Mapping Air Pollution
to Economic Disparity

Fine particulate matter (PM2.5) is a major air pollutant repeatedly linked to morbidity, environmental degradation, and chronic disease. Its exposure distribution is not uniform between populations, and PM2.5 levels vary substantially across regions: existing evidence associates PM2.5 exposure with socioeconomic status. Despite this, there is a notable lack of a uniform county-level evaluation integrating air quality and economic status across the United States.

This study examines the relationship between PM2.5 concentrations in air and median household income in the United States. Merging data from the EPA Air Quality System and the U.S. Census, data for 615 counties were analyzed through regression, correlation, and various distributional comparisons. After analysis, significant national-level disparities have been observed between high-income and low-income communities in terms of PM2.5 concentration and exposure.

All PM2.5 data was obtained from the U.S. Environmental Protection Agency's Air Quality System (AQS), which provides daily county-level averages of airborne fine particulate matter (aerodynamic diameter ≤2.5 μm). Household income data was obtained from the U.S. Census Bureau's American Community Survey (ACS) 5-year estimate of median household income.

All data were analyzed and all relevant columns were standardized before data analysis began. Dates were all parsed into datetime objects, and all PM2.5 data and economic fields were standardized using numeric conversions. County Federal Information Processing Standard (FIPS) codes are constructed for the data by matching state-level and county-level codes.

Daily PM2.5 measurements were aggregated to county-level summaries. Each county's dataset includes:

The AQS and Census Bureau datasets were merged on the standardized FIPS code obtained in pre-processing. Only counties with valid PM2.5 and income data were retained. A merged table of 615 counties was produced.

The following analyses were executed automatically:

After producing a local clone of the repository on your machine and acquiring and using all necessary API keys, the complete analysis can be replicated by running:

Please remember to leave some time for the data to download and for the analysis to run.

Across the 615 counties in the analytical dataset, OLS regression identified a statistically significant negative relationship between annual mean PM2.5 and median household income (β ≈ $1,350 per μg/m³; p < 0.001). This finding was corroborated by Huber robust regression and median quantile regression, both of yielding coefficients of comparable magnitude and direction, indicating the result is not driven by outliers.

The lowest income decile, decile 0, exhibited the highest mean PM2.5 at 7.823 μg/m³ (SD = 2.373). The overall minimum was observed in decile 8 at 6.629 μg/m³ (SD = 1.619), resulting in a difference of 1.194 μg/m³. The highest income decile, decile 9, recorded a mean of 6.988 μg/m³ (SD = 1.421), yielding a gap of 0.835 μg/m³ between the income extremes.

The general downward trend is not perfectly monotonic. Decile 4 shows a local elevation (7.633 μg/m³), suggesting that middle-income counties may face heightened pollution conditions, potentially reflecting a mix of suburban and industrial geographies. The highest decile, Decile 9's mean, 6.988 μg/m³ (SD = 1.421), shows a local elevation as well, hinting towards the air pollution suffered by high-income residents of urban environments. The Spearman rank trend test, however, confirmed an overall statistically significant negative trend across deciles.

The lowest income decile also exhibits the highest variability (SD = 2.373), indicating a notably great heterogeneity in pollution exposure among the lowest-income counties. In contrast, the wealthiest decile (SD = 1.421) shows the most consistent exposure (consistently the lowest).

ANOVA and Kruskal–Wallis tests both returned significant results (p < 0.001), confirming that PM2.5 distributions differ across income groups.

Temporal analysis showed that the negative income–PM2.5 correlation persisted across all months and seasons, though the magnitude varied. Summer months (JJA) showed the strongest associations, likely reflecting elevated photochemical activity and wildfire contributions that disproportionately affect lower-income, rural counties.

Welch's t-test confirmed a statistically significant difference in mean PM2.5 between the lowest and highest income quintiles. Bootstrap confidence intervals for the mean difference excluded zero, which reinforces the strength of the observed effects arising from wealth disparity.

The consistently-negative correlation between PM2.5 and income across multiple statistical methods and analyses underscores the economic realities of air pollution exposure in the United States. These findings align with broader literature about environmental justice and ethical environmentalism, which tends to document the reality of lower-income and minority communities bearing a disproportionate pollution burden.

Several mechanisms could help to explain this pattern. Lower-income counties may be located closer to industrial facilities, transportation corridors such as highways, or agricultural sources of particulate matter. Wealthier communities may also benefit from being able to afford more effective environmental regulation, superior urban planning, and greater political capacity.

The non-monotonic behavior of decile 4, which stands as the highest value outside of the bottom two deciles, is notable among the data. This may stand to reflect a concentration of middle-income counties in regions possessing mixed industrial and suburban land use, where PM2.5 sources could include agriculture, industrial sites, or city pollution.

The elevated standard deviation in the lower deciles, and specifically in decile 0, with a SD of 2.373, shows that the lowest-income groups are the most internally diverse in pollution exposure. They may potentially be rural with clean and untouched air, while other low-income groups may sit near major pollution sources or in city centers. This diversity of results suggests that further sub-county investigation may be warranted.

The fully reproducible and automated nature of AirHealthLink's pipeline enables other researchers to replicate our findings, extend this analysis to additional years, economic indicators, or pollutants, and facilitate potential longitudinal or comparative studies.

Several limitations exist in our analysis. First of all, the analysis is purely observational and cross-sectional: causal inference cannot be drawn from association alone. Confounding variables such as population density, urbanization, industrial composition, racial demographics, or immigration status were not controlled for in this analysis.

Second of all, although the 615 counties with complete data represent a large subset of the more than 3,100 counties in the United States, they by no means represent every county available. FIPS mismatches and ACS data unavailability, such as some counties not possessing daily records, made many counties' data unusable in the study.

Third, using county-level aggregation masks any intra-county disparities in both wealth and pollution exposure. If intra-county disparities were acknowledged through the use of a finer-grain analysis at the census tract or ZIP code level, potentially greater results may have emerged.

Finally, the study uses a single year's data (2022), which may not capture longer-term trends or the effects of episodal or occasional events such as wildfires. A multi-year analysis would considerably strengthen the generalizability of our findings.