Air Pollution and Individual and Neighborhood Socioeconomic Status: Evidence from the Multi-Ethnic Study of Atherosclerosis (MESA)

Background: Although research has shown that low socioeconomic status (SES) and minority communities have higher exposure to air pollution, few studies have simultaneously investigated the associations of individual and neighborhood SES with pollutants across multiple sites. Objectives: We characterized the distribution of ambient air pollution by both individual and neighborhood SES using spatial regression methods. Methods: The study population comprised 6,140 participants from the Multi-Ethnic Study of Atherosclerosis (MESA). Year 2000 annual average ambient PM2.5 and NOx concentrations were calculated for each study participant’s home address at baseline examination. We investigated individual and neighborhood (2000 U.S. Census tract level) SES measures corresponding to the domains of income, wealth, education, and occupation. We used a spatial intrinsic conditional autoregressive model for multivariable analysis and examined pooled and metropolitan area–specific models. Results: A 1-unit increase in the z-score for family income was associated with 0.03-μg/m3 lower PM2.5 (95% CI: –0.05, –0.01) and 0.93% lower NOx (95% CI: –1.33, –0.53) after adjustment for covariates. A 1-SD–unit increase in the neighborhood’s percentage of persons with at least a high school degree was associated with 0.47-μg/m3 lower mean PM2.5 (95% CI: –0.55, –0.40) and 9.61% lower NOx (95% CI: –10.85, –8.37). Metropolitan area–specific results exhibited considerable heterogeneity. For example, in New York, high-SES neighborhoods were associated with higher concentrations of pollution. Conclusions: We found statistically significant associations of SES measures with predicted air pollutant concentrations, demonstrating the importance of accounting for neighborhood- and individual-level SES in air pollution health effects research. Citation: Hajat A, Diez-Roux AV, Adar SD, Auchincloss AH, Lovasi GS, O’Neill MS, Sheppard L, Kaufman JD. 2013. Air pollution and individual and neighborhood socioeconomic status: evidence from the Multi-Ethnic Study of Atherosclerosis (MESA). Environ Health Perspect 121:1325–1333; http://dx.doi.org/10.1289/ehp.1206337


Supplemental Material, Methods
In this supplemental methods section we provide additional detail on several SES measures and modeling approaches used in this study.

Socioeconomic status
Individual SES variables: In addition to the four individual SES variables described in the methods section (income, wealth, education and occupation), we also examined the incomewealth index and working outside the home. The income-wealth index was specified as a 9 point scale (0 being the lowest level of income and no assets and 8 being the highest level of income and all 4 assets) was the sum of a 5 category income variable and the 5 point wealth index. The income-wealth index was described in depth previously (Hajat et al. 2010). Participants who reported being employed either full or part-time or who were retired but working for pay or volunteering were classified as working outside the home, while those who never worked outside the home, were employed but not currently working, were unemployed or retired and not working were classified as not working outside the home. Work outside the home and the income-wealth index are not presented here because the other individual SES variables were more representative of the 4 domains of SES.
Neighborhood SES index: The 16 variables included in the principal components analysis were: percent of occupied housing units, percent of housing units that are owner occupied out of total housing units, median value of occupied housing units, percent of housing units without telephone, percent of housing units without vehicle, percent living in same house in 1995, percent of person 25 or older with at least high school education, percent of persons 25 or older with at least a Bachelor's degree, percent unemployed among civilians 16 and over in the labor force, percent of civilians 16 and over not in the labor force, percent with management, professional and related occupation, median household income, percent households with household income greater than $50,000, percent of households with interest, dividends, or net rental income, percent of households with public assistance and percent of persons below the poverty level.

Modeling approaches
Priors for ICAR model: Prior specifications for the conditional and marginal variance were set to 0.05. Prior specification can be difficult for the conditional variance because of its conditional rather than marginal interpretation (Fong et al. 2010). Thus a range of other priors were tested; they are: 0.5, 0.1, 0.01, 0.005 and 0.001. These priors were used for both the conditional and marginal variances. In general parameter estimates changed only at the third decimal place and beyond. These small changes would not affect the conclusions of the study.
Multilevel Models: In addition to the spatial ICAR models, we used multilevel models to account for the clustering of individuals within neighborhoods. Multilevel models assume that people living within the same census tracts are more similar to each other than persons living in different tracts, but do not make assumptions about the relationship between census tracts as ICAR models do. Multilevel models are similar to ICAR models in that they are also random effects models. αεβThe equation for the multilevel model is below: where Y ij = air pollution concentration for i th person within census tract j γ 00 = average of the neighborhood level intercepts γ 01 = neighborhood SES characteristic for census tract j γ 02 = area level covariates for subject i within census tract j (i.e. population density, high density land use) β 1 = coefficient for individual SES characteristic β 2 = coefficients for individual level covariates (i.e. age, race/ethnicity and sex) a ij = within census tract error term (random effect) ε i = between census tract error term Supplemental Material, Table S1: Differences from mean PM 2.5 and percent difference from geometric mean of NO x associated with an increase in individual and neighborhood socioeconomic status (SES) characteristics estimated from crude and minimally adjusted intrinsic conditional autoregressive (ICAR) models a   metropolitan area. f Model 4: adjusted for variables in model 3 plus population density and high density land use. g Parameter estimates for family income, wealth index and education refer to a one unit increase in the z-score for these variables, which were originally ordinal variables that were transformed into z-scores (see methods section for more details). h Management occupation is a dichotomous variable (management versus non-management occupations). i Parameter estimates for neighborhood SES variables refer to a one-standard deviation unit increase in that variable.   estimates for family income and wealth index refer to a one unit increase in the z-score for these variables, which were originally ordinal variables that were transformed into z-scores (see methods section for more details).  family income, wealth index and education refer to a one unit increase in the z-score for these variables, which were originally ordinal variables that were transformed into z-scores (see methods section for more details). h Management occupation is a dichotomous variable (management versus non-management occupations). i Parameter estimates for neighborhood SES variables refer to a one-standard deviation unit increase in that variable.