The full text on this page is automatically extracted from the file linked above and may contain errors and inconsistencies.
Race and Environmental Worries
WP 21-15
Ranie Lin
Rice University
Lala Ma
University of Kentucky
Toan Phan
Federal Reserve Bank of Richmond
Race and Environmental Worries
Ranie Lin∗ , Lala Ma† & Toan Phan∗∗
September 8, 2021
Abstract
We use survey data to document a strong heterogeneity in stated degrees of
worry about environmental problems across racial groups. Minorities are significantly more worried about air and water pollution than their white counterparts, even after controlling for socioeconomic factors and pollution exposure.
Our finding implies that residential sorting based on heterogeneous financial
resources and heterogeneous levels of environmental concern is unlikely to be
the only driver of uneven exposure to pollution across racial groups.
Keywords: Residential sorting; pollution; race
∗
Student, Rice University, † Assistant Professor, University of Kentucky, ∗∗ Senior Economist,
The Federal Reserve Bank of Richmond. We thank Kartik Athreya, Rajashri Chakrabarti, and the
seminar participants at the Federal Reserve Bank of New York and the Federal Reserve Bank of
Richmond for helpful comments and suggestions. The views expressed here are those of the authors
and should not be interpreted as those of the Federal Reserve Bank of Richmond or the Federal
Reserve System. Contacts: rl65@rice.edu, lala.ma@uky.edu, and toanvphan@gmail.com.
1
1
Introduction
Decades of social sciences research has documented the disproportionate exposure of
pollution by socioeconomic status, a field that is broadly known as Environmental
Justice (EJ) (Banzhaf et al., 2019). A landmark study in 1987 found that race was
the most important factor for siting toxic wastes sites in the United States, where
the average percentage of people of color in a ZIP code with at least one commercial
hazardous waste facility was two times that of ZIP codes with none (Chavis and
Lee, 1987). These disparities remain persistent: an update of the study in 2007 finds
similar results (Bullard et al., 2007).
While the pattern of uneven exposure is clear, the empirical evidence on the
causal mechanisms that contribute to these patterns is limited. Work in economics
and related disciplines have posited that firms may site their polluting activities
near disadvantaged populations for reasons ranging from discriminatory preferences
(Becker, 1957) to lower labor/land costs and more lax zoning laws (Been, 1994;
Hamilton, 1995; Wolverton, 2002). On the other hand, disadvantaged populations
may move or “sort” toward polluted neighborhoods (Tiebout, 1956). A large hedonic
literature shows that housing prices are lower in areas with more pollution,1 lowering
the costs of living in these communities. The sorting explanation is thus often used to
attribute the drivers of uneven pollution burden to income inequality (Banzhaf and
Walsh, 2008; Banzhaf et al., 2019). Aside from income-based sorting, heterogeneous
tastes is seen as another driver of sorting, where certain groups have lower preferences
for environmental quality (Kuminoff et al., 2013; Bakkensen and Ma, 2020). However,
the literature has not empirically tested between these alternative drivers of sorting
in practice.
In this paper, we aim to evaluate the sorting explanation for uneven pollution
exposure by using a novel survey data on individuals’ concerns over the environment.
To our knowledge, this is the first paper to explicitly incorporate stated preferences
for the environment with actual pollution exposure to evaluate potential sorting
1
The hedonic literature is vast. Some recent examples include Davis (2011); Currie et al. (2015);
Haninger et al. (2017).
2
mechanisms behind environmental justice correlations. We map individuals in the
Gallup Poll Social Series (GPSS) survey from 2000 to 2020 to pollution data from
the Environmental Protection Agency’s (EPA) Risk-Screening Environmental Indicators (RSEI) dataset. We first test whether minorities have systematically lower
preferences for environmental quality, conditional on income. We then investigate
how the well-documented correlation between pollution and race is affected once we
condition on environmental concern. We frame our discussion using a stylized model
of residential sorting, simplified from Banzhaf and Walsh (2008). Our model explicitly allows for heterogeneous income and heterogeneous environmental preferences.
The sorting model predicts that if minorities are more exposed to pollution and all
else is equal, then uneven pollution burden must be because minorities have weaker
concern over pollution than non-minorities.
In contrast to the model’s prediction, our empirical analysis documents that
minorities are significantly more worried about air and water pollution than their
white counterparts. Interestingly, this is true even after we control for socioeconomic
factors and pollution exposure. Moreover, conditioning on environmental concern
actually does not significantly change the correlation between pollution exposure
and race. These findings suggest that sorting based on heterogeneous environmental
preferences is unlikely to be the only driver of uneven exposure to air and water
pollution. Furthermore, we also find that minorities are more worried about global
warming and their climate beliefs are more aligned with scientific consensus. This
suggests that environmental issues are generally a more salient problem for minorities.
Our findings have implications to a growing literature that aims to understand potential mechanisms that lead to uneven exposure to pollution. Individuals may sort
toward pollution based on various reasons, including heterogeneous income (Banzhaf
and Walsh, 2008), land-use restrictions (Colas et al., 2019), and beliefs (Bakkensen
and Barrage, 2017). Other recent papers have provided more nuanced sorting explanations, from constraints posed by housing discrimination (Christensen et al., 2020)
to different pollution information sets (Hausman and Stolper, 2020; Ma, 2019). By
combining administrative data on pollution exposure with surveyed attitudes on
pollution, our analysis allows us to directly control for preferences for environmental
3
quality in studying the correlation between race and pollution. In doing so, we can
assess whether uneven pollution exposure can be fully explained by heterogeneous
preferences. If this were the case, then maybe the uneven exposure to pollution
would be “efficient” (as in economies where a heterogeneous allocation of resources is
driven by trades between agents with heterogeneous preferences and where the first
welfare theorem holds). However, this hypothesis does not seem consistent with our
empirical findings. Instead, our paper lends support to the idea that other important
factors besides heterogeneous tastes (such as choice constraints or disproportionate
siting by polluting firms, as surveyed in Banzhaf et al. 2019) are likely to be at play
in contributing to uneven pollution exposure.
Our paper proceeds as follows. We construct a stylized sorting model in section
2. Section 3 presents our empirical analysis, including a description of our data,
empirical model, and results. Section 4 concludes.
2
Motivating model
We motivate our empirical analysis through the lens of a simple model. Our intentionally parsimonious model (based on Banzhaf and Walsh, 2008) aims to capture
the essence of residential sorting over environmental quality. An important ingredient is the heterogeneity in environmental preferences, which is key in mapping the
theory to the survey data.
Consider a set of individuals with heterogeneous endowment y and heterogeneous
preference parameter α in their utility u over environmental quality E and consumption c:
u(E, c) = α ln E + ln c.
All else equal, an individual with a higher α is more concerned about the environmental quality of his or her residential location. The heterogeneity in this environmental
concern parameter α could represent, for example, the differences in the degree of
knowledge, attention, or belief about the negative effects of pollution on health and
well-being. The heterogeneity in the endowment parameter y could represent, for
4
example, the differences in income, assets, or human capital.
Each individual chooses a location ` ∈ {0, 1} to reside, where for simplicity we
assume there are two types of locations with heterogeneous environmental quality E`
and cost of living P` , with
E1 > E0 ; P1 > P0 .
In other words, type 1 location has better environmental quality (e.g., farther away
from sources of pollution) and also a higher cost of living. This ranking reflects
the well-documented correlation between environmental amenities and cost of living
(e.g., Davis 2011; Currie et al. 2015). The utility from choosing location ` is then
given by u(E` , y − P` ).
Given endowment y and utility u, an individual chooses the more polluted location type (` = 0) if and only if
u(E0 , y − P0 ) > u(E1 , y − P1 ).
(1)
α
E1
0
>
.
With our parametrization of u, this inequality is equivalent to y−P
y−P1
E0
An immediate implication of (1) is that for a given degree of environmental concern, individuals with lower endowments are more likely to sort toward the polluted
location. Formally, for a fixed environmental concern parameter α, there exists an
endowment threshold ȳ, such that those choosing ` = 0 have y < ȳ. Figure 1a
illustrates this simple result. The red solid line represents the set of α and y of individuals who would be indifferent between the two location types. Points above this
line represent individuals who would strictly prefer the cleaner location type (` = 1),
and points below represent those who strictly prefer the more polluted location type
(` = 0). The dashed horizontal line represents the fact that for a fixed α, there exists an endowment cutoff ȳ, above which individuals strictly prefer ` = 1 and below
which individuals strictly prefer ` = 0. In words, all else equal, poorer people are
more likely to live in more polluted areas. This maps to the standard sorting by
income that has been well documented in the literature (e.g., Banzhaf and Walsh,
2008).
Another immediate implication of (1) is that for a given endowment, individuals
5
with less concern for environmental quality are more likely to sort toward the polluted
location type. Formally, for a fixed endowment y, there exists a threshold ᾱ for the
environmental concern parameter such that those choosing ` = 0 have α < ᾱ. Similar
to before, the dashed vertical line in Figure 1b illustrates this implication. In simple
words, our model implies that after controlling for observable socioeconomic proxies
for endowment (e.g., education, income, or wealth), individuals who live in more
polluted areas must have lower degrees of concern for environmental quality.
A corollary and most important implication of the sorting model is that if minorities are disproportionately exposed to pollution compared to whites, even after
controlling for various socioeconomic factors (as the aforementioned environmental
justice literature has documented and as is the case in our data), then minorities must
have lower degrees of concern for environmental quality. This is the key implication
that motivates our subsequent empirical analysis.
3
Empirical analysis
3.1
Data
Our primary source of data on environmental concern is the environmental edition
of the Gallup Poll Social Series, which is conducted in March of each year. Gallup
provides individual-level responses to a variety of environmental issues. The key
variables we use are the respondents’ stated degrees of worry or concern2 over air
pollution and concern over water pollution. The respondents’ levels of concern are
coded as a great deal (-1), a fair amount (-2), a little (-3), or not at all (-4).3
In additional analyses (to be described later), we also include a variable for the
respondents’ stated perception on the seriousness of global warming. Gallup also
provides detailed social, economic, and political attributes, including respondents’
locations (ZIP code and/or county), political leanings, education level, race, sex,
2
Throughout, we will use the terms worry and concern interchangeably.
We invert Gallup’s original values (e.g., we convert a value of 2 to a value of -2). This way, a
higher index intuitively represents a higher degree of concern.
3
6
age, and income level. In total, there are over 6,000 responses gathered from 2000 to
2020; ZIP code-level data is available beginning in 2008. Table A2 in the Appendix
provides more details of the Gallup data.
For exposure to environmental pollution, we use the Environmental Protection
Agency’s (EPA) Risk-Screening Environmental Indicators (RSEI) dataset.4 These
data combine information on the location and amount of chemicals emitted from
facilities that report to the Toxics Release Inventory (TRI) with chemical transport
models and toxicity data to better reflect human exposure and health damages from
TRI emissions. For a measure of pollution exposure for each respondent in our
Gallup data, we use the average RSEI index of pollution for the respondent’s county.
A higher index means more pollution exposure.
Lastly, for additional control variables, we obtain the following: county-level data
from 2010 to 2018 on total population sizes from the U.S. Census Bureau American
Community Survey (ACS) 1-year estimates, county-level data from 2000 to 2019
on wages and employment from the Bureau of Labor Statistics Quarterly Census
on Employment and Wages (QCEW), as well as the county-level median housing
values obtained from the Decennial Census. Table A1 in the Appendix provides the
summary statistics of our data.
For a preliminary look at the data, Figure 2 plots the time series of the average
degrees of worry over air pollution (left panel) and over water pollution (right panel)
by race. It reveals an interesting pattern: minority respondents, especially black and
Hispanic, are on average more concerned about environmental quality than white
respondents. Our subsequent analysis shows that this ranking remains even after we
control for various socioeconomic differences between racial and ethnic groups.
4
The data can be downloaded here: https://www.epa.gov/rsei.
7
3.2
Main result: heterogeneous environmental worries
We now formally investigate whether there are heterogeneous degrees of environmental concerns across race. Our main regression specification is as follows:
worry = α0 + βrace + γX + αyear + αstate + ,
(2)
where worry is the respondents’ stated degree of concern about either air or water
pollution, race is a dummy variable for whether the respondent is minority (we
will also break down the minority dummy into finer racial groups), X is a vector
of controls, αyear is the survey year fixed effect, and αstate is the fixed effect for
the state in which the respondent lives.5 Our controls include: stated employment
status, degree of worry about unemployment risk, income, age, political ideology,
education status, gender, marital status, and whether the respondent has children
under 18 years old.
The key coefficient of interest is β. If β is zero, then it means that, after controlling for observable differences in socioeconomic factors and the relevant differences
across states and time, there is no heterogeneity in the degree of environmental concern across race. If β < 0, then it means that minorities are less concerned about
environmental pollution (as the sorting model in Section 2 predicted). If β > 0, then
the reverse would be true: minorities are instead more concerned about pollution
(the opposite of the model’s prediction).
There is a potential omitted variable bias with regression (2), as it does not control
for pollution exposure. If minorities are more exposed to pollution (as documented in
the literature), then they would be naturally more worried about the quality of their
environment. To help address this issue, we include the RSEI index for pollution
exposure, measured at the county where each respondent lives. In addition, pollution
exposure could be endogenous: individuals less concerned about the environment
may choose to live in areas more exposed to pollution. To help address this issue,
we use the RSEI index of pollution in 2000, before our survey data was collected,
5
Our data does not have enough spatial variation across counties for us to include county fixed
effects.
8
rather than concurrent pollution levels. This leads us to our second specification:
worry = α0 + βrace + δ log(pollution2000 ) + γX + αyear + αstate + ,
(3)
where pollution2000 is the pre-sample county-level average RSEI pollution index.
Table 1 presents our main results. Column 1 reports the estimates for our main
specification (2) for worry about air pollution. Column 2 further replaces the race
dummy of whether the respondent is a minority with a race category for whether
the respondent is black, Hispanic, Asian, or other. Columns 3 and 4 are similar to
columns 1 and 2, except that they include the pollution control (specification (3)).
Columns 5 to 8 then repeat the exercises in columns 1 to 4 for worry about drinking
water pollution.
Throughout Table 1, we find a striking and robust pattern: all else equal, minorities are more concerned about pollution than whites. The estimated coefficients for
β in columns 1, 3, 5, and 7 are statistically significant at the 1 percent level. They are
also economically significant, with the magnitude of the estimated coefficients being
about a quarter to a third of the standard deviation of worry variables. That is, the
racial factor explains about a quarter to a third of the variation in environmental
worries in our data. Columns 2, 4, 6, and 8 paint a similar picture: all else equal,
different minority groups (especially black, Hispanic, and Asian) are more concerned
about air and water pollution.
Some additional results are noteworthy. Income is negatively correlated with
environmental worry. This could be partially because individuals with higher income
have more resources to protect themselves from the perverse effects of pollution and
are thus less worried. Furthermore, political ideology seems to matter: compared to
moderates, liberals are more concerned, and conservatives are less concerned about
environmental quality.
The apparent contradiction between our main empirical results and the main
implication of our sorting model is informative. Our evidence suggests that sorting, especially the kind of sorting based on heterogeneous environmental concern, is
unlikely a main driver of disproportionate exposure to air and water pollution. In
9
other words, lack of environmental concern cannot explain minority groups’ relatively
higher exposure to pollution.
3.3
Robustness checks
Our main result that minorities are more concerned about the environment than nonminorities is robust to a battery of checks, as reported in Appendix Table A3. We
consider four robustness exercises to verify heterogeneity in environmental worries.
First, we repeat the regression specification (3) controlling for Gallup-provided survey
respondent weights. Gallup assigns each respondent a weighting factor to correct
for non-response bias and unequal selection probability and makes its final survey
representative of the U.S. population. Demographic weighting factors are computed
according to the Census Bureau Current Population Survey (CPS). The first row of
Table A3 shows that the coefficients on the minority variable remain positive and
statistically significant after re-weighting each response.
Next, we consider replacing specification (3) with an ordered probit specification,
as the worryair and worrywater response variables are encoded as ordinal rather
than continuous quantities. In the ordered probit specification, the coefficients on
the minority variable remain positive and significant, as shown in the second row of
Table A3. We also attempt to control for individual-level savings and debt in specification (3) using two additional questions in the Gallup survey that ask respondents
about their current levels of savings and debt, respectively. Additional information
about respondents’ savings and debt levels may better reflect respondents’ wealth
levels than their income alone. Unfortunately, data for these questions is only available between 2003 to 2005, hence the sample size for this specification is drastically
reduced. These estimates, which lose statistical significance, are listed in row three
of Table A3.
Lastly, we repeat regression (3) with an additional control for local employment
growth, which may be correlated with both economic activity and pollution (for
example, through the opening of new manufacturing sites). We use data on employment from the BLS Quarterly Census on Employment and Wages and define local
10
employment growth as the February (the month before the Gallup survey) over-theyear percent change in employment in each respondent’s county. The fourth row
of Table A3 shows the estimated coefficients, which are positive and statistically
significant.
3.4
Heterogeneous exposure, revisited
Recall from the discussion of our model in Section 2, the motivation of our main
empirical analysis in Section 3.2 hinges on the stylized fact that minorities are more
exposed to pollution than their white counterparts. Even though the aforementioned
environmental justice literature has documented this fact, a concern is that this
heterogeneous pollution exposure may not be a feature of our data sample.
In this section, we confirm the fact that minorities are disproportionately exposed
to pollution within the context of our survey data. Specifically, we revisit the race
and exposure relationship, via the following specification:
log(pollution) = α0 + βrace + γX + αyear + αstate ,
(4)
where the left-side variable is the log of the average pollution index for the county
in which the survey respondent lives and for the year in which the respondent was
surveyed. As before, race is either a dummy for minority or a category variable for
various racial groups. X is the vector of controls, which include the control variables
as described in specifications (2) and (3). We further include the county-level housing
price index as a control variable. To help reduce endogeneity concerns, we include
the pre-sample housing price index from 2000.
Columns 1 and 2 of Table 2 report the estimates for equation (4). They show that,
in our sample, minority groups, in particular black respondents, are more exposed to
pollution relative to white respondents. This correlation resonates with the finding
in the environmental justice literature. Our results thus confirm that there is uneven
exposure to pollution in our data, as previously documented in the literature (e.g.,
Banzhaf et al. 2019).
There is a further implication of the sorting model. If those with a lower degree
11
of worry about environmental quality sort toward more polluted areas, as the theory
would predict, then controlling for environmental worry should reduce the correlation between race and pollution in equation (4). To see if this is the case, we include
the stated worry about air and water pollution as additional control variables on the
right hand side of (4). Columns 3 and 4 of Table 2 report the estimates for these
regressions. They show that the correlation between race and pollution exposure
remains robust after controlling for environmental worry. The estimates for the race
dummy, and especially the estimates for the black category, remain statistically significant. Their magnitudes are only slightly smaller than those in the corresponding
columns 1 and 3.
3.5
Seriousness of global warming
We have documented that minorities are more concerned about local air and water
pollution relative to white counterparts. But could it simply be because of a survey bias? For example, could it be because minorities are more likely to respond
pessimistically to surveys about their outlooks? In other words, is the difference in
stated degrees of worry largely “nominal?” This section will provide some additional
results: they are also more concerned about global warming, and interestingly, their
views about the cause and effects of global warming are also more aligned with the
scientific consensus. These findings suggest that environmental problems are generally a more salient issue for minorities, and thus it is unlikely that the difference in
degrees of worry is purely nominal.
We rerun our main specification (2) but replace worry over air or water with
respondents’ views about global warming. Gallup provides three useful variables:
view of seriousness of global warming, whether human activities is the main cause
of global warming, and whether global warming causes more intense hurricanes (see
Table A2 for details about these variables). We also include worry about air and
water as control variables.
Table 3 reports the results. Column 1 shows that minorities are significantly
more likely to worry about the seriousness of global warming (and there is correlation
12
between worry over local pollution and view over the seriousness of global warming).
Column 2 shows that minorities are also more likely to think that emissions from
human activities are the main cause of global warming. Similarly, column 3 shows
that minorities are more likely to think that global warming increases the intensity
of hurricanes. Together, columns 2 and 3 suggest that minorities’ views on the cause
and effects of global warming are more aligned with the scientific consensus.
4
Conclusion
We document a significant heterogeneity in stated concerns over air and water pollution across racial groups, even after controlling for differences in socioeconomic
variables and pollution exposure. All else equal, minority respondents are more worried about pollution than white respondents. Our analysis implies that, conditional
on socioeconomic factors such as income and education, uneven pollution exposure
across racial groups cannot be fully explained by residential sorting with heterogeneous preferences for environmental quality. To our knowledge, our paper is the first
to explicitly incorporate stated preferences for environmental quality to evaluate a
potential sorting explanation for the well documented correlation between race and
pollution exposure.
Our findings imply that other important factors besides heterogeneous preferences
and heterogeneous economic resources could be at play in explaining uneven pollution
exposure. The literature has suggested alternative theories relating to firm decisions,
bargaining between various stakeholders, and political economy (see the survey in
Banzhaf et al., 2019). For example, Christensen and Timmins (2018) argue that
discrimination constrains neighborhood choices of minorities: minorities are more
likely to be steered in their housing search towards less desirable neighborhoods,
including areas that are more polluted. Going forward, we believe that it is important
to empirically explore these alternative reasons behind uneven exposure.
13
α
α
y
α
y
y
(a) Sorting by endowment
(b) Sorting by environmental concern
Figure 1: Equilibrium sorting. In both figures, the solid red line represents the
set of endowment y and environmental preference parameter α of individuals who
are indifferent between the two location types. Individuals strictly prefer the more
polluted location (` = 0) if and only if the points associated with their y and α lie
below this indifference line. The horizontal dashed line in Panel (a) represents the
fact that for a fixed α, individuals choose ` = 0 if and only if y < ȳ. Similarly, the
vertical dashed line in Panel (b) represents the fact for a fixed y, individuals choose
` = 0 if and only if α < ᾱ.
Figure 2: Average stated worry over air pollution or drinking water pollution, separated by race.
14
(1)
minority
worry air
(2)
(3)
0.309∗∗∗
(0.0433)
(4)
0.274∗∗∗
(0.0368)
(5)
worry water
(6)
(7)
0.255∗∗∗
(0.0345)
(8)
0.244∗∗∗
(0.0275)
black
0.335∗∗∗
(0.0559)
0.283∗∗∗
(0.0498)
0.347∗∗∗
(0.0497)
0.306∗∗∗
(0.0401)
hispanic
0.319∗∗∗
(0.0605)
0.295∗∗∗
(0.0516)
0.192∗∗∗
(0.0470)
0.234∗∗∗
(0.0274)
asian
0.288∗∗∗
(0.0952)
0.295∗∗∗
(0.0771)
0.243∗∗∗
(0.0555)
0.210∗∗∗
(0.0618)
other
0.175∗∗
(0.0828)
0.126∗
(0.0735)
0.136
(0.0859)
0.0297
(0.0817)
log pollution
15
0.0119∗∗
(0.00521)
0.0115∗∗
(0.00526)
0.000493
(0.00528)
-0.000278
(0.00534)
income
-0.0122∗
(0.00610)
-0.0120∗
(0.00610)
-0.0138∗∗
(0.00536)
-0.0137∗∗
(0.00531)
-0.0167∗∗∗
(0.00611)
-0.0168∗∗∗
(0.00613)
-0.0183∗∗∗
(0.00481)
-0.0180∗∗∗
(0.00481)
liberal
0.199∗∗∗
(0.0285)
0.196∗∗∗
(0.0288)
0.198∗∗∗
(0.0265)
0.196∗∗∗
(0.0268)
0.110∗∗∗
(0.0335)
0.109∗∗∗
(0.0339)
0.0797∗∗
(0.0300)
0.0777∗∗
(0.0300)
conservative
-0.351∗∗∗
(0.0290)
-0.352∗∗∗
(0.0289)
-0.360∗∗∗
(0.0268)
-0.361∗∗∗
(0.0266)
-0.259∗∗∗
(0.0329)
-0.257∗∗∗
(0.0337)
-0.299∗∗∗
(0.0224)
-0.299∗∗∗
(0.0229)
Yes
Yes
Yes
0.201
8446
Yes
Yes
Yes
0.201
8446
Yes
Yes
Yes
0.190
6184
Yes
Yes
Yes
0.190
6184
Yes
Yes
Yes
0.169
8440
Yes
Yes
Yes
0.171
8440
Yes
Yes
Yes
0.163
6182
Yes
Yes
Yes
0.164
6182
Other controls
State FE
Year FE
Adjusted R-squared
N
Table 1: Main results: Disproportionate environmental worries. Other controls include stated employment
status, degree of worry about unemployment risk, age, education status, gender, marital status, and status
of having children under 18 or not. Standard errors in parentheses. ∗ p < 0.10, ∗∗ p < 0.05, ∗∗∗ p < 0.01
log pollution
(2)
(3)
(1)
minority
0.194∗∗
(0.0895)
(4)
0.177∗
(0.0918)
black
0.637∗∗∗
(0.130)
0.620∗∗∗
(0.131)
hispanic
-0.234
(0.164)
-0.250
(0.164)
asian
0.0321
(0.299)
0.0183
(0.300)
other
-0.130
(0.339)
-0.143
(0.337)
worry air
0.102
(0.0722)
0.102
(0.0725)
worry water
-0.0370
(0.0591)
-0.0406
(0.0598)
Yes
Yes
Yes
0.0248
3508
Yes
Yes
Yes
0.0299
3508
Other controls
State FE
Year FE
Adjusted R-squared
N
Yes
Yes
Yes
0.0238
3509
Yes
Yes
Yes
0.0290
3509
Table 2: Disproportionate pollution exposure, revisited. Other controls include
stated employment status, degree of worry about unemployment risk, income, age,
education status, gender, marital status, political ideology, status of having children
under 18 or not, and county-level housing price index for year 2000. Standard errors
in parentheses. ∗ p < 0.10, ∗∗ p < 0.05, ∗∗∗ p < 0.01.
16
(1)
gw serious
(2)
gw cause
(3)
gw hurricanes
minority
0.179∗∗∗
(0.0443)
0.0313∗
(0.0180)
0.280∗∗∗
(0.0850)
worry air
0.216∗∗∗
(0.0157)
0.128∗∗∗
(0.00884)
0.206∗∗∗
(0.0504)
worry water
0.0636∗∗∗
(0.0171)
0.0343∗∗∗
(0.0120)
0.0950∗∗
(0.0470)
income
-0.0129∗∗
(0.00608)
-0.00593∗
(0.00322)
-0.0262
(0.0163)
conservative
-0.375∗∗∗
(0.0381)
-0.255∗∗∗
(0.0183)
-0.319∗∗∗
(0.0788)
liberal
0.342∗∗∗
(0.0267)
0.114∗∗∗
(0.0136)
0.119
(0.0952)
Yes
Yes
Yes
0.317
8302
Yes
Yes
Yes
0.285
7591
Yes
Yes
Yes
0.278
535
Other controls
State FE
Year FE
Adjusted R-squared
N
Table 3: Views about global warming. First column: view of seriousness of global
warming. Second column: belief over whether pollution from human activities is the
main cause of global warming. Third column: belief over whether global warming
causes more intense hurricanes. Other controls include stated employment status,
degree of worry about unemployment risk, age, education status, gender, marital
status, and status of having children under 18 or not. Standard errors in parentheses.
∗
p < 0.10, ∗∗ p < 0.05, ∗∗∗ p < 0.01.
17
References
Bakkensen, L. A. and Barrage, L. (2017). Flood risk belief heterogeneity and coastal
home price dynamics: Going under water? Technical report, National Bureau of
Economic Research.
Bakkensen, L. A. and Ma, L. (2020). Sorting over flood risk and implications for
policy reform. Journal of Environmental Economics and Management, 104:102362.
Banzhaf, H. S. and Walsh, R. P. (2008). Do people vote with their feet? an empirical
test of tiebout’s mechanism. American Economic Review, 98(3):843–863.
Banzhaf, S., Ma, L., and Timmins, C. (2019). Environmental justice: The economics
of race, place, and pollution. Journal of Economic Perspectives, 33(1):185–208.
Becker, G. S. (1957). The economics of discrimination. University of Chicago press.
Been, V. (1994). Locally undesirable land uses in minority neighborhoods: Disproportionate siting or market dynamics? Yale Law Journal, 103:1383–2593.
Bullard, R. D., Mohai, P., Saha, R., and Wright, B. (2007). Toxic wastes and race
at twenty 1987–2007: Grassroots struggles to dismantle environmental racism in
the united states. Cleveland OH: United Church of Christ Justice and Witness
Ministry.
Chavis, B. F. and Lee, C. (1987). Toxic wastes and race in the united states. Commission for Racial Justice, United Church of Christ, New York.
Christensen, P., Sarmiento-Barbieri, I., and Timmins, C. (2020). Housing discrimination and the toxics exposure gap in the united states: Evidence from the rental
market. Review of Economics and Statistics, pages 1–37.
Christensen, P. and Timmins, C. (2018). Sorting or steering: Experimental evidence
on the economic effects of housing discrimination. Technical report, National
Bureau of Economic Research.
18
Colas, M., Morehouse, J. M., et al. (2019). The environmental cost of land use
restrictions. Working Paper.
Currie, J., Davis, L., Greenstone, M., and Walker, R. (2015). Environmental health
risks and housing values: Evidence from 1,600 toxic plant openings and closings.
American Economic Review, 105(2):678–709.
Davis, L. W. (2011). The effect of power plants on local housing values and rents.
Review of Economics and Statistics, 93(4):1391–1402.
Hamilton, J. T. (1995). Testing for environmental racism: Prejudice, profits, political
power? Journal of Policy Analysis and Management, 14(1):107–132.
Haninger, K., Ma, L., and Timmins, C. (2017). The Value of Brownfield Remediation.
Journal of the Association of Environmental and Resource Economists, 4(1):197–
241.
Hausman, C. and Stolper, S. (2020). Inequality, information failures, and air pollution. National Bureau of Economic Research No. w26682, pages 1–75.
Kuminoff, N. V., Smith, V. K., and Timmins, C. (2013). The new economics of equilibrium sorting and policy evaluation using housing markets. Journal of Economic
Literature, 51(4):1007–1062.
Ma, L. (2019). Learning in a hedonic framework: Valuing brownfield remediation.
International Economic Review, 60(3):1355–1387.
Tiebout, C. M. (1956). A pure theory of local expenditures. Journal of Political
Economy, pages 416–424.
Wolverton, A. (2002). Does race matter? an examination of a polluting plant’s
location decision. US Environmental Protection Agency, National Center for Environmental Economics, mimeograph.
19
Online Appendix
Table A1: Mean, conditional mean, and standard deviation of main variables
mean
worry air
-1.965
worry water
-1.765
global warming serious
0.655
worry unemployment
-2.224
employed
0.938
income indicator
6.437
age
44.670
conservative
0.392
liberal
0.230
some college
0.277
college graduate
0.246
postgraduate
0.240
female
0.424
has child
0.397
married
0.573
white
0.770
black
0.089
hispanic
0.093
asian
0.027
other race
0.021
savings index
-2.734
debt index
-3.467
pollution indicator (RSEI) 13935
std
mean, minority
mean, white
0.937
0.930
0.674
1.027
0.242
2.519
14.139
0.488
0.421
0.447
0.431
0.427
0.494
0.489
0.495
0.421
0.285
0.291
0.161
0.144
1.090
1.088
148364
-1.627
-1.458
0.817
-2.034
0.900
5.701
39.630
0.322
0.264
0.282
0.218
0.190
0.431
0.472
0.430
0.000
0.386
0.406
0.116
0.092
-2.771
-3.479
11138
-2.066
-1.856
0.607
-2.281
0.949
6.672
46.175
0.413
0.220
0.275
0.255
0.255
0.421
0.374
0.616
1.000
0.000
0.000
0.000
0.000
-2.726
-3.464
14661
20
Table A2: Gallup variable definitions
21
Variable Name
envworry air
envworry drnkwater
gw serious
gw cause
gw hurricane
employ
worry unemploy
income
conservative
moderate
liberal
high school
some col
col grad
post grad
white
black
hispanic
asian
other race
male
female
other_race
male
female
Description
Worry: Air Pollution
Worry: Pollution of Drinking Water
View of Seriousness of Global Warming
Main Cause of Global Warming
Global Warming Contribution to Strengthened Hurricanes
Current employment status
Worry: unemployment
Household income
Conservative political ideology dummy variable
Moderate political ideology dummy variable
Liberal political ideology dummy variable
High school education or less education level dummy variable
Some college education level dummy variable
College graduate only education level dummy variable
Post-graduate education level dummy variable
Non-hispanic white race dummy variable
Non-hispanic black race dummy variable
Hispanic race dummy variable
Asian race dummy variable
Other race dummy variable
Male gender dummy variable
Female gender dummy variable
Other race dummy variable
Male gender dummy variable
Female gender dummy variable
Scale
A great deal (-1), A fair amount (-2), Only a little (-3), Not at all (-4)
A great deal (-1), A fair amount (-2), Only a little (-3), Not at all (-4)
Generally exagerrated (1), Generally correct (2), Generally underestimated (3)
Effects of pollution from human activities (-1), Natural changes in environment (-2)
Major cause (-1), minor cause (-2), not a cause (-3)
Employed full-time or employed part-time (1), Unemployed but looking for work (0)
A great deal (-1), A fair amount (-2), Only a little (-3), Not at all (-4)
Less than 10K(1),10-20K(2),20-30K(3), . . . ,250k-499k(10),500K and over (11)
1/0
1/0
1/0
1/0
1/0
1/0
1/0
1/0
1/0
1/0
1/0
1/0
1/0
1/0
1/0
1/0
1/0
(1)
worry air
(2)
worry water
Weighted regression
0.267∗∗∗
(0.0372)
0.236∗∗∗
(0.0352)
Ordered probit
0.400∗∗∗
(0.0533)
0.401∗∗∗
(0.0398)
Savings, debt controls
0.186
(0.129)
-0.00608
(0.110)
Local employment growth controls
0.275∗∗∗
(0.0368)
0.244∗∗∗
(0.0275)
Table A3: Robustness checks. Table lists coefficients on the minority dummy variable. All controls from the main specification in Table 2 are included. Row 1 weights
each response by its Gallup-provided weight factor, row 2 uses an ordered probit,
row 3 controls for respondents’ self-reported savings and debt, and row 4 controls
for employment growth in respondents’ counties. Standard errors in parentheses. ∗
p < 0.10, ∗∗ p < 0.05, ∗∗∗ p < 0.01.
22
@FedFRASER