To our knowledge, this is the first study of how exposures to PM2.5
components may differ by population for race/ethnicity, age, and SES. In an earlier study, Marshall (2008)
from diesel sources and hexavalent chromium based on individual-level exposure estimates in California and found higher exposures for persons who were younger (< 7 years vs. > 80 years of age), less educated (< high school vs. college), or nonwhite. Previous studies compared exposure levels of various populations for other pollutants, including PM2.5
. U.S. counties in the lowest quantile of air quality had a higher fraction of non-Hispanic blacks and persons in poverty than did counties in the highest quantile of air quality for PM2.5
and ozone (Miranda et al. 2011
). In the same study, the investigators found that 20% of the counties with the worst air quality for PM2.5
and for ozone had more persons > 64 years of age and more children < 5 years of age, respectively. Areas with parks or adjacent to parks in Los Angeles, California, had higher NO2
levels in low SES or high minority neighborhoods (Su et al. 2011
). In the United States, Hispanic, African-American, or Asian/Pacific Islander women had higher air pollution exposures during pregnancy than did white women after adjusting for education and other factors, based on an air pollution index that incorporated levels of PM with aerodynamic diameter ≤ 10 μm (PM10
), ozone, carbon monoxide, NO2
, and sulfur dioxide (Woodruff et al. 2003
). In that study, Woodruff et al. (2003)
found that lower education was associated with higher pollution levels, after adjustment for race/ethnicity. In Hamilton, Ontario, Canada, levels of total suspended particles (TSP) were higher in census tracts with more Latin Americans or fewer Asian Canadians, with no observable trends between TSP and black Canadians, after adjusting for SES (Buzzelli and Jerrett, 2004
). In the same area, Jerrett et al. (2001)
observed that TSP levels were higher in census tracts with higher dwelling values and lower income.
In Tampa, Florida, blacks, Hispanics, and persons in poverty resided in neighborhoods closer to toxic release inventory (TRI) sites, whereas whites lived closer to air pollutant monitors (Stuart et al. 2009
). In California, census tracts within a mile of TRI facilities had higher fractions of minorities, especially Latinos, lower rates of home ownership, and lower incomes (Pastor et al. 2004
). In Orange County, Florida, Chakraborty and Zandbergen (2007)
reported that Hispanic or black children were more likely to live or attend school near TRI sources than were white children. In regions of West Virginia, Louisiana, and Maryland, African Americans lived closer to TRI sites than did whites (Perlin et al. 2001
Our estimates are consistent with these overall trends, indicating the highest PM2.5 exposures for non-Hispanic blacks, the least educated, the unemployed, and those in poverty. However, overall differences were small in magnitude, with the largest difference at 9.9% higher for non-Hispanic blacks than for whites. We estimated larger disparities for exposures to PM2.5 components than to PM2.5. Whereas PM2.5 levels for those without a high school education were 6.2% higher than those with college, Zn levels were 29% higher. Unemployed persons had 2.3% higher PM2.5 than employed persons, but 11% higher levels for V. Similarly, estimated differences among race/ethnicity, earnings, or age categories were larger for many components than for PM2.5. The directions of the associations were different among components. For example, those in the lowest earnings category (< $15,000/year) had higher levels than those earning ≥ $50,000/year for seven components (18% higher for Al) and lower levels for seven components (26% lower for Ni).
We used community-level exposures for census tracts. More precise measures would incorporate spatial heterogeneity (Peng and Bell 2010
), as well as daily activity patterns, indoor exposures (e.g., environmental tobacco smoke), inhalation rates, and occupational exposures at the individual level. Many of these factors (e.g., occupation) may differ by population. Exposures were estimated from ambient monitors, and thus do not reflect the personal exposures of all individuals within the census tract.
Our research does not disentangle demographic characteristics of race/ethnicity, education, unemployment, poverty, and earnings; and many population characteristics co-vary [see Supplemental Material, Table S6
) for correlations]. For example, race, education, earnings, and poverty were correlated. Future work could examine patterns in population characteristics in relation to PM2.5
component exposures and patterns related to community factors, such as urbanicity and property values.
Only 215 census tracts had PM2.5
component monitors meeting the inclusion criteria, covering 0.3% of the population. The monitor coverage hinders ability to fully investigate equity issues, especially for rural populations, which likely have different characteristics. As population demographics and chemical composition of particles differ dramatically by region (Bell et al. 2007
), the geographical distribution of monitors could affect results. In this study, 37% of monitors were in the South (defined by U.S. Census regions), 27% in the Midwest, 19% in the West, and 17% in the Northeast. Future research may consider alternative methods of estimating exposure, such as air quality modeling and satellite imagery (Anderson et al. 2012
; Bell 2006
; Boldo et al. 2011
; Fann et al. 2012
), to estimate exposures for a larger population.
Our results show that populations potentially at risk for higher exposures to components do not appear to be underrepresented in areas with monitors compared with areas without monitors. This contrasts with the study by Miranda et al. (2011)
that found that U.S. counties without sufficient monitoring for PM2.5
and ozone had fewer non-Hispanic blacks, Hispanics, and persons < 5 years of age and a higher percentage of persons > 64 years of age. Our findings may differ because of the use of census tracts (median land area = 5.06 km2
; SD = 571 km2
) rather than counties (median land area = 1,582 km2
; SD = 3,375 km2
) and because of differences between monitoring networks for PM2.5
components. Other studies also have shown links between population characteristics and monitoring networks. In São Paulo, Brazil, areas with higher SES were more likely to have PM10
and ozone monitors (Bravo and Bell 2010
Additional challenges in this area of research include the choice and interpretation of SES indicators, because true SES relates to historical conditions, full sources of income, as well as access to resources beyond official earnings, neighborhood-level SES, insurance, access to health care, use of health care systems, and social networks (Bell et al. 2002
; O’Neill et al. 2003
). The interpretation of SES indicators can vary by region or subculture. Subjective measures of SES include factors such as satisfaction with position, comparison to peers, and perception of financial security. Perceived and actual SES may differ and can have different trends by population (Brown et al. 2008
). Traditional measures of SES (e.g., income, education) can be supplemented with subjective social status measures, which in some cases may be more closely linked to health outcomes than to conventional measures (Dennis et al. 2012
; Singh-Manoux et al. 2003
The 14 PM2.5 chemical components investigated here were selected because they contribute ≥ 1% to PM2.5 total mass and/or were found to be potentially harmful to health in earlier studies. However, the full health impacts of various particle mixtures and the identities of the most harmful components or set of components are unknown. A further complication is that all components come from multiple sources, although some components are more strongly linked to some sources than to others (e.g., Ni and V from oil combustion, SO42– from coal combustion, Si from road dust).
A growing body of scientific literature, including epidemiological and toxicological studies, indicates health associations with various PM2.5
chemical components (U.S. EPA 2009
). For example, results of toxicological studies using animal models and human-cell cultures suggest the possibility of adverse respiratory effects for Zn (Gerlofs-Nijland et al. 2007
; Wu et al. 2003
), Al (Graff et al. 2007
), V (Veranth et al. 2007
(Riley et al. 2005
), and NO3–
(Huang et al. 2003
). Animal models have shown associations with cardiovascular outcomes, such as for zinc (Bagaté et al. 2004
). As additional information becomes available on which chemical components and related sources are most harmful, future studies could examine how such exposures differ by population.