of race/ethnicity.

If, as economic theory posits, individuals trade off private consumption for environmental quality, the distribution of environmental quality will reflect an equilibrium that is optimal, implying that it is welfare optimal.¹⁰ This follows from individuals being bound by a budget constraint to purchase essential consumption goods such as housing, requiring that they trade off environmental quality for desirable and essential consumption goods such as housing. In equilibrium, individuals with lower incomes will locate to communities with lower environmental quality, implying they have optimized their consumption and supply of environmental quality in their geography. Such an outcome can be deemed “Market Justice” with respect to the distribution of environmental quality, as it reflects the optimal tradeoffs available to individuals in the market.

While there are several definitions of what constitutes environmental justice, that environmental quality is conditioned on race/ethnicity suggests the same for the distribution of environmental justice.^11,12 To the extent that individuals sort into optimal consumption-environmental quality choices, the distribution of environmental quality can be viewed as market justice,” as such choices reflect the income constraints they face conditional upon their preferences. Given this optimizing choice, if individual demand for environmental quality equals the supply, the distribution of environmental quality can be viewed as “Environmental Justice” as individuals are satisfied with their housing consumption-environmental quality choice dictated by what was enabled by the market. With data from the U.S. General Social Survey (GSS) linked with regional environmental quality data, this article considers the individual demand for environmental quality, as measured by stated preferences, relative to its supply, to determine if after accounting for individual equilibrium market sorting into consumption-environmental quality choices, race/ethnicity matter for environmental justice.

The remainder of this article is organized as follows. The second section provides a theoretical framework that motivates our empirical strategy. A simple model is provided that enables a consideration of whether, conditional on individuals sorting into communities on the basis of housing consumption-environmental quality combinations, there is a distribution of environmental justice that is not conditioned on race/ethnicity. The data and methodology are discussed in the third section. In section 4, the parameter estimates are reported. The last section concludes.

A SIMPLE MODEL OF NEIGHBORHOOD CHOICE, ENVIRONMENTAL QUALITY AND MARKET/ENVIRONMENTAL JUSTICE

We appeal to an economic-theoretic supply-and-demand approach in our empirical methodology. In particular, we posit that there exists both a supply and demand of environmental quality that are measurable and estimable via parameterization―at least approximately. The central aim is to determine the extent to which the supply and demand of environmental quality are aligned for individuals and whether or not the relationship between demand and supply is conditioned on race/ethnicity, which can inform the extent to which the distribution of environmental quality satisfies a particular notion of environmental justice―theoretically developed and exposited upon below.

Our theoretical framework is an equilibrium excess demand extension of Banzhaf and Walsh,¹³ whereby the price of environmental quality is accounted for with a Lindahl public good price.¹⁴ While the model exposition is formal, at a fundamental level it is simply a demand and supply model that is the foundation of elementary principles of economics. Let household preferences be characterized along a continuum by the indirect utility function V(y, P, E, p), where $y$ is income, $P$ is the unit price of housing, $E$ is a stock/supply of non-rival environmental quality (e.g., clean air/water), and $p$ is the personalized Lindahl price of environmental quality faced by a household. The Lindahl price is a hypothetical price reflecting what an individual would be willing to pay for more of environmental quality to consume.¹⁵ Individuals choose to live in one of $J$ communities, and choose their optimal level of housing conditional on locational choice. $V$ ( $\cdot$ ) contains a numeraire with price normalized to unity, and assumed to be continuous with bounded derivatives that satisfy $V_{y} >$ 0, $V_{P} <$ 0, $V_{E} >$ 0, and $V_{p} <$ 0. For each community, the housing supply is assumed to be continuous, and the exogenous supply of environmental quality is $E_{j}$ . The housing demand function, which determines optimal housing choices is continuous and given by $D$ ( $P$ , $y$ ), with $D_{P} <$ 0, and $D_{y} >$ 0.

To enable a sorting of households from low-housing price/low environmental quality to high-housing price/high environmental quality communities, it is assumed that the slope of the indirect utility function is increasing in $y$ . This single-crossing property generates income stratification across communities with different housing price and environmental quality combinations. Given the equilibrium sorting of households across communities with housing-price/environmental quality combinations, the following can be established: Proposition.

The market justice enabled by sorting into housing-price/environmental quality combinations across communities is also environmental justice if the demand for environmental quality equals its supply.

A proof of this proposition follows from considering the equilibrium of a typical individual given by $V^{*}$ , and the individual’s demand for environmental quality conditional upon the equilibrium stock of $S^{*}$ available in the community sorted into. The equilibrium outcome $V^{*}$ is trivially what is enabled by the Tiebout-like sorting by individuals into the continuum of housing-price/environmental quality combinations. As such, the equilibrium constitutes market justice as it reflects what is optimal regarding the possible tradeoffs between housing and environmental quality. From Roy’s Lemma, the demand for environmental quality for an individual in community $j$ is e $_{i j}^{d}$ = −( $\partial$ V $^{*}$ / $\partial p_{i}$ )/( $\partial$ V $^{*}$ / $\partial y$ ), where $p_{i}$ is the unit price of housing faced by individual $i$ . If for all individuals, e $_{i j}^{d}$ - $S_{j}$ = $ϕ_{i j}$ = 0, the individual demand and supply of environmental quality are identical—an outcome consistent with environmental justice as individual preferences are compatible with the supply of environmental quality, and excess demand is neither positive or negative—it is zero.

For the continuous distribution of environmental justice across individuals defined by $f$ ( $ϕ_{i j}$ ), let $ϕ_{i j}^{N W}$ and $ϕ_{i j}^{W}$ be the equilibrium outcome for a non-White and White individual respectively.¹⁶ This establishes: Corollary.

The distribution of environmental justice is racially biased and unjust if relative to Whites there is a difference in the excess demand of environmental quality for non-Whites.

This notion of environmental justice is perhaps a new one, and may differ from standard notions of it in the extant literature. It does however, cohere with John Rawls’ first principle of justice:¹⁷

First: each person is to have an equal right to the most extensive scheme of basic liberties compatible with a similar scheme of liberties for others.

We attempt to empirically operationalize this principle by permitting the possession of an individual equal right to environmental quality to be equal to its demand for that right relative to its supply. In a Rawlsian sense, if in a population of individuals, the excess demand for environmental quality for non-Whites and Whites is no different, there is environmental justice. On the other hand, if relative to Whites, the excess demand for environmental quality is higher, there is environmental injustice—conditioned on race/ethnicity.

Below, we exploit the above proposition and corollary to empirically determine if there is a racial bias in the distribution of environmental justice. In particular we attempt to estimate $E$ ( $ϕ_{i j}^{N W}$ − $ϕ_{i j}^{W}$ ), while controlling for market sorting into communities on the basis of voluntary and budget-constrained decisions about housing and amenity combinations.

There are, as identified by Laura Pulido, some possible shortcomings to our economic approach to environmental justice that may reduce its explanatory power.¹⁸ Pulido argues that most environmental racism research cast in the framework of economic theory typically relies on narrow, apolitical methods that reduce racism to individual discriminatory acts and polluting emitting facility siting decisions, which obscures its structural and ideological nature. Mainstream economics research methodologies view knowledge as what can be counted, replicated, and rendered independent of human interpretation. The economic approach typically tests whether polluting facilities/emissions are disproportionately located in communities of color, and focus on technical questions of statistical significance, and treat racism and polluting facilities/emissions as something measurable mainly through discrete, and locational measurements in data.

Pulido’s critique privileges research on environmental justice that integrates qualitative methods and historical analysis to trace how environmental racism emerges from long-term processes like segregation, deindustrialization, historical white privilege, and state regulation, rather than only from economic approaches that abstract away from such considerations. There is however some scientific virtue to our economic approach to environmental justice. Our econometric method estimates parameters are at least correlations between race and environmental outcomes. Banzhaf, Ma and Timmins note that the correlation between race and environmental outcomes is robust across research studies.¹⁹ To the extent that causality requires correlation, economic approaches to environmental racism that estimate parameters without accounting for more nuanced and unobserved ideological and structure factors can inform potential causal mechanisms. Indeed Banzhaf, Ma, and Timmins conclude (p. 190) that correlations estimated through parameterized economic models are useful “even if these correlations are the result of socioeconomic processes. Simply because the inequity is mediated through some mechanism does not mean it isn’t there.”²⁰ In this context, our economic approach to environmental racism is a sensible strategy to identify the extent to which race is a possible causal driver of environmental justice.

DATA AND METHODOLOGY

Our data are from two sources. We use the 2018 U.S. GSS for data on individual demand for environmental quality, and relevant sociodemographic characteristics. The GSS is a nationally representative repeated cross section sample of adults living in the United States.²¹ Conducted biennially, the GSS includes a standard core of demographic, behavioral, and attitudinal questions, with variation across survey years on topics of special interest.²²

Given the inquiry of this article, we utilize as dependent regressands, four variables identified in the 2018 GSS that are sensible approximations and/or proxies for an individual’s demand for environmental quality. These four questions are (1) In general, do you think that air pollution caused by industry is: [Ordered Categorical Response], (2) Are you very interested, moderately interested, or not at all interested in issues about environmental pollution?: [Ordered Categorical Response], (3) Are we spending too much, too little, or about the right amount on Improving and protecting the environment?: [Ordered Categorical Response], and (4) Do you think that pollution of America’s rivers, lakes and streams is: [Ordered Categorical Response].²³

The second source of our data is the U.S. News and World Report 2019 ranking of the U.S. states on environment and pollution—which aligns temporally approximate with the 2018 GSS.²⁴ These data rank each state according to (1) Industrial toxins based on 2019 Environmental Protection Agency (EPA) data on total chemical pollution per square mile of land area, and (2) Low Pollution Health Risk, a risk score based on 2019 EPA data capturing the risk pollution poses to health per capita due to the chronic health effects of chemical pollution from manufacturing, mining, electric power generation, and hazardous waste treatment.²⁵ We used these rankings as a measure of the aggregate supply of environmental quality with the property that the lower (higher) a state’s ranking the lower (higher) is the supply of environmental quality. Each of these supply measures captures environmental quality in a general sense as they capture multiple causes and consequences of environmental degradation. As the public version of the GSS only reports the region of respondents, we construct for each ranking measure the average ranking—rounded to the nearest integer—for the region in which the respondent lives as a measure of the supply of environmental quality available to the individual in the community/geography resided in.²⁶

Our measure of the supply of environmental quality has low geographical resolution, as it is constructed on the basis of the lowest level of geographical resolution possible in the publicly available GSS―9 Census divisions. This crude level of supply aggregation may indeed mask significant variation across the communities in which individuals live. As such, our parameter estimates may render an inference about the demand and supply of environmental quality that is an ecological fallacy, an attempt to infer individual behavior on the basis of highly aggregated measures. The ecological fallacy results from an aggregation bias due to the high level of aggregation of one outcome, and a specification bias due to a differential distribution of confounding variables engendered by the high level of aggregation²⁷. Given the practical limitations of disaggregating, our environmental supply variable in the publicly available GSS, our estimation framework attempts to mitigate and/or eliminate this ecological bias by including the same confounding covariates as controls for both the individual demand for environmental quality and the highly aggregated low geographic resolution supply of environmental quality.

Given the environmental quality demand and supply variables of interest are measured in ordered response categories, we utilize a Bivariate Ordered Probit framework to estimate the relevant parameters for the demand and supply of environmental quality:²⁸

Demand : D_{i 1}^{*} = X_{i 1} β_{i 1} + U_{i}

Supply : S_{i 2}^{*} = X_{i 2} β_{i 2} + V_{i}

where

D_{i 1}^{*}

and

S_{i 2}^{*}

are latent demand and supply for environmental quality respectively,

X_{i 1}

and

X_{i 2}

are row vectors of covariates with coefficient vector

β_{i 1}

and

β_{i 2}

respectively, and

U_{i}

and

V_{i}

are residual vectors that are possibly correlated. The likely correlation between

U_{i}

and

V_{i}

follows from standard supply and demand principles requiring that away from the price that equates supply and demand, quantity demanded and supplied will be inversely related. The correlation between

U_{i}

and

V_{i}

could also result from any correlations between the unobservable determinants of supply and demand.

Observable supply and demand are determined by the following threshold—or category—crossing conditions:

D_{i j}^{*}, S_{i j}^{*} = r iff Γ_{r j} \leq D_{i j}^{*}, S_{i j}^{*} < Γ_{r + 1 j} r = 1, \dots R_{j} and j = 1,2

where

R_{j}

is the number of categories of

D_{i j}^{*}

and

S_{i j}^{*}

, and the

Γ_{r j}

are threshold parameters with the property that

Γ_{1 j}

= −

\infty

and

Γ_{R_{j} j}

= +

\infty

We include in $X_{i 1}$ and $X_{i 2}$ identical covariates that account for individual equilibrium sorting into their communities—which can mitigate and/or eliminate the specification bias that results from, relative to individual demand, the high level of aggregation in our supply measure. The covariates of interest are binary variables for seven non-White race/ethnic groups, which measure the impact of being non-White on being in a particular ordered category of the ordered supply and demand regressands. The non-race/ethnicity regressors are not of interest per se, but are included as controls that explain the market determinants of individual’s sorting into their optimal housing consumption-environmental quality choices. As the excluded group is Whites, the relevant estimated $β_{i 1}$ and $β_{i 2}$ will measure the relative impact of being non-White on the demand and supply of environmental quality. In particular, the estimates will inform if for non-Whites, does environmental justice differ from Whites, as the relevant coefficients on the non-White binary variables inform how the demand and supply of environmental quality is distributed relative to Whites—an approximation of $E$ ( $ϕ_{i j}^{N W}$ − $ϕ_{i j}^{W}$ ).

The estimated coefficients on the non-White racial dummy variables enable a taxonomy for a characterization of relative environmental justice. Define the difference between the differences between the demand and supply of environmental quality between non-Whites and Whites excess demand for environmental as ( $D^{n w}$ − $D^{w}$ )−( $S^{n w}$ − $S^{w}$ ), where $D^{n w}$ is non-White individual demand for environmental quality, $D^{w}$ is White individual demand for environmental quality, $S^{n w}$ is the supply of environmental quality in a non-White individual’s community, and $S^{w}$ is the supply of environmental quality in a White individual’s community. The estimated coefficients on the non-White dummy variables in capture the effects, relative to Whites, of being a non-White on the demand and supply of environmental quality.

The estimated magnitudes of the relative effects enable a characterization and taxonomy of how environmental justice is conditioned on race/ethnicity is depicted in Table 1. As an illustrative interepretive example, suppose both ( $D^{n w}$ − $D^{w}$ ) and ( $S^{n w}$ − $S^{w}$ ) are positive. In this instance, the higher relative demand of non-Whites for environmental quality is associated with a higher relative supply of environmental quality. As such, this is compatible with our notion of environmental justice. If ( $D^{n w}$ − $D^{w}$ ) were positive and ( $S^{n w}$ − $S^{w}$ ) was zero or negative, this would be consistent with environmental injustice for non-Whites, as their relatively higher demand for environmental quality is not associated with a higher supply of environmental quality. Of the nine possible coefficient patterns, three are consistent with environmental justice not conditioned on race/ethnicity, three are consistent with environmental injustice with respect to non-Whites, and three are consistent with environmental justice with respect to Whites.

Table 1.

Taxonomy of Relative Environmental Justice and Race/Ethnicity From Estimated Ordinal Probit Parameter Estimates

Estimated magnitude of (D^nw−D^w)	Estimated magnitude of (s^nw – S^w)	Characterization of environmental justice
+	+	Environmentally Just
+	−	Environmentally unJust with respect to non-Whites
+	0	Environmentally unJust with respect to non-Whites
−	+	Environmentally unJust with respect to Whites
−	−	Environmentally Just
−	0	Environmentally unJust with respect to Whites
0	+	Environmentally unJust with respect to Whites
0	−	Environmentally unJust with respect to non-Whites
0	0	Environmentally Just

RESULTS

A statistical summary for the estimating sample, along with a definition of the regressands and covariates, is reported in Table 2. Our racial/ethnic measurement follows the detail that the GSS enables, which is not completely compatible with the U.S. Census Classifications. In particular we use the GSS variable RACECEN1, which allows individual respondents to report across 16 racial/ethnic classifications. This is in contrast to the 6 U.S. Census classifications. The RACECEN1 variable also allows categorization of Asians into those from the Indian subcontinent and Asians from Southeast Asia.

Table 2.

Covariate Summary

Covariate	Definition	Mean	Standard deviation	Number of observations
Demand for clean air:
In general, do you think that air pollution caused by industry is: (INDUSGEN)^a	1 = Extremely dangerous	1.54	1.27	453
	2 = Very dangerous
	3 = Somewhat dangerous
	4 = Not very dangerous
	5 = Not dangerous at all
Demand for clean water:
In general, do you think that pollution of America’s rivers, lakes, and streams is: (WATERGEN)^a	1 = Extremely dangerous	1.34	1.31	455
	2 = Very dangerous
	3 = Somewhat dangerous
	4 = Not very dangerous
	5 = Not dangerous at all
Demand for environmental protection:
Are we spending too much, too little, or about the right amount on improving and protecting the environment? (NATENVIR)^a	1 = Too little	1.35	.588	438
	2 = About right
	3 = Too much
Demand for pollution-free environment:
Are you very interested, moderately interested, or not at all interested in issues about environmental pollution? (INTENVIR)^a	1 = Very interested	1.65	.651	683
	2 = Moderately interested
	3 = Not at all interested
Supply of air/water pollution:
Regional Rank for industrial toxins	Ordinal numeric Rank^b	27.97	8.53	683
Supply of health-toxic air/water pollution:
Regional Rank for low-pollution health risk	Ordinal numeric Rank^b	28.42	8.66	683
Individual Covariates:
Age (AGE)^a	Numeric value of age in years	44.32	14.33	680
Real income (REALRINC)^a	Numeric value of real income in constant 1986 dollars	24336.6	28692.7	683
Education (EDUC)^a	Numeric value of number of years of schooling completed	14.01	2.89	683
Home owner (DWELOWN)^a	Binary variable equal to 1 if respondent owns home	0.574	0.495	683
Integrated neighborhood (RACLIV)^a	Binary variable equal to 1 if respondent lives in a neighborhood where there are individuals of a different race/ethnicity	0.574	0.495	683
African American (RACECEN1)^a	Binary variable equal to 1 if respondent is African American	0.165	0.372	683
Native American (RACECEN1)^a	Binary variable equal to 1 if respondent is American native or Alaskan native	0.019	0.137	683
Asian Indian (RACECEN1)^a	Binary variable equal to 1 if respondent is Asian Indian	0.006	0.076	683
Southeast Asian (RACECEN1)^a	Binary variable equal to 1 if respondent is Chinese, Filipino, Japanese, Korean, Vietnamese, or Other Asian	0.032	0.177	683
Pacific islander (RACECEN1)^a	Binary variable equal to 1 if respondent is Native Hawaiian, Guamanian or Chammoro, Samoan, or Other Pacific islander	0.003	0.054	683
Other race (RACECEN1)^a	Binary variable equal to 1 if respondent is Some other race	0.010	0.101	683
Hispanic American (RACECEN1)^a	Binary variable equal to 1 if respondent is Hispanic	0.064	0.246	683
Sample weight (WTSSALL)^a	GSS sampling weight for the number of nonrespondents, and adults in household	0.998	0.571	683

GSS variable name for the question relevant to the construction of the covariate. The definition is slightly edited for grammatical coherency.

U.S. News and World Report 2018 regional state average rounded to nearest integer.

Given that our GSS sample reflects complete case counts for the regressand and regressors in our estimating specifications, the shares of each racial/ethnic group in the sample are not necessarily representative of their populations shares in the U.S. This may introduce bias in our parameter estimates. To mitigate or eliminate this bias, we also report parameters estimates in which the specification is weighted by WTSSALL, which accounts for any differential probability of selection into the sample, subsampling of nonrespondents, and number of adults in the household of the respondent. The weighted parameter estimates, as they account for the possible non-representativeness of the sample relative to the population, will enable inference on the true population effect of race/ethnicity on the supply and demand of environmental quality.

Bivariate Ordinal Probit parameter estimates across two distinct supply and four distinct demand for environmental quality are reported in Tables 3–6.²⁹ In each instance, the joint distribution of the correlated residuals of the individual demand and aggregate supply specifications is assumed to be bivariate normal. To assess the adequacy of the regressors—as both the supply and demand specifications include the same regressors—a chi-square test for the joint zero equality of all the estimated parameters is reported. As a goodness-of-fit measure, we report the Pseudo-R $^{2}$ constructed from the relevant likelihoods.³⁰

Table 3.

Bivariate Ordinal Probit Parameter Estimates: The Relative Effect of Race/Ethnicity On Environmental Quality Demand and Supply (CLEAN AIR)

Specification	(Unweighted) supply = air/water pollution	(Weighted) supply = air/water pollution	(Unweighted) supply = health-toxic air/water pollution	(Weighted) supply = health-toxic air/water pollution
DEMAND: Clean Air
Age	−0.003 (0.494)	−0.002 (0.725)	−0.003 (0.493)	−0.002 (0.725)
Real income	4.14e-06 (0.031)^b	2.90e-06 (0.032)^b	3.21e-06 (0.031)^b	3.52e-06 (0.032)^b
Education	−0.022 (0.221)	−0.017 (0.402)	−0.021 (0.219)	−0.017 (0.399)
Home owner	0.054 (0.624)	−0.024 (0.856)	0.055 (0.622)	−0.024 (0.858)
Integrated	−0.072 (0.584)	−0.024 (0.856)	−0.072 (0.584)	−0.024 (0.876)
African American	−0.069 (0.645)	−0.006 (0.975)	−0.071 (0.642)	−0.007 (0.972)
Native American	0.146 (0.687)	0.234 (0.512)	0.146 (0.686)	0.234 (0.511)
Asian Indian	−0.291 (0.510)	−0.189 (0.636)	−0.291 (0.510)	0.234 (0.511)
Southeast Asian	−0.642 (0.057)^c	−0.612 (0.060)^c	−0.639 (0.058)^c	−0.609 (0.060)^c
Pacific Islander	−6.51 (0.000)^a	−6.25 (0.000)^a	−6.54 (0.000)^a	−6.22 (0.000)^a
Other race	−0.685 (0.204)	−0.130 (0.825)	−0.688 (0.203)	−0.132 (0.823)
Hispanic American	−0.004 (0.986)	0.113 (0.600)	−0.004 (0.984)	0.112 (0.602)
$H_{o}$ : $β_{i}$ = $β_{j}$ , $\forall$ $i$ $\neq$ $j$ ( $χ^{2}_{k - 1}$ )	1069.98^a	893.43^a	948.10^a	736.68^a
Pseudo-R²	0.014	0.016	0.015	0.017
Number of Observations	452	452	452	452
SUPPLY: Clean Air
Age	−0.003 (0.414)	−0.003 (0.515)	−0.003 (0.493)	−0.002 (0.575)
Real income	−4.46e-06 (0.033)^b	−4.81e-06 (0.024)^c	−5.51e-06 (0.021)^c	−4.61e-06 (0.012)^c
Education	−0.015 (0.437)	−0.026 (0.187)	−0.015 (0.447)	−0.024 (0.215)
Home owner	0.275 (0.009)^a	0.258 (0.026)^b	0.238 (0.024)^b	0.219 (0.057)^b
Integrated	0.069 (0.580)	0.134 (0.391)	0.100 (0.422)	0.159 (0.305)
African American	0.349 (0.003)^a	0.295 (0.019)^b	0.315 (0.009)^a	0.249 (0.047)^b
Native American	0.547 (0.192)	0.464 (0.267)	0.570 (0.122)	0.468 (0.188)
Asian Indian	0.094 (0.407)	0.180 (0.158)	0.139 (0.397)	0.242 (0.194)
Southeast Asian	0.144 (0.574)	0.204 (0.499)	0.011 (0.967)	−0.009 (0.976)
Pacific islander	−0.476 (0.000)^a	−0.469 (0.000)^a	−0.896 (0.000)^a	−0.985 (0.000)^a
Other race	0.038 (0.913)	−0.115 (0.804)	0.156 (0.656)	0.089 (0.821)
Hispanic American	0.061 (0.681)	0.061 (0.702)	−0.279 (0.058)^c	−0.329 (0.036)^b
$H_{o}$ : $β_{i}$ = $β_{j}$ , $\forall$ $i$ $\neq$ $j$ ( $χ^{2}_{k - 1}$ )	1069.98^a	893.43^a	948.10^a	736.68^a
Pseudo-R²	0.014	0.016	0.015	0.017
Number of Observations	452	452	452	452

Estimated standard errors are robust.

Approximate p-values are in parentheses.

Significant at the 0.01 level.

Significant at the 0.05 level.

Significant at the 0.10 level.

Table 4.

Bivariate Ordinal Probit Parameter Estimates: The Relative Effect of Race/Ethnicity On Environmental Quality Demand and Supply (CLEAN WATER)

Specification	(Unweighted) supply = air/water pollution	(Weighted) supply = air/water pollution	(Unweighted) supply = health-toxic air/water pollution	(Weighted) supply = health-toxic air/water pollution
DEMAND: Clean Water
Age	−0.001 (0.796)	−0.002 (0.735)	−0.001 (0.793)	−0.002 (0.732)
Real income	3.83e–06 (0.048)^c	3.21e–06 (0.118)	2.74e–06 (0.049)^b	3.24e–06 (0.122)
Education	−0.084 (0.000)^a	−0.079 (0.000)^a	−0.085 (0.000)^a	−0.079 (0.000)^a
Home owner	0.154 (0.184)	0.158 (0.236)	0.155 (0.181)	0.159 (0.233)
Integrated	−0.052 (0.712)	0.051 (0.758)	−0.052 (0.709)	0.051 (0.761)
African American	−0.089 (0.583)	−0.037 (0.825)	−0.090 (0.580)	−0.038 (0.821)
Native American	0.267 (0.557)	0.271 (0.529)	0.269 (0.553)	0.273 (0.524)
Asian Indian	1.63 (0.000)^a	1.63 (0.000)^a	1.63 (0.000)^a	1.63 (0.000)^a
Southeast Asian	−0.295 (0.337)	−0.374 (0.265)	−0.293 (0.340)	−0.374 (,266)
Pacific islander	−4.90 (0.000)^a	−4.81 (0.000)^a	−4.53 (0.000)^a	−4.44 (0.000)^a
Other race	−0.590 (0.229)	−0.309 (0.587)	−0.594 (0.227)	−0.312 (0.584)
Hispanic American	−0.038 (0.885)	0.059 (0.833)	−0.037 (0.887)	0.060 (0.832)
$H_{o}$ : $β_{i}$ = $β_{j}$ , $\forall$ $i$ $\neq$ $j$ ( $χ^{2}_{k - 1}$ )	755.72^a	650.08^a	435.23^a	325.39^a
Pseudo-R²	0.019	0.021	0.020	0.022
Number of Observations	454	454	454	454
SUPPLY: Clean Water
Age	−0.003 (0.398)	−0.003 (0.494)	−0.003 (0.411)	−0.022 (0.539)
Real income	−5.51e–06 (0.031)^b	−5.27e–06 (0.022)^b	−4.61e–06 (0.019)^b	−4.94e–06 (0.010)^a
Education	−0.012 (0.516)	−0.022 (0.247)	−0.010 (0.585)	−0.017 (0.365)
Home owner	0.286 (0.006)^a	0.269 (0.020)^b	0.252 (0.017)^b	0.235 (0.041)^b
Integrated	0.064 (0.612)	0.129 (0.406)	0.093 (0.457)	0.153 (0.323)
African American	0.362 (0.002)^a	0.306 (0.014)^a	0.329 (0.006)^a	0.265 (0.035)^b
Native American	0.562 (0.176)	0.483 (0.244)	0.591 (0.104)^c	0.497 (0.157)
Asian Indian	0.096 (0.379)	0.177 (0.149)	0.138 (0.379)	0.231 (0.188)
Southeast Asian	0.155 (0.544)	0.214 (0.480)	0.025 (0.925)	0.003 (0.993)
Pacific islander	−0.451 (0.000)^a	−0.445 (0.000)^a	−0.862 (0.000)^a	−0.947 (0.000)^a
Other race	0.049 (0.884)	−0.101 (0.826)	0.172 (0.619)	0.112 (0.770)
Hispanic American	0.081 (0.589)	0.083 (0.602)	−0.253 (0.084)^c	−0.297 (0.059)^c
$H_{o}$ : $β_{i}$ = $β_{j}$ , $\forall$ $i$ $\neq$ $j$ ( $χ^{2}_{k - 1}$ )	755.72^a	650.08^a	435.23^a	325.39^a
Pseudo-R²	0.019	0.021	0.020	0.022
Number of Observations	454	454	454	454

The reported coefficients on Real income are approximated for brevity and range in actual value from ± 2.57 $\times$ 10 $^{- 6}$ to ± 4.54 $\times$ 10 $^{- 6}$ .

Estimated standard errors are robust.

Approximate p-values are in parentheses.

Significant at the 0.01 level.

Significant at the 0.05 level.

Significant at the 0.10 level.

Table 5.

Bivariate Ordinal Probit Parameter Estimates: The Relative Effect of Race/Ethnicity On Environmental Quality Demand and Supply (ENVIRONMENTAL PROTECTION)

Specification	(Unweighted) supply = air/water pollution	(weighted) supply = air/water pollution	(Unweighted) supply = health-toxic air/water pollution	(Weighted) supply = health-toxic air/water pollution
DEMAND: Environmental Protection
Age	0.014 (0.001)^a	0.012 (0.010)^a	0.015 (0.001)^a	0.012 (0.011)^a
Real income	4.52e–06 (0.288)	4.16e–06 (0.097)^c	3.81e–06 (0.287)	3.28e–06 (0.096)^c
Education	−0.055 (0.023)^b	−0.050 (0.042)^b	−0.055 (0.023)^b	−0.050 (0.042)^b
Home owner	0.042 (0.750)	0.089 (0.545)	0.039 (0.761)	0.087 (0.550)
Integrated	−0.058 (0.669)	−0.127 (0.410)	−0.058 (0.669)	−0.128 (0.410)
African American	−0.098 (0.616)	−0.122 (0.567)	−0.101 (0.606)	−0.129 (0.548)
Native American	−0.651 (0.209)	−0.782 (0.141)	−0.657 (0.203)	−0.784 (0.137)
Asian Indian	0.590 (0.378)	0.377 (0.590)	0.586 (0.384)	0.363 (0.607)
Southeast Asian	−0.166 (0.639)	−0.094 (0.806)	−0.167 (0.635)	−0.100 (0.794)
Pacific islander	−5.06 (0.000)^a	−4.88 (0.000)^a	−5.37 (0.000)^a	−5.19 (0.000)^a
Other race	0.349 (0.495)	0.841 (0.131)	0.349 (0.494)	0.838 (0.136)
Hispanic American	0.298 (0.173)	0.381 (0.108)	0.301 (0.169)	0.382 (0.106)
$H_{o}$ : $β_{i}$ = $β_{j}$ , $\forall$ $i$ $\neq$ $j$ ( $χ^{2}_{k - 1}$ )	638.99^a	569.03^a	785.13^a	695.44^a
Pseudo-R²	0.020	0.027	0.021	0.026
Number of Observations	436	436	436	436
SUPPLY: Environmental Protection
Age	−0.003 (0.498)	−0.004 (0.364)	−0.003 (0.506)	−0.003 (0.445)
Real income	−5.12e–06 (0.017)^b	−4.71e–06 (0.003)^a	−4.94e–06 (0.011)^a	−4.58e–06 (0.001)^a
Education	−0.009 (0.594)	−0.009 (0.621)	−0.012 (0.481)	−0.006 (0.738)
Home owner	0.170 (0.100)^c	0.247 (0.025)^b	0.147 (0.159)	0.238 (0.030)^b
Integrated	−0.020 (0.871)	−0.094 (0.485)	0.035 (0.780)	−0.039 (0.774)
African American	0.297 (0.017)^b	0.323 (0.012)^b	0.294 (0.019)^b	0.325 (0.013)^a
Native American	−0.644 (0.016)^b	−0.846 (0.003)^a	−0.618 (0.042)^b	−0.781 (0.010)^a
Asian Indian	0.247 (0.375)	0.299 (0.286)	−0.068 (0.525)	−0.133 (0.268)
Southeast Asian	−0.131 (0.558)	−0.058 (0.823)	−0.334 (0.101)^c	−0.302 (0.249)
Pacific islander	−0.217 (0.021)^b	−0.188 (0.091)^c	−0.279 (0.003)^a	−0.235 (0.026)^b
Other race	0.039 (0.911)	−0.134 (0.754)	0.184 (0.591)	0.109 (0.757)
Hispanic American	0.210 (0.243)	0.163 (0.372)	−0.027 (0.877)	−0.078 (0.688)
$H_{o}$ : $β_{i}$ = $β_{j}$ , $\forall$ $i$ $\neq$ $j$ ( $χ^{2}_{k - 1}$ )	638.99^a	569.03^a	785.13^a	695.44^a
Pseudo-R²	0.020	0.027	0.021	0.026
Number of Observations	436	436	436	436

Estimated standard errors are robust.

Approximate p-values are in parentheses.

Significant at the 0.01 level.

Significant at the 0.05 level.

Significant at the 0.10 level.

Table 6.

Bivariate Ordinal Probit Parameter Estimates: The Relative Effect of Race/Ethnicity On Environmental Quality Demand and Supply (POLLUTION-FREE ENVIRONMENT)

Specification	(Unweighted) supply = air/water pollution	(Weighted) supply = air/water pollution	(Unweighted) supply = health-toxic air/water pollution	(Weighted) supply = health-toxic air/water pollution
DEMAND: Pollution-free Environment
Age	−0.005 (0.112)	−0.007 (0.064)^c	−.005 (0.112)	−.007 (0.064)^c
Real income	5.14e–06 (0.142)	5.37e–06 (0.026)^b	5.26e–06 (0.142)	5.62e–06 (0.026)^b
Education	−0.043 (0.013)^a	−0.036 (0.071)^c	−.043 (0.013)^a	−.036 (0.071)^c
Home owner	0.115 (0.229)	0.086 (0.434)	.115 (0.230)	.086 (0.433)
Integrated	−0.175 (0.109)	−0.156 (0.211)	−.175 (0.109)	−.157 (0.211)
African american	−0.054 (0.684)	−0.171 (0.218)	−.054 (0.684)	−.171 (0.218)
Native american	−0.621 (0.181)	−0.629 (0.217)	−.621 (0.181)	−.628 (0.217)
Asian indian	−0.778 (0.182)	−0.979 (0.091)^c	−.778 (0.182)	−.981 (0.091)^c
Southeast asian	−0.304 (0.164)	−0.223 (0.270)	−.304 (0.164)	−.223 (0.271)
Pacific islander	−5.54 (0.000)^a	−5.29 (0.000)^a	−4.77 (0.000)^a	−4.85 (0.000)^a
Other race	−0.230 (0.509)	−0.021 (0.942)	−.230 (0.509)	−.021 (0.941)
Hispanic american	−0.065 (0.768)	−0.095 (0.716)	−.065 (0.767)	−.095 (0.716)
$H_{o}$ : $β_{i}$ = $β_{j}$ , $\forall$ $i$ $\neq$ $j$ ( $χ^{2}_{k - 1}$ )	749.33^a	546.64^a	234.37^a	206.48^a
Pseudo-R²	0.013	0.015	0.014	0.016
Number of Observations	680	680	680	680
SUPPLY: Pollution-free Environment
Age	−0.002 (0.496)	−0.002 (0.507)	−0.003 (0.395)	−0.003 (0.454)
Real income	−4.74e–06 (0.017)^b	−4.81e–06 (0.012)^a	−4.63e–06 (0.010)^a	−5.18e–06 (0.006)^a
Education	−0.013 (0.382)	−0.022 (0.140)	−0.014 (0.340)	−0.020 (0.184)
Home owner	0.215 (0.012)^a	0.212 (0.022)^b	0.205 (0.018)^b	0.216 (0.021)^b
Integrated	0.024 (0.810)	0.007 (0.953)	0.065 (0.524)	0.037 (0.756)
African American	0.303 (0.002)^a	0.257 (0.010)^a	0.294 (0.002)^a	0.249 (0.014)^a
Native American	−0.096 (0.752)	−0.329 (0.299)	−0.074 (0.805)	−0.287 (0.344)
Asian Indian	0.189 (0.220)	0.269 (0.110)	0.079 (0.568)	0.121 (0.448)
Southeast Asian	−0.031 (0.866)	0.009 (0.962)	−0.211 (0.248)	−0.228 (0.300)
Pacific islander	−0.355 (0.009)^a	−0.319 (0.014)^a	−0.564 (0.017)^b	−0.487 (0.035)^b
Other race	0.070 (0.815)	−0.084 (0.835)	0.220 (0.465)	0.143 (0.672)
Hispanic American	0.112 (0.401)	0.079 (0.587)	−0.142 (0.289)	−0.198 (0.193)
$H_{o}$ : $β_{i}$ = $β_{j}$ , $\forall$ $i$ $\neq$ $j$ ( $χ^{2}_{k - 1}$ )	749.33^a	546.64^a	234.37^a	206.48^a
Pseudo-R²	0.013	0.015	0.014	0.016
Number of Observations	680	680	680	680

Estimated standard errors are robust.

Approximate p-values are in parentheses.

Significant at the 0.01 level.

Significant at the 0.05 level.

Significant at the 0.10 level.

As long as the either or both of the row vectors of covariates are not perfect indicators of the latent $D^{*}$ and $S^{*}$ , neither the supply and demand specifications can be expressed as indistinguishable combinations. In this context, Pudney and Shields (2000) conclude that no exclusion restriction is necessary, and the Bivariate Ordinal Probit estimator of $D^{*}$ and $S^{*}$ is identified through the non-linearity induced by $D^{*} \neq D$ and/or $S^{*} \neq S$ , where $D$ and $S$ are actual demand and supply, respectively. In particular, as long as the value of the latent demand/supply measures is not equal to the value of their actual measured values in the data, the parameters of our demand and supply function are identifiable. Given this, and as instruments to include as an exclusion restriction are in principle difficult to justify, and have untestable assumptions, our specification refrains from exclusion restriction, and identifies the supply and demand parameters on the basis on the nonlinearities that follow from $D^{*} \neq D$ and/or $S^{*} \neq S$ .³¹

The pseudo-R²’s for all the parameter estimates in Tables 3–6 are low, suggesting that our specifications are a poor fit relative to a Null model with no regressors. However, the pseudo-R² is based upon a pseudo-likelihood of the specifications, which only approximates the true likelihood function for the specifications.³² As pseudo-likelihoods estimates are biased relative to standard likelihood estimates the pseudo-R² also is also a biased measure of fit³³. In this context, the chi-square test for the joint zero equality of all the estimated parameters is likely a better measure of the adequacy of the specifications—and the null hypothesis that the parameters are jointly zero is always rejected in all the specifications.

As White Americans are the excluded racial group in our specifications, the included racial/ethnic classifications capture the demand and supply effects of non-Whites relative to Whites. Given that low values of each of our demand variables indicate stronger demand for the environmental quality under consideration―see Table 2―if a particular racial/ethnic dummy variable has a positive (negative) sign, it suggests that relative to Whites, non-Whites have lower (higher) demand. For the supply specification, higher ordinal values indicate lower supply, so a positive (negative) sign on a particular racial/ethnic dummy variable suggests that relative to Whites, non-Whites have lower (higher) supply. A statistically insignificant or zero value for a particular racial/ethnic dummy variable suggests no differences in demand or supply relative to Whites. The pattern of sign and significance informs conclusions about environmental justice based upon our theory-based taxonomy in Table 1.

Table 3 reports weighted and unweighted parameter estimates for the demand and supply of clean air across the two measures of supply. Following the taxonomy of environmental justice in Table 1, relative to Whites, the parameter estimates suggest that with respect to clean air the distribution of environmental justice is “just” for Native Americans, and Asian Indians. For African Americans, Southeast Asians, and Other Race, the parameter estimates suggest that the distribution of environmental justice is “unjust.” With respect to relative environmentally unjust outcomes for Hispanic Americans, the parameter estimate suggests that while Hispanic Americans and Whites have no difference in the demand for clean air, Hispanic Americans have―as indicated by the estimated supply function in two instances―a relatively higher supply of clean air. Thus, relative to Hispanic Americans, the distribution of clean air is “unjust” for White Americans.

Parameter estimates for the demand and supply of clean water across the two measures of supply are reported in Table 4. The parameter estimates suggest that with respect to clean water, it’s distribution is “just” for Southeast Asians, Pacific Islanders, Other Race, and in one instance for Native Americans across both measures of supply. For Asian Indians, Pacific Islanders and at least on average for Hispanic Americans, the parameter estimates suggest that the distribution of clean water is “unjust.” For African Americans, and in at least one instance for Native Americans and Hispanic Americans, the parameter estimates suggest that the distribution of clean water is “unjust.” With respect to Asian Indians, their relatively lower demand for clean water is not offset by a lower supply, suggesting that relative to Asian Indians, the distribution of clean water is “unjust” for White Americans. For Hispanic Americans, the signs on the parameter estimate suggests that while they have no difference in the demand for clean air, Hispanic Americans have―as indicated by the estimated supply function in two instances―a relatively higher supply of clean air. Thus, relative to Hispanic Americans, the distribution of clean air is “unjust” for White Americans.

Table 5 reports parameter estimates for the demand and supply of environmental protection across the two measures of supply. The parameter estimates suggest that with respect to environmental protection its distribution is “just” for Asian Indians, Pacific Islanders, Other Race, and Hispanic Americans across both measures of supply. For African Americans, the parameter estimates suggest that the distribution of environmental protection is “unjust.” For Native Americans, and in one instance for Southeast Asians, the signs on the parameter estimate suggests that while they have no difference in the demand for environmental protection relative to White Americans, they have a relatively higher supply of environmental protection. Thus, relative to Native Americans and Southeast Asians, the distribution of environmental protection is “unjust” for White Americans.

Parameter estimates for the demand and supply of a pollution-free environment across the two measures of supply are reported in Table 6. The parameter estimates suggest that with respect to a pollution-free environment, the distribution is “just” for Native Americans, Southeast Asians, Pacific Islanders, Other Race, and Hispanic Americans. For African American, and in two instances for Asian Indians, the parameter estimates suggest that the distribution of a pollution-free environment is “unjust.”

In general, the parameter estimates in Tables 3–6 suggest that relative to Whites, the distribution of environmental quality is unjust for African Americans, Native Americans, Asian Indians Southeast Asians, and Other Race. Given the taxonomy in Table 3, the parameter estimates do suggest that in some instances, in particular specifications, the distribution of environmental quality is unjust for Whites with respect to Native Americans, Asian Indians, Southeast Asians, and Hispanic Americans. The preponderance of the findings for African Americans suggest that among non-Whites, they are the racial/ethnic group subject to the most environmental injustice. This suggests that there may be racial/ethnic discrimination in the provisioning of environmental quality (Christensen, Sarmiento-Barbieri, and Timmins, 2002), particularly as it relates to African Americans.

CONCLUSION

With data from the 2018 U.S. GSS linked with regional environmental quality data, this article considered a new supply and demand theory of environmental justice. We test an economic supply and demand model to determine if after accounting for equilibrium sorting in the market whereby individuals make their optimal housing-environmental amenity choices, race/ethnicity matter for the demand and supply of environmental quality, and the distribution of environmental justice. Parameter estimates from Bivariate Ordinal Probit specifications of the demand and supply of environmental quality revealed that relative to White Americans, the distribution of environmental quality is unjust for African Americans, Native Americans, Asian Indians, Southeast Asians, and other races. For African Americans, the preponderance of environmentally unjust outcomes across our measures of environmental quality suggests that they are the racial/ethnic group subject to the most environmental injustice and subject to racial/ethnic discrimination in the provisioning of environmental quality. This suggests that while there may be “market justice” with respect to the allocation of environmental quality, it does not constitute an outcome of environmental justice, as race/ethnicity appears to matter for the supply and demand of environmental quality.

There are at least three noteworthy limitations of our analysis. First, the GSS environmental quality demand proxies we deploy are stated preferences, which may be biased measures of actual preferences, resulting in biased parameter estimates. However, as revealed preferences are positively correlated with stated preferences³⁴, our use of stated preferences as actual demand may constitute an approximation of actual demand with little or no bias.³⁵ Lastly, our measure of the supply of environmental quality has low geographical resolution, as it is constructed on the basis of the lowest level of geographical resolution possible―census divisions―in the publicly available GSS. This coarse level of supply aggregation may mask significant variation across the communities in which individuals live. In this context, our parameter estimates may render an inference about the demand and supply of environmental quality that constitutes an ecological fallacy―an erroneous inference of individual demand behavior on the basis of highly aggregated supply measures.³⁶ However, to the extent that the inclusion of individual covariate information for highly aggregated measures reduces or eliminates ecological bias―which we include as regressors in our low geographic resolution supply measures―our parameter estimates may not suffer from any significant bias.³⁷ A fruitful avenue for future research on environmental justice with GSS data would be to secure private information from GSS respondents on their specific place of residence, to link them to the supply of environmental quality that face at a more fine-grained community level—perhaps with U.S. EPA region/community specific data on measures such as particulate matter, air quality, etc.³⁸ Finally, race-specific environmental outcomes are not necessarily an individual discrete act of optimizing malicious intent. They may reflect a deeply embedded structural phenomenon operating perhaps through white privilege or racial capitalism.³⁹ As we do not account for the structural nature of racism, there could be some bias in our results. However, as our parameter estimates are correlations that do not account for structural socioeconomic processes, they at least inform the extent to which race is a possible causal determinant of environmental justice.⁴⁰

Our findings provide reasons as to why the changes implemented by the current U.S. Presidential administration to the U.S. Environmental Protection Agency equity action plan should raise concerns.⁴¹ The initial action plan articulates as EPA policy objectives, policy interventions to promote and embed environmental justice at the federal, tribal, state, and local levels. In addition, the action plan calls for the strengthening of civil rights enforcement in communities with environmental justice concerns. The proposed recent changes to the EPA’s policies deemphasize, and remove considerations of race-specific environmental justice initiatives in favor of “colorblind” and “merit-based” approaches. This includes, the termination of the EPA’s Diversity, Equity, and Inclusion programs and offices, and the national Office of Environmental Justice and External Civil Rights.⁴² Our parameter estimates suggest that as race/ethnicity continues to condition the distribution of environmental justice, there still remains an unmet demand for environmental quality in the U.S. among non-Whites. As such, the recent EPA policy reversals on deprioritizing environmental justice is not warranted, and environmental policy that acknowledges the intersection of race/ethnicity, civil rights, and environmental justice are still needed.

AUTHOR’S CONTRIBUTIONS

G.N.P.—the sole author—was solely responsible for review and editing, conceptualization, securing necessary data, formal econometric/statistical analysis, revisions, and methodology.

Race,Ethnicity,and the Distribution of Environmental Justice in the United States

Abstract

Keywords

INTRODUCTION

A SIMPLE MODEL OF NEIGHBORHOOD CHOICE, ENVIRONMENTAL QUALITY AND MARKET/ENVIRONMENTAL JUSTICE

DATA AND METHODOLOGY

RESULTS

CONCLUSION

AUTHOR’S CONTRIBUTIONS

Footnotes

AUTHOR DISCLOSURE STATEMENT

FUNDING INFORMATION