Temporal Trends in Air Pollution Exposure Inequality in Massachusetts

Abstract

Mounting evidence over the past several decades has demonstrated inequitable distribution of pollutants of ambient origin between sociodemographic groups in the United States. Most environmental inequality studies to date are cross-sectional and used proximity-based methods rather than modeled air pollution concentrations, limiting the ability to examine trends over time or the factors that drive exposure inequalities. In this paper, we use 1 km² modeled PM_2.5 and NO₂ concentrations in Massachusetts over an 8-year period and Census demographic data to quantify inequality between sociodemographic groups and to develop a more nuanced understanding of the drivers and trends in longitudinal air pollution inequality. Annual-average population-weighted PM_2.5 and NO₂ concentrations were highest for urban non-Hispanic black populations (11.8 μg/m³ in 2003 and 8.4 μg/m³ in 2010, vs. 11.3 μg/m³ and 8.1 μg/m³ for urban non-Hispanic whites) and urban Hispanic populations (15.9 ppb in 2005 and 13.0 ppb in 2010, vs. 13.0 ppb and 10.2 ppb for urban non-Hispanic whites), respectively. While population groups experienced similar absolute decreases in exposure over time, disparities in population-weighted concentrations increased over time when quantified by the Atkinson Index, a relative inequality measure. Exposure inequalities were approximately one order of magnitude greater for NO₂ compared to PM_2.5, were more pronounced in urban compared to rural geographies, and between racial/ethnic groups compared to income and educational attainment groups. Our results also revealed similar longitudinal PM_2.5 and NO₂ inequality trends using Census 2000 and Census 2010 data, indicating that spatio-temporal shifts in air pollution may best explain observed trends in inequality. These findings enhance our understanding of factors that contribute to persistent inequalities and underscore the importance of targeted exposure reduction strategies aimed at vulnerable populations and neighborhoods.

1. Background

Ambient exposure to nitrogen dioxide (NO₂) and fine particulate matter (PM_2.5) have been associated with a range of adverse health effects including increased risk of asthma and respiratory infections (Brauer et al., 2002; O’Connor et al., 2008; Xing et al., 2016), adverse birth outcomes such as early gestational age and low birth weight (Brauer et al., 2008; Stieb et al., 2012; Zheng et al., 2016), increased risk of autism spectrum disorders (Raz et al., 2015; Volk et al., 2013), and all-cause mortality (Franklin et al., 2008; Shi et al., 2016). Mounting evidence over the past several decades has demonstrated inequitable distribution of exposure to PM_2.5 and NO₂ in the United States among children and older adults, non-Hispanic black and Hispanic populations, low educational attainment and low income populations, potentially contributing to environmental health disparities (Bell and Ebisu, 2012; Brugge et al., 2015; Clark et al., 2014; Morello-Frosch and Lopez, 2006; Su et al., 2009).

However, there are three key limitations in the exposure inequality literature to date. First, much of the environmental inequality (EI) research is cross-sectional, examining environmental inequalities at one point in time. This limits the ability to examine longitudinal trends or the causal mechanisms that drive inequality (Legot et al., 2012; Mohai et al., 2011; Pastor et al., 2004). In particular, there is limited insight about whether disparities are driven by population shifts subsequent to siting of hazardous facilities or roadways, disparate siting practices in poor communities and communities of color, or policies focused on decreasing ambient pollution that simply do not examine distributional consequences. Investigators in both the sociological and environmental health literature argue that residential segregation is a main driver of environmental health disparities (Mohai and Saha, 2015; Morello-Frosch and Lopez, 2006), so demographic shifts over time could have an influence on land use practices, declining social capital and local economies and ultimately, community-level environmental exposures (Mohai and Saha, 2015a; Pastor et al., 2004, 2001). Further, demographic change over time could modify inequalities even in the absence of changes in air quality. Therefore, it is imperative to incorporate demographic time trends in air pollution exposure inequality studies.

Second, a limited number of studies have used quantitative metrics to assess EI over space and time. Quantifiable measures of exposure inequality allow regulators to formally assess patterns of EI and to maximize efficiency in exposure reduction policies that seek to reduce environmental exposures, while simultaneously incorporating social equity into distributional assumptions (Boyce et al., 2016; Harper et al., 2013; Levy et al., 2007, 2006). A handful of environmental studies to date have incorporated formal inequality indices to assess geographic and social distribution of environmental hazards (Boyce et al., 2016; Clark et al., 2014; Levy et al., 2007, 2006; Post et al., 2011; Su et al., 2009). These previous studies have adopted welfare-based or health-based measures of inequality to assess sociodemographic distributions of exposure to a single hazard (Boyce et al., 2016; Clark et al., 2014; Fann et al., 2011; Levy et al., 2006; Post et al., 2011) or cumulative environmental hazards (Su et al., 2009). This paper employs the Atkinson Index (AI) (Atkinson, 1970), a relative measure of inequality, discussed in further detail below. Although some previous studies have used the AI to quantify exposure inequality (Clark et al., 2014; Levy et al., 2009, 2006; Post et al., 2011), these studies focused to a greater extent on understanding the inequality implications of air pollution control strategies, and not on longitudinal patterns of inequality.

Most EI studies examine inequitable distributions of hazardous facilities among population subgroups (Mohai and Saha, 2015a, 2015b). A limited, but growing number of EI studies have examined inequalities with respect to both hazardous facilities and traffic-related air pollution using modeled or measured ambient concentrations. However, many are at coarse geographic resolutions, ignore chemical fate and transport and local meteorological conditions, and do not address longitudinal trends in EI (Clark et al., 2014; Hajat et al., 2015; Kravitz-Wirtz et al., 2016; Mohai and Saha, 2015b; Morello-Frosch and Jesdale, 2006; Pope et al., 2016). Pollutants such as NO₂ and PM_2.5 have significant public health burdens but are not typically dominated by local emissions from hazardous facilities, reinforcing the importance of an exposure-based analytical approach to identify EI occurring at smaller spatial scales.

In this paper, we quantify inequality in modeled ambient PM_2.5 and NO₂ concentrations between racial, ethnic, income and education groups across Massachusetts between 2003 and 2010 using methods to address the three major limitations in this area of research. The work applies a formal inequality index to examine patterns of exposure among rural and urban populations as a means to identify populations most vulnerable to air pollution exposure within the state. The availability of demographic data from the decennial 2000 and 2010 Census at the block group level and modeled ambient air pollution at a 1 km² resolution over an eight-year period provides us the unique opportunity to examine inequalities over time and develop a more nuanced understanding of whether PM_2.5 and NO₂ exposure inequalities are driven by demographic shifts or longitudinal pollution source distribution.

2. Methods

2.1 Data Sources

2.1.1 Ambient air pollution for Massachusetts, 2003–2010

Daily surface PM_2.5 at a 1 km² resolution was modeled from 2003–2010 using a 3-stage statistical modeling approach (Kloog et al., 2014). This modeling approach used a combination of aerosol optical depth (AOD) satellite data retrieved using the multi-angle implementation of atmospheric correction (MAIAC) algorithm, land use and meteorological predictors of variation in surface-PM_2.5, and monitored PM_2.5 concentrations (Kloog et al., 2014). This produced an overall “out-of-sample” R² for daily values of 0.88, and cross validation results produced a slope of observed versus predicted of 0.99. Details of the PM_2.5 prediction models can be found in in Kloog et al. (2014).

We used daily ground NO₂ concentrations that were estimated for the New England region from 2005–2010 at a 1 km² resolution from a combination of ground-level NO₂ data at monitoring sites, satellite Ozone Monitoring Instrument NO₂ vertical column density data, and land use regression (Lee and Koutrakis, 2014). Predictors in mixed effects models included population density, distance to major highways, percent developed area, NO₂ source emissions, elevation, and temperature data. This model produced an R² of 0.79 and cross validation results produced a slope of observed versus predicted of 0.98, demonstrating high predictive reliability. NO₂ model details can be found in Lee and Koutrakis (2014).

2.1.2 Demographic Data

We gathered geographic distributions of race/ethnicity, income, and educational attainment from the US Census and American Community Survey (ACS) at the block group unit of analysis. Measures of educational attainment and income were not collected in the decennial 2010 Census. Therefore, we obtained race/ethnicity data from Census 2010, and measures of income and educational attainment from ACS 2006–2010 5-year estimates. We categorized block groups as rural and urbanized centers according to Census classifications, which rely on population density (Ratcliffe et al., 2016). We utilize Census data at two distinct time periods, 2000 and 2010, rather than at 1-year intervals over the decade under study because the non-decennial 1-year summaries from the ACS are less-reliable, constitute a smaller sample size, and were only collected starting in 2005.

We categorized population characteristics into the following groups:

• Race/ethnicity: individuals in each block group that self-identify as non-Hispanic white, non-Hispanic Black, non-Hispanic Asian, Hispanic or other

• Income: 1999 and 2010 inflation-adjusted median household income as <$20,000/year, $20–35,000/year, $35–50,000/year, $50–75,000/year, and >$75,000/year

• Educational Attainment: individuals in each block group ≥25 years of age with less than a high school degree, high school graduate, postsecondary degree, bachelors and graduate degree

We aggregated daily PM_2.5 and NO₂ concentrations to average annual concentrations. Annual PM_2.5 (years 2003–2010) and NO₂ (years 2005–2010) concentrations were assigned to each block group centroid using the closest 1 km² grid cell centroid for each year over the study period. This exposure assignment method was performed separately for Census 2000 and ACS/Census 2010 block groups using ArcGIS 10.3 (ESRI, Inc.).

2.2 Statistical Analysis

2.2.1 Summary Statistics

We calculated summary statistics for Massachusetts of the number and percentage of individuals and households within each racial/ethnic and education group and the percentage change between 2000 and 2010 stratified by urban (densely developed territories with 50,000 or more people (Census 2000, n=4277; Census and ACS 2010, n=4308)) and rural (any territory not defined as urban (Census 2000, n=654; Census and ACS 2010, n=596)) block groups (Table 1). Median household income in 2010 dollars is also presented for both time points. Due to the small number of block groups categorized by the Census Bureau as “urban clusters,” territories containing between 2,500 and 50,000 residents (Census 2000, n=116; Census and ACS 2010, n=75), these block groups were excluded from stratified analyses.

Table 1.

Massachusetts Census 2000, Census 2010 and ACS 2006–2010 demographic and geographic subpopulation characteristics

	Full State					Urban					Rural
	2000		2010		% Changeb	2000		2010		% Changeb	2000		2010		% Changeb
	#	%	#	%		#	%	#	%		#	%	#	%
Race/Ethnicity
Total	6349 097		6547 629			5337 451		5620 943		5.3	885 599		849 293
Non-Hispanic White	5197 124	81.9	4984 800	76.1	−5.7	4248 651	79.6	4126 251	73.4	−6.2	836 963	94.5	790 098	93.0	−1.5
Non-Hispanic Black	3144 72	5.0	3916 93	6.0	1.0	3012 76	5.6	3803 08	6.8	1.1	992 3	1.1	962 5	1.1	0.0
Non-Hispanic Asian	2370 06	3.7	3474 95	5.3	1.6	2232 62	4.2	3314 79	5.9	1.7	101 59	1.1	150 10	1.8	0.6
Non-Hispanic Other	1731 55	2.7	1959 87	3.0	0.3	1567 07	2.9	1786 92	3.2	0.2	129 36	1.5	148 43	1.7	0.3
Hispanic	4273 40	6.7	6276 54	9.6	2.9	4075 55	7.6	6042 13	10.7	3.1	156 18	1.8	197 17	2.3	0.6
Educational Attainment
Total	4273 275		4382 378			3598 210		3745 562		2.8	600 501		583 138
Less than High School	6510 93	15.2	4958 22	11.3	−3.9	5752 86	16.0	4485 84	12.0	−4.0	649 65	10.8	415 50	7.1	−3.7
High School	1165 489	27.3	1171 725	26.7	−0.5	9788 95	27.2	9963 70	26.6	−0.6	163 986	27.3	157 267	27.0	−0.3
Graduate Post-Secon dary	1038 398	24.3	1036 622	23.7	−0.6	8590 29	23.9	8679 66	23.2	−0.7	160 918	26.8	153 673	26.4	−0.4
Bachelors	8345 54	19.5	9615 63	21.9	2.4	6954 03	19.3	8175 00	21.8	2.5	126 717	21.1	134 937	23.1	2.0
Graduate	5837 41	13.7	7166 46	16.4	2.7	4895 97	13.6	6151 42	16.4	2.8	839 15	14.0	957 11	16.4	2.4
			Full State					Urban					Rural
	Mean	SD	5th	95th		Mean	SD	5th	95th		Mean	SD	5th	95th
Median Household Income
Census 2000, 2010 inflation- adjusted $	70,169	32,225	27,152	127,964		69,110	32,221	26,535	126,745		80,098	31,413	42,402	14,2350
Census 2010, 2010 inflation-adjusted $	70,114	34,262	31,125	132,371		69,070	34,785	20,543	132,197		80,172	29,315	42,024	13,9853

^aIncludes all block groups classified as urbanized, urban cluster, or rural by the U.S. Census. Due to the small number of block groups categorized as “urban cluster,” stratified results for this category are not presented.

^bChange in percent of total population

2.2.2 Calculating Population-Weighted Concentrations across Subpopulations

We calculate block group population-weighted PM_2.5 concentrations for each year from 2003 to 2010, and population-weighted NO₂ concentrations for each year from 2005 to 2010. As corresponding annual population data are not available, we calculate separately using each of Census 2000, Census 2010 and ACS 2006–2010, and we evaluate the influence of using alternative population data. Population-weighted concentrations were calculated for each population demographic group stratified by urban and rural classifications. Block-group PM_2.5 and NO₂ (PM_2.5i, NO_2i) were multiplied by the number of people in each population subgroup (p_i). The subgroup block group values were summed for the state and divided by the total population in each subgroup:

$$\frac{\sum P{M}_{{2.5}_i}pi}{\sum pi}\text{or}\frac{\sum N{O}_{2_i}pi}{\sum pi}$$ (1)

2.2.3 Quantifying Inequality Using the AI

We formally quantify air pollution exposure inequality between population subgroups using the Atkinson Index (AI) (Atkinson, 1970). The AI has been traditionally used as a welfare-based measure of income inequality (Kawachi and Kennedy, 1997), but has since been adopted in the environmental inequality literature (Clark et al., 2014; Levy et al., 2007, 2006). The AI can be decomposed into between-group, within-group and total inequality measures (Lasso de la Vega and Urrutia, 2003; Levy et al., 2006). This feature allows investigators to compare distributions of pollutants between population subgroups, and determine whether total inequality in a population can be explained by disproportionate pollution burden between population subgroups (Harper et al., 2013).

For the purposes of the current analysis, we present only “between-group” inequality, as the focus of this paper is examining trends in environmental inequality between population subgroups given the environmental justice implications. The AI ranges from zero to one, zero indicating no inequality and one indicating complete inequality.

The Between-Group AI can be expressed as:

$$1-{\left({\displaystyle {\sum }_{j=1}^n{f}_i{\left[\frac{{\overline{y}}_j}{\overline{y}}\right]}^{1-\epsilon }}\right)}^{\frac{1}{1-\epsilon }}$$ (2)

where n represents the number of individuals in the population, f_j represents the fraction of the total population in each subgroup, ${\overline{y}}_j$ represents mean exposure of each subgroup, $\overline{y}$ represents the mean exposure over the full population within a given geographic boundary (state, rural or urban) and ε represents an explicit inequality aversion parameter, explained below (Atkinson, 1970). The AI is therefore a relative (as opposed to absolute) measure of inequality, so that proportional changes in exposure across the population would not influence the AI, but additive changes would have an effect. By comparing each subgroup’s weighted exposure to the overall population average exposure within a defined geography, the between-group AI represents the magnitude, in relative terms, of exposure disparities between population subgroups. This is an overall measure of inequality between defined subgroups- it does not explicitly provide information about which of those subgroups are most inequitably exposed.

The inequality aversion parameter is a measure of societal concern about inequality. It determines where relative weights should be placed across the exposure distribution. The parameter ranges from zero to infinity, with increasing values reflecting greater weight on the bottom of the distribution. Unlike income, environmental exposures are worse at higher levels, so we perform all AI calculations with the inverse of the pollution concentrations to allow for interpretable calculations (Harper et al., 2013). All calculations presented here apply an aversion parameter of 0.75, consistent with the literature (Clark et al., 2014; Fann et al., 2011; Levy et al., 2009, 2007, 2006; Post et al., 2011). As a sensitivity analysis, we report the AI for multiple alternative inequality aversion parameters (0.25–2) in the Appendix (Figure A4).

Applying the AI (ε =0.75), we quantified between-group PM_2.5 and NO₂ exposure inequality by applying average annual PM_2.5 concentrations for each year between 2003 and 2010 and average annual NO₂ concentrations for each year between 2005 and 2010, keeping the demographic data constant.

3. Results

3.1 Population Characteristics

State-wide demographic characteristics by block group in 2000 and 2010 are presented in Table 1. Overall, the Massachusetts state population grew by 3.1% between 2000 and 2010. The state contained 81.9% Non-Hispanic white in 2000 and 76.1% in 2010. The Hispanic population increased by 47% from 2000 to 2010, growing from 6.7% of the population to 9.6%. The percent of the population with less than a high school education decreased 3.9 percentage points from 2000 to 2010. Average inflation-adjusted median household income among all Massachusetts block groups was essentially unchanged, although the 10^th percentile value decreased by $3,383 (9.8%) and the 95th percentile value increased by $4,407 (3.4%), indicating growing income inequality.

We observe distinct population distribution changes between block groups categorized as rural or urban. The size of the population grew by 5.3% in urban block groups and decreased by 4.1% in rural block groups. Overall, a higher percentage of racial/ethnic minorities live in urban than in rural block groups. Urban non-Hispanic whites experienced the greatest change of any population group between 2000 and 2010, decreasing by 6.2%. Educational attainment distributions are similar between rural and urban block groups in both 2000 and 2010, and both experienced a decrease in the population with less than a high school education. In general, median household incomes were higher in rural than in urban block groups, and the growing income inequality was more pronounced in urban areas.

3.2 Population Weighted Concentrations

Based on modelled PM_2.5 and NO₂, we find that average annual PM_2.5 concentrations across the state decreased by 35% between 2003 and 2010, and that average annual NO₂ concentrations decreased by 24% between 2005 and 2010. Concentrations were consistently lower in rural than urban areas, but patterns of change remained the same between the two strata (Figures 1a and 1b).

Figure 1.

a. Annual Average PM_2.5 Concentrations in Massachusetts, 2003–2010. b. Annual Average NO₂ Concentrations in Massachusetts, 2005–2010

Tables 2 and 3 display snapshots of subgroup population-weighted PM_2.5 and NO₂ concentrations, along with absolute and relative (percent) change in exposure over the study period for the full state and rural/urban classifications. Population-weighted concentrations for PM_2.5 in 2003 and NO₂ concentrations in 2005 were calculated using the Census 2000 population, approximating the spatial and demographic distributions of the population in those years. For PM_2.5 and NO₂ population-weighted concentrations in 2010, the Census 2010 and ACS 2006–2010 populations were used to characterize the population distribution (Tables 2 and 3). Figures 2a–f display population weighted trends in PM_2.5 and NO₂ concentration for each year between 2003 and 2010 using demographic data from Census 2010 and ACS 2006–2010, allowing us to isolate the effects of changing concentrations from any sociodemographic shifts. Longitudinal changes in population-weighted concentrations using demographic data from Census 2000 can be found in the Appendix (Figures A1a–A1f).

Table 2.

Population-weighted annual average PM_2.5 (μg/m³) Concentrations by Census 2000, Census 2010 and ACS 2006–2010 demographic and geographic subpopulations

		Full State				Urban				Rural
	PM2.5 2003, Census 2000	PM2.5 2010, Census 2010	Absolute Changea	Relative Changeb	PM25 2003, Census 2000	PM25 2010, Census 2010	Absolute Changea	Relative Changeb	PM25 2003, Census 2000	PM25 2010, Census 2010	Absolute Changea	Relative Changeb
Race/Ethnicity
Total	11.2	8.0	−3.2	−28.7	11.4	8.1	−3.2	−28.3	10.1	6.8	−3.3	−32.7
non-Hispanic white	11.1	7.8	−3.2	−29.2	11.3	8.1	−3.2	28.6	10.1	6.8	−3.3	−32.7
non-Hispanic black	11.7	8.4	−3.4	−28.6	11.8	8.4	−3.4	−28.6	10.4	6.9	−3.5	−33.6
non-Hispanic Asian	11.6	8.2	−3.4	−29.0	11.7	8.3	−3.4	−28.8	10.8	7.0	−3.8	−35.0
non-Hispanic other	11.4	8.2	−3.3	−28.6	11.6	8.3	−3.3	−28.4	10.2	6.9	−3.3	−32.6
Hispanic	11.6	8.4	−3.2	−27.4	11.6	8.5	−3.2	−27.3	10.5	6.9	−3.6	−34.4
	PM2.5 2003 Census 2000	PM2.5 2010, ACS 2010	Absolute Changea	Relative Changeb	PM2.5 2003, Census 2000	PM2.5 2010 ACS 2010	Absolute Changea	Relative Changeb	PM2.5 2003, Census 2000	PM2.5 2010, ACS 2010	Absolute Changea	Relative Changeb
Educational Attainment
Total	11.2	7.9	−3.2	−28.8	11.4	8.1	−3.2	−28.4	10.1	6.8	−3.3	−32.8
< High School	11.3	8.2	−3.1	−27.6	11.5	8.3	−3.1	−27.3	10.1	6.8	−3.4	−33.2
High School Grad	11.1	7.9	−3.2	−28.7	11.3	8.1	−3.2	−28.2	10.0	6.8	−3.3	−32.7
Post-Secondary	11.1	7.9	−3.2	−28.9	11.3	8.1	−3.2	−28.5	10.0	6.8	−3.3	−32.5
Bachelors	11.2	7.9	−3.3	−29.3	11.3	8.1	−3.3	−28.8	10.2	6.8	−3.4	−33.1
Masters	11.2	8.0	−3.3	−29.0	11.4	8.2	−3.3	−28.6	10.2	6.8	−3.4	−33.0
Median Household Income
Total	11.2	8.0	−3.2	−28.7	11.4	8.2	−3.2	−28.3	10.1	6.8	−3.3	−32.6
<20,000	11.4	8.2	−3.1	−27.6	11.5	8.4	−3.2	−27.4	10.2	6.9	−3.3	−32.4
20–35,000	11.2	8.1	−3.2	−28.1	11.4	8.2	−3.2	−27.8	10.0	6.8	−3.2	−32.0
35–50,000	11.2	8.0	−3.2	−28.3	11.4	8.2	−3.2	−28.0	10.0	6.8	−3.2	−31.7
50–75,000	11.1	8.0	−3.2	−28.3	11.3	8.2	−3.2	−28.1	10.0	6.8	−3.2	−32.1
>75,000	11.1	7.9	−3.3	−29.3	11.3	8.1	−3.2	−28.7	10.2	6.8	−3.4	−33.5

^aAbsolute change in concentrations (μg/m³)

^bPercentage change in concentrations

Table 3.

Population-weighted annual average NO₂ (ppb) Concentrations by Census 2000, Census 2010 and ACS 2006–2010 demographic and geographic subpopulations

		Full State				Urban				Rural
	NO2 2005, Census 2000	NO2 2010, Census 2010	Absolute Changea	Relative Changeb	NO2 2005, Census 2000	NO2 2010, Census 2010	Absolute Changea	Relative Changeb	NO2 2005, Census 2000	NO2 2010, Census 2010	Absolute Changea	Relative Changeb
Race/Ethnicity
Total	13.2	10.4	−2.8	−21.4	13.7	10.8	−2.9	−21.2	10.5	7.9	−2.6	−24.8
non-Hispanic white	12.7	9.8	−2.9	−23.0	13.0	10.2	−2.9	−22.2	10.4	7.9	−2.5	−24.4
non-Hispanic black	14.9	11.8	−3.1	−20.8	14.8	12.0	−2.9	−19.3	10.9	7.5	−3.3	−30.8
non-Hispanic Asian	15.8	12.3	−3.5	−22.2	15.5	12.5	−3.0	−19.2	11.6	7.5	−4.0	−35.0
non-Hispanic other	14.9	11.5	−3.3	−22.5	14.8	11.9	−2.9	−19.6	10.6	7.8	−2.8	−26.3
Hispanic	15.8	12.8	−3.0	−19.2	15.9	13.0	−2.9	−18.0	11.2	7.6	−3.6	−32.0
	NO2 2005 Census 2000	NO2 2010, ACS 2010	Absolute Changea	Relative Changeb	NO2 2005, Census 2000	NO2 2010, ACS 2010	Absolute Changea	Relative Changeb	NO2 2005, Census 2000	NO2 2010, ACS 2010	Absolute Changea	Relative Changeb
Educational Attainment
Total	13.1	10.3	−2.8	−21.7	13.7	10.7	−3.0	−21.7	10.4	7.9	−2.6	−24.5
< High School	14.5	11.9	−2.7	−18.3	15.1	12.3	−2.8	−18.5	10.6	8.0	−2.6	−24.4
High School Grad	13.0	10.2	−2.8	−21.5	13.5	10.6	−2.9	−21.5	10.4	7.9	−2.5	−23.8
Post-Secondary	12.5	9.7	−2.8	−22.4	13.0	10.1	−2.9	−22.4	10.4	7.9	−2.5	−24.0
Bachelors	12.9	10.1	−2.9	−22.0	13.5	10.5	−3.0	−22.1	10.5	7.9	−2.6	−24.7
Masters	13.3	10.5	−2.8	−21.2	13.9	10.9	−2.9	−21.1	10.5	7.8	−2.8	−26.2
Total	13.3	10.5	−2.8	−21.4	13.9	10.9	−3.0	−21.4	10.4	7.9	−2.6	−24.5
Median Household Income
<20,000	14.4	11.7	−2.7	−18.5	15.0	12.2	−2.8	−18.8	10.5	7.8	−2.6	−24.4
20–35,000	13.6	10.9	−2.7	−20.1	14.2	11.3	−2.9	−20.1	10.3	7.8	−2.5	−23.8
35–50,000	13.4	10.6	−2.7	−20.3	13.9	11.1	−2.8	−20.4	10.3	7.8	−2.5	−24.0
50–75,000	13.0	10.4	−2.6	−20.2	13.6	10.8	−2.8	−20.3	10.4	7.9	−2.6	−24.7
>75,000	12.7	9.9	−2.8	−22.1	13.2	10.3	−2.9	−21.9	10.6	7.9	−2.8	−26.2

^aAbsolute change in concentrations (ppb)

^bPercentage change in concentrations

Figure 2.

a–f. Population-Weighted Annual Average PM_2.5 (a–c) and NO₂ (d–f) concentrations by Census 2010 and ACS 2006–2010 Demographic and Geographic Subpopulations

3.2.1 PM_2.5

Overall, weighted PM_2.5 exposures in 2003 across the state ranged from 11.1 to 11.7 μg/m³ across racial/ethnic groups, with ranges in urban block groups from 11.3 to 11.8 μg/m³ and in rural block groups from 10.1 to 10.8 μg/m³ (Table 2). In 2010, weighted PM_2.5 exposures ranged across racial/ethnic groups ranged from 7.8 to 8.4 μg/m³, 8.1 to 8.5 μg/m³ and 6.8 to 7.0 μg/m³ across the state and among urban and rural populations, respectively. Across the state in 2003, PM_2.5 concentrations were highest for the non-Hispanic black (11.7 μg/m³) population among racial/ethnic groups, those with less than a high school education (11.3 μg/m³) among education groups, and those with incomes less than $20,000 per year (11.4 μg/m³) among income groups. Among racial/ethnic groups in 2003, the greatest difference in population weighted concentration was between non-Hispanic whites (11.1 μg/m³) and non-Hispanic blacks (11.7 μg/m³), and in 2010 the greatest difference was between non-Hispanic whites (7.8 μg/m³) and both Hispanic and non-Hispanic black (8.4 μg/m³) populations. These patterns were present in urban but not rural block groups. The absolute decrease in PM_2.5 over time was relatively homogenous across all population groups.

Figures 2a–c display population weighted trends in PM_2.5 exposures from 2003 to 2010, holding Census 2010 and ACS 2006–2010 data constant. Among racial and ethnic groups, non-Hispanic Asian populations experienced the largest decrease in PM_2.5 exposures between 2003 and 2010 in both urban (28.8%) and rural (33.6%) locations, whereas the urban Hispanic population experienced the lowest percentage decrease (27.3%). Among rural income groups, the greatest decrease in PM_2.5 was observed for median incomes above $75,000 per year (33.5%), a pattern that differed from urban income groups. In general, the population groups with the highest exposures in 2003 experienced lower relative decreases in exposures over time, consistent with similar absolute reductions across populations. Holding the Census 2000 population constant over annual PM_2.5 concentrations reveals similar results (Figures A1a–A1c).

3.2.2 NO₂

Table 3 displays population weighted NO₂ concentrations, absolute and percent decrease in exposure for the full state, and patterns of exposure stratified by sociodemographic characteristics and rural/urban status. Across the state in both 2005 and 2010, NO₂ concentrations were highest for Hispanic populations (15.8 ppb in 2005 and 12.8 ppb in 2010), those with less than a high school education (14.5 ppb in 2005 and 11.9 ppb in 2010) and households in the lowest income bracket (14.4 ppb in 2005 and 11.7 ppb in 2010). Patterns were identical in urban block groups. However, in rural block groups, the non-Hispanic Asian population experienced the highest exposure in 2005 (11.6 ppb), while non-Hispanic whites (7.9 ppb) had the highest NO₂ burden in 2010. Those in the highest income bracket in rural block groups also experienced the highest NO₂ concentrations in both 2005 and 2010.

Figures 2d–f display trends in population-weighted NO₂ concentrations from 2005 to 2010 using demographic data from Census 2010 and ACS 2006–2010. In general, the rate of NO₂ concentration decrease was greater in rural than urban block groups. Similar to PM_2.5, population-weighted NO₂ exposure inequality existed for urban, but not rural, sociodemographic groups. Trends in population-weighted NO₂ exposure make clear that patterns of inequalities persisted from 2005 to 2010, and that urban racial/ethnic minorities, low income and education groups remained the highest exposure groups. Results are similar when applied to the Census 2000 population (Figures A1d–A1f).

3.3 Atkinson Index

3.3.1 PM_2.5

We estimate the AI for each year, separately using Census 2000 and Census 2010/ACS 2006–2010, to determine how exposure inequality has evolved over time and whether this is related to concentration patterns or changing demographics. Overall, AI trends and values are relatively insensitive to the choice of population data (Figure 3). The principal difference between the two demographic years is demonstrated in modestly higher rural exposure inequality trends among racial/ethnic and income groups for the Census 2000 compared to the Census 2010 population. These results indicate that both population mobility and shifting PM_2.5 distributions contribute to rural exposure inequality trends. Exposure inequality trends among all subgroups living in urban block groups are nearly identical between 2000 and 2010, indicating that shifting PM_2.5 distributions (and not population mobility) are likely driving observed exposure inequality trends in urban areas.

Figure 3.

Between-Group Inequality in Population-Weighted Annual Average PM_2.5 and NO₂ Concentrations. a displays results for PM_2.5 and NO₂ inequality for the Census 2000 Demographic and Geographic Subpopulations. b displays results for PM_2.5 and NO₂ inequality for the Census 2010 and ACS 2006–2010 Demographic and Geographic Subpopulations.

Between-sociodemographic group PM_2.5 inequality using the AI reveals peaks of increased and decreased inequality over time as a result of PM_2.5 concentration distributions in urban block groups, and a slight decreasing trend among rural block groups (Figure 3a). We additionally find that inequality is generally greater in magnitude, especially after 2005, in urban block groups. Although the AI values are generally low, they are consistently higher for racial/ethnic groups compared to inequality among income and education groups. The AI results seen here are explained by non-Hispanic black and Hispanic populations, low-income, and low educational attainment populations consistently experiencing a greater PM_2.5 burden than the other racial/ethnic, income and education groups (Table 2). As all sociodemographic groups experience a similar absolute decrease in exposure, the lowest exposed groups, such as non-Hispanic whites, undergo greater relative rates of exposure decline over time.

3.3.2 NO₂

We observe distinctly different patterns in quantified NO₂ inequality as compared to PM_2.5 inequality (Figure 3). AI values are generally low but approximately one order of magnitude greater for NO₂ compared to PM_2.5. Between 2005 and 2010 there is a slight increase in NO₂ inequality among all population strata located in urban block groups. Inequality was greatest for racial/ethnic subpopulations in urban block groups, which also experienced the greatest rate of increase in AI over time. AI results were similar between when using Census 2000 and Census 2010/ACS 2006–2010 population data.

4. Discussion

Our study builds on previous environmental inequality analyses that use measures of proximity or coarsely-resolved measures of air pollution exposure and investigate inequality at one point in time by incorporating longitudinal Census data and pollution concentrations. This work additionally builds on the current literature by employing a novel application of the AI to formally quantify inequality between population groups over time.

Although modeled and monitored air pollution data have demonstrated longitudinal reductions in concentrations, our highly-resolved and stratified analyses provided some novel insights with respect to exposure inequalities. For example, we found distinct differences in population-weighted concentration patterns and trends over time between PM_2.5 and NO₂ and between urban and rural geographic areas. Urban areas contain greater densities of low-income, non-white and low-educational attainment populations and PM_2.5 and NO₂ pollution sources, contributing to some exposure heterogeneity and potential inequalities. Greater concentration of urban air pollution sources is reflected in our findings of non-Hispanic blacks, individuals with lower educational attainment, and households with an annual income of <$20K as the most burdened population groups for both NO₂ and PM_2.5 concentrations throughout the state.

That said, PM_2.5 concentrations are more regional than local in nature because they are derived from a wide variety of sources, with a strong contribution from secondary pollutant formation and long-range transport (Zheng et al., 2002). NO₂ is strongly linked to automobiles and other mobile sources and tends to exhibit high intra-urban variability. As such, it has greater potential for exposure inequalities in urban settings. PM_2.5 is a regionally-based pollutant, exhibiting less spatial variability in urban areas, leading to smaller exposure disparities compared to NO₂ (Clougherty et al., 2008). These pollutant-specific characteristics are reflected in our finding of NO₂ inequality that is greater in magnitude than PM_2.5 inequality. Higher NO₂ inequality growth rates within urban areas, especially between racial/ethnic groups, may further be explained by increased local source emissions, such as higher traffic counts over time or increased transportation infrastructure in Boston neighborhoods containing high proportions of non-Hispanic black and Hispanic populations (Brugge et al., 2015; Levy et al., 2001).

In rural settings, NO₂ exposures were disproportionately higher for Hispanic populations, individuals with lower educational attainment, and households with lower income in urban block groups, but higher for non-Hispanic Asians and the wealthiest population groups in rural block groups. This could reflect the fact that roadway proximity tends to decrease property value in urban areas, but may potentially increase them in rural areas (Bateman et al., 2001; Lake et al., 1998). Analyses that did not stratify by urban/rural status would not appropriately characterize exposure inequality or capture key between-group differences.

We additionally found fluctuations in PM_2.5 inequality and increasing trends in between-group NO₂ inequality, despite similar absolute rates of PM_2.5 and NO₂ decline across population groups. Uniform absolute reductions are beneficial to all but do not decrease exposure inequalities, and in fact, tend to increase them for metrics such as the AI given growing relative differences. Similar AI trends using Census 2000 and Census 2010 populations indicates that sociodemographic mobility is not the main driver of urban PM_2.5 and statewide NO₂ inequality trends, although it could remain a contributing factor. It would be informative for future studies to examine inequality trends using annual demographic data where available, holding ambient concentrations constant.

Because most environmental inequality studies rely on cross-sectional data, they do not inform our understanding of the components that contribute to changing inequality over time (Bell and Ebisu, 2012; Lopez, 2002; Miranda et al., 2011; Pastor et al., 2004; Rosofsky et al., 2014). Our application of the AI to characterize exposure inequality addresses this literature gap by separately examining population and air pollution patterns, to determine which best explains changing inequality. To our knowledge, only one study to date has applied the AI to describe spatial patterns of pollution across sociodemographic characteristics and between rural and urban areas (Clark et al., 2014). Clark et al. (2014) findings of population-weighted racial/ethnic and income disparity for NO₂ exposure nationwide were similar to our results: nonwhite and low-income populations experienced the greatest burden of NO₂ exposure, and these disparities were more pronounced in large urban areas than rural areas.

A handful of studies within the environmental inequality literature have also moved to address this gap (Kravitz-Wirtz et al., 2016; Mohai and Saha, 2015b; Pastor et al., 2001). One recent study by Kravitz-Wirtz et al. (2016) examined trends in racial and ethnic disparities in exposure to neighborhood air pollution across the U.S., while controlling for individual and neighborhood-level changes over time. The authors found that black and Hispanic participants were disproportionately exposed to higher concentrations of NO₂, PM_2.5 and PM₁₀ compared to white participants, and that concentrations decreased for all racial and ethnic groups over time. In contrast to our findings, rate of decline in PM_2.5 and NO₂ exposure among black and Hispanic participants were more pronounced than for white participants. The authors hypothesized that these findings are explained by more rapid decreases in pollution in urban areas, where black and Hispanic participants of the study reside. However, our study found the opposite effect, with concentrations of PM_2.5 and NO₂ falling more rapidly in rural areas, and for the non-Hispanic white population.

In general, our AI values are quite small, and we observed small absolute differences in PM_2.5 and NO₂ concentrations between population groups. However, AI values cannot be reasonably compared across contexts (i.e., income spans many orders of magnitude, whereas ambient air pollution has a narrower range within a state), and are most meaningful for comparisons over time or between pollutants analyzed similarly. In addition, the exposure differences may be large enough to contribute to health disparities (Atkinson et al., 2014; Brauer et al., 2008; Clark et al., 2014; Shi et al., 2016). For instance, in a recent study by Shi et al. (2016), all-cause mortality increased by 0.9% per μg/m³ increase in long-term PM_2.5 concentrations even when restricted to ambient concentrations below 10 μg/m³. The 0.6 μg/m³ difference in exposure in urban areas in 2010 for non-Hispanic whites versus Hispanics would therefore translate into a 0.5% increase in mortality rates, all else being equal. Further, the population subgroups found to have the highest population-weighted PM_2.5 and NO₂ concentrations also tend to have higher baseline rates of asthma and cardiovascular disease, leaving them more vulnerable to persistent, longitudinal air pollution exposure (Crain et al., 1994; Jones et al., 2009; O’Neill et al., 2003).

Our findings demonstrating inequitable pollution exposure by SES and race/ethnicity are supported by evidence of environmental inequality that is firmly established in the academic literature (Lopez, 2002; Mohai and Bryant, 1992; Mohai and Saha, 2015a; Morello-Frosch and Lopez, 2006). Further, a growing number of studies have demonstrated environmental inequality specific to PM_2.5 and NO₂ in Massachusetts and nationwide using Census data (Clark et al., 2014; Miranda et al., 2011; Yanosky et al., 2008). For instance, Miranda et al. (2011) used an air quality ranking approach to assess environmental justice dimensions of air pollution exposure, finding that the proportion of non-Hispanic black residents in the 20% of counties in the United States with the poorest air quality was twice that in those counties with the most favorable air quality. Yanosky et al. (2008) evaluated whether predicted NO₂ concentrations are associated with socioeconomic position, after controlling for spatial autocorrelation in Worcester, Massachusetts. They found that block group NO₂ concentrations exhibit a significant negative association with median household income, and that rates of poverty and low educational attainment populations rose by 3.1% and 3.4%, respectively, with every one standard deviation increase in block group mean NO₂.

Despite employing novel inequality-based methods using data of high temporal and geographic resolution, there are some limitations that merit discussion. The use of Census data restricts our ability to examine disparities at the individual/household level. Using personal monitors is not feasible at this scale, so we assigned modeled PM_2.5 and NO₂ concentrations to each block group to approximate individual exposure, thereby limiting potential variability in exposure across the population. Our results consequently do not incorporate individual mobility or characteristics that may provide a more comprehensive understanding of the drivers of inequality. However, the inputs used to assign block-group level exposures are advantageous over proximity-based and aggregation methods that ignore chemical fate and transport and local meteorological conditions (Chakraborty et al., 2011; Lucier et al., 2011; Mohai et al., 2011; Pastor et al., 2001). These predictions are also an improvement over EI studies that use predicted concentrations over coarse geographic and temporal resolutions (Hajat et al., 2015; Kravitz-Wirtz et al., 2016; Mohai and Saha, 2015b; Morello-Frosch and Jesdale, 2006; Pope et al., 2016). A 1 km² resolution is adequate for regionally-based pollutants, such as PM_2.5, but may ignore local hotspots for locally-based pollutants such as NO₂. Conversely, smaller geographic resolutions may introduce bias related to individual mobility (Setton et al., 2011)

We acknowledge that the temporal misalignment of PM_2.5 (years 2003–2010) and NO₂ (years 2005–2010) with Census data for the year 2000 or 2010 prevents a precise characterization of exposure patterns over time. However, as discussed above, the relative stability of our inequality measures across different population data indicates that this is a minor source of error.

As a final overall limitation, we only studied inequalities in outdoor ambient air pollution exposures; low socioeconomic status groups may be disproportionately exposed to indoor-generated exposures or from indoor exposure to outdoor pollutants due to older, leakier housing stock (Adamkiewicz et al., 2011). Taking into account the full exposure profile of both indoor and outdoor-generated air pollutants may reveal a more striking characterization of exposure inequality between population groups.

5. Conclusion

Despite overall reductions in ambient air pollution concentrations and decreased industrialization, we found that air pollution inequalities have slightly increased over time when measured on a relative scale, and that group-specific concentrations are most disparate between racial/ethnic groups. Greater inequalities in urban areas, where there is often substantial segregation, reinforces the importance of targeted exposure reduction strategies within vulnerable populations and neighborhoods. Ultimately, there is a complex dynamic wherein changing sociodemographics over time may impact land use decisions, enforcement policy measures, and other factors influencing emissions. To complement these findings, more studies that utilize longitudinal, individual-level data are needed to understand population mobility and individual factors that affect exposure disparities.

Highlights.

• We characterized longitudinal PM_2.5 and NO₂ inequality trends across Massachusetts

• Exposure inequality increased in urban, but not rural areas

• NO₂ exposure inequality was greater in magnitude than PM_2.5 inequality

• Observed inequality trends likely driven by spatiotemporal pollution shifts