Chapter 5.3 · The air you're assigned

Air pollution

The national traffic-pollution percentile of a census tract has a correlation of −0.41 with the Area Deprivation Index across tracts but +0.19 within counties, and the NO₂ percentile rises with the Hispanic share of residents from 33 in the lowest decile to 73 in the highest.

Author: Vishal Singh
Affiliation: NYU Stern
Published: 9 October 2026

Traffic pollution ranks at a mean national percentile of 75 in the tenth of census tracts that are least deprived by the Area Deprivation Index (ADI) and 39 in the tenth that are most deprived. Across the 82,198 tracts that have both values, the correlation between the traffic percentile and the ADI rank is −0.41 (95% CI −0.49 to −0.33). The sign depends on the comparison. Among metropolitan tracts alone it is still negative (−0.31 (95% CI −0.40 to −0.21)), within a county it is positive (+0.19 (95% CI 0.15 to 0.22), in 79,730 tracts of counties that have enough tracts to compare), and with the tract poverty rate in place of the ADI it is +0.07 (95% CI 0.04 to 0.11) across all tracts. For NO₂, diesel PM and traffic, the percentiles rise with the Hispanic share and with the Black share of a tract, across tracts and within counties: the correlation of the NO₂ percentile with Hispanic share is +0.40 (95% CI 0.35 to 0.46) across tracts and +0.23 (95% CI 0.21 to 0.26) within counties.

Every statement in this chapter is about tracts or about the residents of groups of tracts. None describes what an individual breathes, because people leave their tract during the day and the indicators are modeled estimates for particular years.

What the pollution indicators measure

The indicators come from the Environmental Protection Agency's EJScreen. EPA removed the tool from its website in 2025, and the chapter uses the archived copy of the final 2024 release kept on Zenodo.[1] The tract file is on 2020 census tracts, as are the ADI, the American Community Survey (ACS) estimates for 2019 to 2023 and the CDC PLACES file used later, so the files join on the tract identifier with no crosswalk. Connecticut tracts carry the planning-region codes in all of them.

EPA's technical documentation dates each air indicator.[1] Fine particulate matter (PM₂.₅, in micrograms per cubic meter) and ozone (the annual mean of the ten highest daily maximum eight-hour readings, in parts per billion) are for 2020 and combine monitor readings with an air-quality model. NO₂ (parts per billion) is a 2020 estimate built from satellite data and a regression model. Diesel particulate matter (micrograms per cubic meter) is EPA's 2020 estimate from its air toxics screening. Toxic releases to air are a toxicity-weighted concentration modeled from 2021 reports, and traffic proximity is an index of 2020 traffic counts on nearby major roads, weighted by distance. The release has no air-toxics cancer-risk index. Besides the 6 air indicators, the file carries 4 indicators of proximity and discharge (Superfund sites, risk-management facilities, hazardous-waste facilities and wastewater) and 3 of old housing and infrastructure (housing built before 1960 as a proxy for lead paint, underground storage tanks and drinking-water violations), 13 in all. Each is a modeled or calculated value and none is a monitor reading at the tract.

Each indicator also comes as a national percentile. EPA's documentation describes percentiles ranked across block groups, but the percentiles in this file rank tracts, so a percentile is the share of tracts with a lower value. A percentile compares a tract with other tracts and says nothing about the level of a pollutant. The first and third quartiles of tract PM₂.₅ are 7.3 and 9.0 micrograms per cubic meter, so small differences in concentration separate tracts by many percentile points. EJScreen has no PM₂.₅ or ozone estimate for the 1,524 tracts with residents in Alaska, Hawaii and Puerto Rico. A missing value is not a value of zero, so the analysis leaves those tracts out and every comparison covers the same places. It uses the 83,008 tracts of the contiguous states and the District of Columbia that have a positive ACS population, and they hold 330.1 million residents. 810 of those tracts have no ADI rank, and comparisons with the ADI leave them out. Means and correlations are population-weighted, with intervals from a bootstrap that redraws whole counties.

Pollution percentiles fall as deprivation rises

Figure 1 sorts tracts into tenths by the ADI national rank, with the first tenth the least deprived, and averages the pollution percentile of the tracts in each tenth. Every line falls from the first tenth to the ninth and edges up in the tenth. For traffic the means are 75 in the first tenth, 35 (95% CI 32 to 38) in the ninth and 39 in the tenth, and for PM₂.₅ they are 64 (95% CI 52 to 73), 44 (95% CI 41 to 47) and 48 (95% CI 44 to 51). Across the tracts with both values the correlation with the ADI rank is −0.41 (95% CI −0.49 to −0.33) for traffic, −0.38 (95% CI −0.45 to −0.30) for diesel PM, −0.18 (95% CI −0.27 to −0.10) for NO₂ and −0.18 (95% CI −0.29 to −0.06) for PM₂.₅. Higher deprivation goes with lower pollution percentiles across the country, most strongly for traffic and diesel PM.

Figure 1

Traffic pollution ranks at a mean national percentile of 75 in the least deprived tenth of tracts and 39 in the most deprived

Mean national percentile of four EJScreen air indicators by decile of the Area Deprivation Index national rank, population-weighted, census tracts

Mean national percentile of the tract (0 to 100)

Decile of the Area Deprivation Index national rank (1 = least deprived, 10 = most deprived)

Show data

Data for Traffic pollution ranks at a mean national percentile of 75 in the least deprived tenth of tracts and 39 in the most deprived
conf_high	conf_low	decile	estimate	hazard	n_units
79.18	68.05	1	74.70	Traffic proximity	8,149
71.29	58.96	2	65.23	Traffic proximity	8,293
60.23	53.49	3	56.92	Traffic proximity	8,205
53.77	48.87	4	51.45	Traffic proximity	8,184
50.34	44.93	5	47.72	Traffic proximity	8,416
47.13	41.66	6	44.43	Traffic proximity	8,052
44.32	38.17	7	41.19	Traffic proximity	8,161
40.55	35.39	8	38.01	Traffic proximity	8,288
37.53	32.20	9	34.92	Traffic proximity	8,247
42.13	35.42	10	38.99	Traffic proximity	8,203
75.36	62.13	1	69.69	Diesel PM	8,149
71.11	59.70	2	65.61	Diesel PM	8,293
61.41	54.39	3	58.14	Diesel PM	8,205
56.45	50.53	4	53.64	Diesel PM	8,184
52.62	46.70	5	49.71	Diesel PM	8,416
49.53	43.50	6	46.55	Diesel PM	8,052
46.27	39.94	7	43.07	Diesel PM	8,161
42.12	36.95	8	39.46	Diesel PM	8,288
38.91	33.80	9	36.35	Diesel PM	8,247
42	35.94	10	39.13	Diesel PM	8,203
67.58	54.21	1	61.37	NO₂	8,146
63.92	51.85	2	58.08	NO₂	8,292
56.05	48.39	3	52.30	NO₂	8,204
53.46	47.20	4	50.45	NO₂	8,184
51.82	45.39	5	48.71	NO₂	8,416
50.96	44.17	6	47.59	NO₂	8,052
49.47	42.09	7	45.71	NO₂	8,161
46.95	40.92	8	43.89	NO₂	8,288
45.25	39.68	9	42.52	NO₂	8,247
50.71	44.27	10	47.64	NO₂	8,203
73.24	52.34	1	64.27	PM₂.₅	8,149
65.84	47.61	2	56.82	PM₂.₅	8,293
56.99	47.69	3	52.29	PM₂.₅	8,205
54.11	47.20	4	50.75	PM₂.₅	8,184
51.99	46.09	5	49.21	PM₂.₅	8,416
51.28	44.87	6	48.16	PM₂.₅	8,052
50.70	44.20	7	47.40	PM₂.₅	8,161
47.38	41.64	8	44.47	PM₂.₅	8,288
46.62	41.50	9	44.06	PM₂.₅	8,247
50.97	44.29	10	47.76	PM₂.₅	8,203

Every line falls from the least deprived decile on the left to the ninth decile and edges up in the tenth, and the traffic and diesel lines fall furthest. A line plots the population-weighted mean national percentile of the tracts in each decile, tracts tied on the ADI rank share a decile, and the bands are 95% intervals from a bootstrap over counties. Across the tracts with both values the correlation with the ADI rank is −0.41 for traffic, −0.38 for diesel PM, −0.18 for NO₂ and −0.18 for PM₂.₅.Note: National percentiles rank modeled values, so a gap in percentiles is not a gap in concentrations: the mean PM₂.₅ concentration is 9.6 micrograms per cubic meter in the least deprived decile and 8.1 in the most deprived.Note: Each point rests on the tracts of a decile that have a value for the indicator, and the tooltip gives the count. The 1,524 tracts with residents in Alaska, Hawaii and Puerto Rico have no PM₂.₅ or ozone estimate and are outside the analysis, and 810 tracts in it have no ADI rank.Note: The Area Deprivation Index descends from an index that includes median home value and median gross rent, so it rates dense, high-cost metropolitan tracts as less deprived than their incomes suggest.Source: EPA EJScreen 2024, Area Deprivation Index 2023 and ACS 2019-2023; population-weighted means, modeled estimates

A gap in percentiles is a gap in rank. The mean PM₂.₅ concentration is 9.6 (95% CI 8.7 to 10.4) micrograms per cubic meter in the least deprived tenth and 8.1 (95% CI 7.9 to 8.2) in the most deprived. The intervals describe which places are in the file and leave out the error in EPA's models.

The fall persists among metropolitan tracts for traffic and diesel PM, and the sign reverses within counties

One explanation is that the most deprived tracts are rural and that rural air ranks cleaner. Figure 2 tests it in three tabs. The share of residents living in small-town and rural tracts, as the USDA classifies tracts, rises from 0.6% (95% CI 0.4% to 1.1%) in the least deprived tenth to 20.3% (95% CI 18.4% to 22.3%) in the ninth and eases to 18.5% (95% CI 16.2% to 21.0%) in the tenth, and it is 6.9% of residents overall. Those tracts hold a minority of residents in every tenth. Among metropolitan tracts alone the traffic percentile still falls from 76 (95% CI 69 to 80) in the first tenth to 48 (95% CI 45 to 50) in the ninth and diesel PM from 71 (95% CI 64 to 76) to 47 (95% CI 44 to 50). The correlations with the ADI rank are −0.31 (95% CI −0.40 to −0.21) and −0.28 (95% CI −0.36 to −0.19) among metropolitan tracts, against −0.41 and −0.38 in all tracts.

For NO₂ and PM₂.₅ the metropolitan correlations are −0.09 (95% CI −0.17 to 0.00) and −0.10 (95% CI −0.22 to 0.03), small and with intervals that include zero, and most of the fall in the metropolitan lines comes across the first three tenths. Leaving out the small-town, rural and micropolitan tracts changes the correlation less for traffic and diesel PM than for NO₂ and PM₂.₅, so rural tracts account for more of the national fall in the NO₂ and PM₂.₅ percentiles than of the traffic and diesel PM falls.

Figure 2

The fall in traffic and diesel PM percentiles with deprivation persists among metropolitan tracts, and the sign reverses when tracts are compared within a county

Three views of the link between the Area Deprivation Index and national pollution percentiles: rural share, metropolitan tracts only, and correlations by comparison

The share of residents in small-town and rural tracts rises from 0.6% in the least deprived decile to 20.3% in the ninth and 18.5% in the tenth

Share of residents living in small-town and rural tracts by decile of the Area Deprivation Index national rank, population-weighted

Share of residents in small-town and rural tracts (%)

Decile of the Area Deprivation Index national rank (1 = least deprived, 10 = most deprived)

Bars rise through the ninth decile and ease in the tenth. Each bar is the share of residents in tracts that the USDA codes as small town or rural, with 95% intervals from a bootstrap over counties; Connecticut has no code and is left out.

Show data

Data for The fall in traffic and diesel PM percentiles with deprivation persists among metropolitan tracts, and the sign reverses when tracts are compared within a county, Rural share
conf_high	conf_low	decile	n_units	rural
1.10	0.400	1	8,077	0.600
1.40	0.700	2	8,222	1
1.70	1	3	8,064	1.30
2.80	1.80	4	8,050	2.20
4.20	2.90	5	8,271	3.50
6.90	5	6	7,946	5.90
10.30	7.80	7	8,077	8.90
15.50	12.50	8	8,207	13.90
22.30	18.40	9	8,214	20.30
21	16.20	10	8,194	18.50

The first tab shows small-town and rural tracts holding a larger share of residents as deprivation rises, up to 20.3% in the ninth decile. The second compares metropolitan tracts with all tracts for four indicators: the traffic and diesel PM percentiles keep falling, and the NO₂ and PM₂.₅ percentiles fall little. The third compares correlations across all tracts, across metropolitan tracts and within counties.Source: EPA EJScreen 2024, Area Deprivation Index 2023 and ACS 2019-2023; population-weighted means, modeled estimates

The sign also depends on how deprivation is measured. The ADI descends from an index of census measures that includes median home value and median gross rent,[2] so places with high housing costs read as less deprived, a limit that Chapter 10.2, The anatomy of a deprivation index takes up. These data do not isolate that component, but the tract poverty rate does not show the negative sign. Across all tracts it correlates +0.07 (95% CI 0.04 to 0.11) with the traffic percentile and +0.18 (95% CI 0.15 to 0.21) with the NO₂ percentile.

It depends as well on which tracts are compared. Across all tracts, the comparison mixes places: a deprived tract in one county is compared with a less deprived tract in another county, which can differ in traffic, industry and housing costs. Within a county, each tract is compared only with the other tracts of its county, by removing the county's population-weighted mean from both variables. In that comparison the ADI rank correlates +0.19 (95% CI 0.15 to 0.22) with the traffic percentile, +0.19 (95% CI 0.15 to 0.23) with diesel PM and +0.28 (95% CI 0.25 to 0.31) with NO₂, and the poverty rate correlates +0.27 (95% CI 0.24 to 0.29) with the traffic percentile. Within a county, more deprived tracts have higher pollution percentiles. Across counties the relation runs the other way: the correlation between a county's mean traffic percentile and its mean ADI rank is −0.66 (95% CI −0.73 to −0.58) across 3,095 counties, each weighted by its residents. The fall across the country reflects differences between counties.

Early environmental-justice research documented waste facilities sited in Black communities in the South.[3] The same indicators that fall with the ADI rise with the racial and ethnic composition of a tract. Figure 3 repeats the tenths design with the share of residents who are Hispanic or Latino, of any race, and the share who are Black or African American alone, of any ethnicity. The groups overlap, because a Hispanic resident can be Black, and they are 19.0% and 12.1% of residents. The NO₂ percentile averages 33 (95% CI 30 to 36) in the tenth of tracts with the lowest Hispanic share and 73 (95% CI 68 to 76) in the tenth with the highest, and 38 (95% CI 35 to 41) and 58 (95% CI 53 to 63) across Black-share tenths. Tracts with the same share go to the same tenth, so the lowest Black-share tenth holds all 8,815 tracts with no Black alone residents and the second holds correspondingly fewer.

Figure 3

Pollution percentiles rise with the Hispanic share of a tract, from 33 to 73 for NO₂ across deciles, and with its Black share up to the fifth decile and in the tenth

Mean national percentile of four EJScreen air indicators by decile of the tract's Hispanic or Latino share and of its Black alone share, population-weighted

Mean national percentile of the tract (0 to 100)

Decile of the tract's share of residents (1 = lowest share, 10 = highest)

Show data

Data for Pollution percentiles rise with the Hispanic share of a tract, from 33 to 73 for NO₂ across deciles, and with its Black share up to the fifth decile and in the tenth
conf_high	conf_low	decile	estimate	group	hazard	n_units
33.11	27.53	1	30.28	Hispanic or Latino share	Traffic proximity	8,295
37.41	32.83	2	35.08	Hispanic or Latino share	Traffic proximity	8,301
42.07	37.43	3	39.80	Hispanic or Latino share	Traffic proximity	8,301
46.83	42.22	4	44.55	Hispanic or Latino share	Traffic proximity	8,301
51.21	46.17	5	48.74	Hispanic or Latino share	Traffic proximity	8,300
54.44	48.92	6	51.80	Hispanic or Latino share	Traffic proximity	8,300
58.25	51.88	7	55.26	Hispanic or Latino share	Traffic proximity	8,301
61.57	54.47	8	58.22	Hispanic or Latino share	Traffic proximity	8,301
65.67	58.11	9	62.12	Hispanic or Latino share	Traffic proximity	8,298
73.75	61.15	10	67.72	Hispanic or Latino share	Traffic proximity	8,300
38.13	30.82	1	34.38	Black alone share	Traffic proximity	8,811
41.66	33.58	2	37.52	Black alone share	Traffic proximity	7,786
49.06	40.54	3	44.80	Black alone share	Traffic proximity	8,301
54.03	46.09	4	50.15	Black alone share	Traffic proximity	8,301
57.57	50.14	5	54.04	Black alone share	Traffic proximity	8,301
59.11	51.61	6	55.41	Black alone share	Traffic proximity	8,300
57.72	51.71	7	54.72	Black alone share	Traffic proximity	8,301
57.64	51.82	8	54.74	Black alone share	Traffic proximity	8,299
57.05	51.07	9	54.11	Black alone share	Traffic proximity	8,300
63.65	56.11	10	60.09	Black alone share	Traffic proximity	8,298
34.27	29.53	1	31.90	Hispanic or Latino share	Diesel PM	8,295
37.78	33.56	2	35.65	Hispanic or Latino share	Diesel PM	8,301
41.92	37.69	3	39.83	Hispanic or Latino share	Diesel PM	8,301
46.83	42.35	4	44.62	Hispanic or Latino share	Diesel PM	8,301
51.98	46.55	5	49.36	Hispanic or Latino share	Diesel PM	8,300
55.05	49.39	6	52.40	Hispanic or Latino share	Diesel PM	8,300
58.95	52.77	7	56.06	Hispanic or Latino share	Diesel PM	8,301
62.36	55.56	8	59.19	Hispanic or Latino share	Diesel PM	8,301
66.91	59.68	9	63.61	Hispanic or Latino share	Diesel PM	8,298
74.51	61.29	10	68.86	Hispanic or Latino share	Diesel PM	8,300
38.10	30.66	1	34.37	Black alone share	Diesel PM	8,811
41.66	34.07	2	37.81	Black alone share	Diesel PM	7,786
48.49	40.80	3	44.63	Black alone share	Diesel PM	8,301
53.41	45.86	4	49.76	Black alone share	Diesel PM	8,301
57.92	50.84	5	54.49	Black alone share	Diesel PM	8,301
60.10	52.85	6	56.63	Black alone share	Diesel PM	8,300
59.47	53.58	7	56.62	Black alone share	Diesel PM	8,301
59.25	53.41	8	56.45	Black alone share	Diesel PM	8,299
58.75	52.52	9	55.78	Black alone share	Diesel PM	8,300
65.14	56.84	10	61.15	Black alone share	Diesel PM	8,298
35.63	29.91	1	32.64	Hispanic or Latino share	NO₂	8,300
37.85	33.41	2	35.66	Hispanic or Latino share	NO₂	8,300
40.50	36.41	3	38.47	Hispanic or Latino share	NO₂	8,299
44.55	40.42	4	42.50	Hispanic or Latino share	NO₂	8,300
48.59	43.65	5	46.07	Hispanic or Latino share	NO₂	8,301
52	46.50	6	49.30	Hispanic or Latino share	NO₂	8,300
56.32	50.50	7	53.53	Hispanic or Latino share	NO₂	8,301
61.67	55.20	8	58.61	Hispanic or Latino share	NO₂	8,301
66.71	60.36	9	63.74	Hispanic or Latino share	NO₂	8,301
76.39	67.95	10	72.62	Hispanic or Latino share	NO₂	8,300
40.85	34.68	1	37.81	Black alone share	NO₂	8,813
42.45	36.05	2	39.17	Black alone share	NO₂	7,786
48.13	41.45	3	44.78	Black alone share	NO₂	8,300
52.13	45.61	4	48.97	Black alone share	NO₂	8,299
56.35	50.06	5	53.35	Black alone share	NO₂	8,301
57.99	51.28	6	54.73	Black alone share	NO₂	8,300
57.40	51.41	7	54.50	Black alone share	NO₂	8,301
57.79	51.05	8	54.50	Black alone share	NO₂	8,301
57.51	49.72	9	53.74	Black alone share	NO₂	8,301
62.87	53.48	10	58.35	Black alone share	NO₂	8,301
42.72	37.18	1	39.90	Hispanic or Latino share	PM₂.₅	8,300
41.66	37.05	2	39.37	Hispanic or Latino share	PM₂.₅	8,301
43.40	39.08	3	41.23	Hispanic or Latino share	PM₂.₅	8,301
46.77	41.99	4	44.43	Hispanic or Latino share	PM₂.₅	8,301
49.67	44.33	5	47.04	Hispanic or Latino share	PM₂.₅	8,301
52.85	46.43	6	49.59	Hispanic or Latino share	PM₂.₅	8,300
56.43	48.56	7	52.56	Hispanic or Latino share	PM₂.₅	8,301
60.75	51.13	8	56.22	Hispanic or Latino share	PM₂.₅	8,301
67.09	56.02	9	62.01	Hispanic or Latino share	PM₂.₅	8,301
76.80	60.46	10	69.44	Hispanic or Latino share	PM₂.₅	8,301
49.02	40.71	1	44.78	Black alone share	PM₂.₅	8,815
48.02	40.19	2	44	Black alone share	PM₂.₅	7,786
52.73	43.89	3	48.27	Black alone share	PM₂.₅	8,301
55.36	46.28	4	50.74	Black alone share	PM₂.₅	8,301
57.32	48.36	5	52.76	Black alone share	PM₂.₅	8,301
58.38	49.06	6	53.58	Black alone share	PM₂.₅	8,300
57.34	49.22	7	53.36	Black alone share	PM₂.₅	8,301
57.32	49.17	8	53.42	Black alone share	PM₂.₅	8,301
55.94	48.20	9	52.14	Black alone share	PM₂.₅	8,301
58.29	48.65	10	53.65	Black alone share	PM₂.₅	8,301

Hispanic-share lines rise from the second decile to the tenth in every panel. Black-share lines rise from the second decile to the fifth, change little from the fifth to the ninth, and rise again in the tenth for traffic, diesel PM and NO₂. At the tenth decile the Hispanic-share line is above the Black-share line for NO₂ and PM₂.₅, and the two intervals overlap for traffic and diesel PM. A point is the population-weighted mean percentile of the residents of the tracts in a decile, so it describes tracts and not the exposure of any person. Across the tracts the correlation of the NO₂ percentile is +0.40 with Hispanic share and +0.14 with Black share.Note: Deciles hold nearly equal numbers of tracts, and tracts with the same share go to the same decile. The lowest Black-share decile therefore holds all 8,815 tracts with no Black alone residents, and the second holds fewer than a tenth of the tracts.Note: Hispanic or Latino residents can be of any race and Black alone residents can be of any ethnicity, so the two groups overlap. Both shares are ACS 2019-2023 tract estimates.Note: Weighting each tract by its number of Hispanic or Latino residents instead, the mean NO₂ percentile is 63; weighting by Black alone residents it is 56, and weighting by all residents it is 50. These are averages over tracts of residence and not measurements of individuals.Note: The analysis covers the contiguous states and the District of Columbia, because EJScreen has no PM₂.₅ or ozone estimate for the 1,524 tracts in Alaska, Hawaii and Puerto Rico.Source: EPA EJScreen 2024 and ACS 2019-2023; population-weighted means, modeled estimates

In every panel the Hispanic-share lines rise from the second tenth to the tenth, and the Black-share lines rise through the fifth tenth, change little to the ninth and rise again in the tenth for traffic, diesel PM and NO₂. Across tracts, Hispanic share correlates +0.40 (95% CI 0.35 to 0.46) with the NO₂ percentile, +0.34 (95% CI 0.26 to 0.41) with diesel PM, +0.33 (95% CI 0.27 to 0.40) with traffic and +0.32 (95% CI 0.22 to 0.40) with PM₂.₅. Black share correlates +0.14 (95% CI 0.10 to 0.19) with NO₂, +0.18 (95% CI 0.13 to 0.23) with diesel PM and +0.17 (95% CI 0.12 to 0.22) with traffic, and +0.06 (95% CI 0.00 to 0.12) with PM₂.₅, which is close to zero. For NO₂, diesel PM and traffic the correlations with both shares stay positive when states are redrawn instead of counties, and they persist within counties. There the NO₂ percentile correlates +0.23 (95% CI 0.21 to 0.26) with Hispanic share and +0.17 (95% CI 0.14 to 0.21) with Black share, and the traffic percentile +0.19 (95% CI 0.16 to 0.21) and +0.16 (95% CI 0.14 to 0.19).

To see whether race and deprivation each go with pollution on their own, a joint model puts them together. In a weighted regression of the traffic percentile on the ADI rank, Hispanic share and Black share, each variable measured from its county mean and scaled to one standard deviation, the coefficients are +0.06 (95% CI 0.02 to 0.10) for the ADI rank, +0.17 (95% CI 0.14 to 0.21) for Hispanic share and +0.15 (95% CI 0.12 to 0.19) for Black share. For NO₂ they are +0.16 (95% CI 0.13 to 0.19), +0.18 (95% CI 0.15 to 0.21) and +0.12 (95% CI 0.08 to 0.16). All are positive and the ADI rank is the weakest for traffic. With the poverty rate in place of the ADI rank, the traffic coefficients are +0.19 (95% CI 0.17 to 0.21) for poverty, +0.14 (95% CI 0.10 to 0.17) for Hispanic share and +0.10 (95% CI 0.07 to 0.14) for Black share. Within a county, poverty and the racial and ethnic composition of a tract each go with higher pollution percentiles, and these data do not rank them.

Weighting tracts by the residents of a group describes where the group lives. The tract of the average Hispanic resident has an NO₂ percentile of 63 (95% CI 59 to 67), that of the average Black alone resident 56 (95% CI 52 to 60), and that of the average resident 50 (95% CI 47 to 53). These are averages over tracts of residence and not measurements of individuals. A national model that attributes PM₂.₅ exposure to the consumption of goods and services finds that Black and Hispanic residents are exposed to more particulate matter than their own consumption accounts for, and non-Hispanic white residents to less.[4]

Hazards stack in the same tracts

Single indicators understate a tract's burden when hazards coincide. Count, for each tract, how many of the 5 hazards PM₂.₅, diesel PM, NO₂, traffic proximity and toxic releases to air have a national percentile of at least 75. Figure 4 shows the mean count by tenth for 82,993 tracts. It rises from 0.8 (95% CI 0.6 to 1.0) in the lowest Black-share tenth to 1.7 (95% CI 1.4 to 2.0) in the highest, and from 0.6 (95% CI 0.5 to 0.7) to 2.3 (95% CI 1.8 to 2.8) across Hispanic-share tenths. Across ADI tenths it falls from 2.1 (95% CI 1.7 to 2.5) to 1.0 (95% CI 0.8 to 1.2). The Black-share line is flat from the sixth tenth to the ninth, so the highest tenth stands apart. Ozone is not counted, because it forms in the air from other pollutants and does not mark a local source as the five counted hazards do; adding it as a sixth hazard gives the same ordering of the three lines.

Figure 4

Residents of tracts in the highest Black-share decile live with 1.7 stacked air hazards on average, against 0.8 in the lowest, and the score falls across deprivation deciles

Mean number of five air hazards in the top quarter of tracts, by decile of tract Black alone share, Hispanic or Latino share and ADI national rank, population-weighted, up to 82,993 tracts

Mean number of stacked air hazards (0 to 5)

Show data

Data for Residents of tracts in the highest Black-share decile live with 1.7 stacked air hazards on average, against 0.8 in the lowest, and the score falls across deprivation deciles
conf_high	conf_low	decile	estimate	measure	n_units	sorted_by
0.990	0.640	1	0.800	score	8,809	Black alone share
15.06	6.03	1	10.17	heavy	8,809	Black alone share
1.04	0.650	2	0.840	score	7,786	Black alone share
17.05	6.94	2	11.71	heavy	7,786	Black alone share
1.27	0.830	3	1.04	score	8,300	Black alone share
22.10	10.42	3	16.08	heavy	8,300	Black alone share
1.43	0.980	4	1.20	score	8,299	Black alone share
25.58	13.37	4	19.37	heavy	8,299	Black alone share
1.58	1.12	5	1.35	score	8,301	Black alone share
28.92	16.40	5	22.71	heavy	8,301	Black alone share
1.68	1.17	6	1.43	score	8,300	Black alone share
31.73	18.17	6	25.08	heavy	8,300	Black alone share
1.57	1.16	7	1.36	score	8,301	Black alone share
28.24	17.10	7	22.71	heavy	8,301	Black alone share
1.60	1.16	8	1.38	score	8,299	Black alone share
28.79	17.03	8	22.94	heavy	8,299	Black alone share
1.59	1.16	9	1.37	score	8,300	Black alone share
30.34	17.82	9	24.06	heavy	8,300	Black alone share
2	1.40	10	1.70	score	8,298	Black alone share
40.20	23.19	10	31.97	heavy	8,298	Black alone share
0.750	0.470	1	0.600	score	8,295	Hispanic or Latino share
9.79	3.68	1	6.44	heavy	8,295	Hispanic or Latino share
0.720	0.530	2	0.620	score	8,300	Hispanic or Latino share
8.77	4.74	2	6.68	heavy	8,300	Hispanic or Latino share
0.850	0.650	3	0.750	score	8,299	Hispanic or Latino share
10.63	6	3	8.29	heavy	8,299	Hispanic or Latino share
1.01	0.780	4	0.890	score	8,300	Hispanic or Latino share
14.68	8.62	4	11.60	heavy	8,300	Hispanic or Latino share
1.21	0.930	5	1.07	score	8,300	Hispanic or Latino share
18.85	10.90	5	14.90	heavy	8,300	Hispanic or Latino share
1.34	1	6	1.17	score	8,300	Hispanic or Latino share
22.30	12.90	6	17.60	heavy	8,300	Hispanic or Latino share
1.53	1.14	7	1.34	score	8,301	Hispanic or Latino share
27.59	16.80	7	22.27	heavy	8,301	Hispanic or Latino share
1.78	1.30	8	1.54	score	8,301	Hispanic or Latino share
34.26	20.32	8	27.64	heavy	8,301	Hispanic or Latino share
2.07	1.55	9	1.82	score	8,298	Hispanic or Latino share
42.31	27.17	9	35.04	heavy	8,298	Hispanic or Latino share
2.79	1.84	10	2.34	score	8,299	Hispanic or Latino share
60.69	34.80	10	48.84	heavy	8,299	Hispanic or Latino share
2.49	1.67	1	2.13	score	8,146	Area Deprivation Index rank
56.35	29.54	1	44.46	heavy	8,146	Area Deprivation Index rank
2.26	1.34	2	1.79	score	8,292	Area Deprivation Index rank
47.03	22.05	2	34.58	heavy	8,292	Area Deprivation Index rank
1.63	1.16	3	1.38	score	8,204	Area Deprivation Index rank
30.43	17.64	3	24.07	heavy	8,204	Area Deprivation Index rank
1.39	1.03	4	1.21	score	8,184	Area Deprivation Index rank
23.21	13.37	4	18.35	heavy	8,184	Area Deprivation Index rank
1.30	0.920	5	1.11	score	8,416	Area Deprivation Index rank
21.04	10.84	5	15.92	heavy	8,416	Area Deprivation Index rank
1.24	0.850	6	1.04	score	8,052	Area Deprivation Index rank
20.28	9.73	6	14.88	heavy	8,052	Area Deprivation Index rank
1.22	0.780	7	0.980	score	8,161	Area Deprivation Index rank
19.91	8.44	7	13.88	heavy	8,161	Area Deprivation Index rank
1.04	0.710	8	0.870	score	8,288	Area Deprivation Index rank
16.11	7.58	8	11.47	heavy	8,288	Area Deprivation Index rank
0.970	0.660	9	0.800	score	8,247	Area Deprivation Index rank
14.70	6.77	9	10.40	heavy	8,247	Area Deprivation Index rank
1.17	0.790	10	0.980	score	8,203	Area Deprivation Index rank
19.56	9.71	10	14.49	heavy	8,203	Area Deprivation Index rank

The Hispanic-share line climbs most, from 0.6 to 2.3, and the deprivation line runs the other way, from 2.1 in the least deprived decile to 1.0 in the most deprived. A hazard is stacked when the tract's national percentile is at least 75, and the score counts the 5 hazards PM₂.₅, diesel PM, NO₂, traffic proximity and toxic releases to air. The second view shows the share of residents in tracts with 3 or more stacked hazards.Note: Deciles hold nearly equal numbers of tracts, tracts with the same value go to the same decile, and the lowest Black-share decile holds all 8,815 tracts with no Black alone residents. Bands are 95% intervals from a bootstrap over counties. The lines are not monotone in every step: the Black-share line is level from the sixth to the ninth decile and the deprivation line rises slightly in the tenth.Note: Ozone is not counted. It forms in the air from other pollutants and does not mark a local source, as the other five hazards do. A count of six that adds ozone has the same ordering across the three lines.Note: The analysis covers the contiguous states and the District of Columbia, because EJScreen has no PM₂.₅ or ozone estimate for the 1,524 tracts in Alaska, Hawaii and Puerto Rico.Note: Hispanic or Latino residents can be of any race and Black alone residents can be of any ethnicity, so the groups overlap.Source: EPA EJScreen 2024 and ACS 2019-2023; population-weighted means, modeled estimates

The second view shows the share of residents living in tracts with 3 or more stacked hazards. It is 10% (95% CI 6% to 15%) in the lowest Black-share tenth and 32% (95% CI 23% to 40%) in the highest, and 6% (95% CI 4% to 10%) and 49% (95% CI 35% to 61%) across Hispanic-share tenths.

Two components summarize the indicators

The indicators overlap, and principal components summarize them. The analysis standardizes each of the 13 national percentiles to mean zero and standard deviation one across the 80,754 tracts that have all of them, counts every tract once, and finds the weighted combinations of the indicators that capture the most variance. The first component holds 38.6% (95% CI 36.2% to 41.7%) of the variance and the second 11.7% (95% CI 10.5% to 13.1%). A population-weighted version gives almost the same components, with 37.1% in the first.

The first component has positive weights on every indicator except drinking-water violations, and the largest are on diesel PM (+0.40 (95% CI 0.38 to 0.41)), traffic proximity (+0.40 (95% CI 0.39 to 0.41)), hazardous-waste facilities (+0.39 (95% CI 0.37 to 0.40)) and NO₂ (+0.38 (95% CI 0.36 to 0.39)). It marks tracts that are dense and close to roads and facilities, and the chapter calls it the urban-industrial component. The second weights underground storage tanks (+0.55 (95% CI 0.33 to 0.66)), lead paint (+0.45 (95% CI 0.33 to 0.56)) and drinking-water violations (+0.32 (95% CI 0.19 to 0.43)) positively and PM₂.₅ (−0.47 (95% CI −0.55 to −0.38)) and ozone (−0.31 (95% CI −0.38 to −0.16)) negatively, so it separates tracts with old housing and infrastructure from tracts with high PM₂.₅ and ozone (Figure 5). The sign of a component is arbitrary, and the names describe the weights without testing them.

Figure 5

The first component weights diesel PM, traffic, hazardous-waste facilities and NO₂ most, and the second weights storage tanks, lead paint and drinking-water violations against PM₂.₅ and ozone

Weights of the 13 standardized EJScreen percentiles on the first two principal components, 80,754 tracts, each tract counted once

Air
Proximity and discharge
Old housing and infrastructure

Show data

Data for The first component weights diesel PM, traffic, hazardous-waste facilities and NO₂ most, and the second weights storage tanks, lead paint and drinking-water violations against PM₂.₅ and ozone
group	indicator	pc1	pc1_high	pc1_low	pc2	pc2_high	pc2_low
Air	PM2.5	0.220	0.260	0.170	−0.470	−0.380	−0.550
Air	Ozone	0.250	0.280	0.220	−0.310	−0.160	−0.380
Air	Diesel PM	0.400	0.410	0.380	−0.020	0.050	−0.080
Air	Nitrogen dioxide	0.380	0.390	0.360	0.040	0.110	−0.030
Air	Traffic proximity	0.400	0.410	0.390	0.020	0.080	−0.050
Air	Toxic releases to air	0.230	0.260	0.200	0.090	0.280	−0.200
Proximity and discharge	Superfund proximity	0.230	0.260	0.210	0.220	0.300	0.130
Proximity and discharge	Risk-management facilities	0.260	0.280	0.210	−0.080	0.110	−0.330
Proximity and discharge	Hazardous-waste facilities	0.390	0.400	0.370	0	0.070	−0.060
Proximity and discharge	Wastewater discharge	0.220	0.250	0.190	−0.040	0.060	−0.150
Old housing and infrastructure	Lead paint (housing built before 1960)	0.140	0.170	0.100	0.450	0.560	0.330
Old housing and infrastructure	Underground storage tanks	0.150	0.260	0.030	0.550	0.660	0.330
Old housing and infrastructure	Drinking-water violations	−0.080	−0.020	−0.120	0.320	0.430	0.190

Indicators at the right weigh most on the first component, which holds 38.6% of the variance, and indicators near the top weigh most on the second, which holds 11.7%. A dot is an indicator, colored by group, and the data table gives the 95% intervals from a bootstrap over counties. The sign of a component is arbitrary and is set so that traffic is positive on the first and storage tanks on the second.Note: Weights are the entries of unit-length eigenvectors of the correlation matrix, so they are the weights on standardized percentiles. A population-weighted version gives almost the same components: the first holds 37.1% of the variance.Note: The first component is labeled by its largest weights and the second by its largest positive weights; the labels are descriptions, not findings.Source: EPA EJScreen 2024, national percentiles of the 13 indicators; principal components of the correlation matrix, each tract once, modeled estimates

The deprivation index relates differently to the two kinds of indicator. Storage tanks correlate +0.25 (95% CI 0.13 to 0.36) with the ADI rank and lead paint +0.16 (95% CI 0.06 to 0.26), against negative correlations for most of the air indicators.

The urban-industrial component is highest in a few dense counties

Figure 6 averages the tract scores on the first component within each county, weighting tracts by population. Cook County, Illinois, has the highest mean (+1.69, in standard deviations of the tract score), followed by Hudson County, New Jersey (+1.67), New York County (+1.58), the independent city of St. Louis (+1.50) and the Bronx (+1.45). Los Angeles County has rank 6 among the mapped counties. The median county scores −1.2, so most counties sit below the tract average. The map covers 3,094 counties. The 49 gray counties are in Alaska and Hawaii, where EJScreen has no PM₂.₅ or ozone estimate, or are counties where fewer than 50% of residents live in tracts with all indicators. Puerto Rico is not drawn.

Figure 6

The urban-industrial component is highest in Cook County, Illinois, and in dense counties around New York, and below the tract average in most counties

Population-weighted county mean of the tract score on the first principal component of 13 EJScreen percentiles, in standard deviations of the tract score

Loading county values

Mean score on the first component (standard deviations of the tract score)

Lower combined percentilesHigher combined percentiles

Values outside −2.0 to 0.9 take the end colors.

No data (3,143 counties)

Select a class to show only those places.

Find a county

Hover or tap a county for its value, or search by name. Zoom with the buttons on the map, then drag to move it.

Counties toward the high end of the key hold tracts with high combined percentiles for traffic, diesel PM, NO₂ and hazardous-waste proximity, and most counties sit toward the low end. Cook County has a mean score of +1.69 and the median county −1.2. The 49 gray counties have no score, and the tooltip says why.Note: A county is mapped when at least 50% of its residents live in tracts with all 13 indicators. Gray counties have no score: Alaska and Hawaii (no PM₂.₅ or ozone estimate), and a few counties where fewer tracts have all of them. Puerto Rico has no PM₂.₅ or ozone estimate and is not drawn.Note: Connecticut is drawn as its eight legacy counties, with tracts assigned to them by the staged crosswalk. Scores are county means of modeled tract values and carry no sampling interval.Note: The key runs from the first to the 99th percentile of the county scores, and counties outside that range take the colors of its two ends; the tooltip gives the value.Source: EPA EJScreen 2024, first principal component of the 13 national percentiles; population-weighted county means, modeled estimates

Figure 7 lists the correlation of each of the 13 indicators with the ADI rank, Hispanic share and Black share. It shades a cell by the wider of two 95% intervals, one from a bootstrap over counties and one from a bootstrap over states, and the counts below require both to exclude zero. Of the 6 air indicators, 4 correlate negatively with the ADI rank: ozone, diesel PM, NO₂ and traffic. The PM₂.₅ correlation is negative under the county bootstrap and its interval includes zero when states are redrawn, and the correlation for toxic releases to air cannot be told from zero. 5 correlate positively with Hispanic share, all but toxic releases, and 4 with Black share: diesel PM, NO₂, traffic and toxic releases. The PM₂.₅ and ozone cells for Black share include zero. Of the 3 old-housing and infrastructure indicators, 1 is clearly positive with the ADI rank, lead paint. Storage tanks are positive under the county bootstrap and include zero when states are redrawn, and the interval for drinking-water violations includes zero.

Figure 7

Pollution percentiles correlate negatively with deprivation for 4 of 6 air indicators and positively with Hispanic share for 5

Population-weighted correlation of 13 EJScreen national percentiles with the ADI national rank, Hispanic or Latino share and Black alone share, across tracts

−0.500.00+0.50

Faded: the interval includes 0.00

A grid of 13 pollution indicators by three tract characteristics, with the correlation printed in each cell. In the air rows the deprivation column is negative for ozone, diesel PM, NO₂ and traffic and faded for PM₂.₅ and toxic releases. The Hispanic column is positive except for toxic releases, and the Black column is positive for diesel PM, NO₂, traffic and toxic releases and faded for PM₂.₅ and ozone. Lead paint is positive with deprivation, and storage tanks and drinking-water violations are faded.
Indicator	Area Deprivation Index	Hispanic or Latino share	Black alone share
PM2.5	−0.18, interval −0.38 to +0.10, n = 82,198	+0.32, interval +0.07 to +0.42, n = 83,008	+0.06, interval −0.07 to +0.24, n = 83,008
Ozone	−0.24, interval −0.36 to −0.04, n = 82,198	+0.34, interval +0.17 to +0.45, n = 83,008	−0.07, interval −0.17 to +0.06, n = 83,008
Diesel PM	−0.38, interval −0.46 to −0.25, n = 82,198	+0.34, interval +0.17 to +0.46, n = 82,998	+0.18, interval +0.09 to +0.28, n = 82,998
Nitrogen dioxide	−0.18, interval −0.29 to −0.05, n = 82,193	+0.40, interval +0.34 to +0.46, n = 83,003	+0.14, interval +0.06 to +0.24, n = 83,003
Traffic proximity	−0.41, interval −0.52 to −0.25, n = 82,198	+0.33, interval +0.22 to +0.42, n = 82,998	+0.17, interval +0.06 to +0.29, n = 82,998
Toxic releases to air	+0.06, interval −0.06 to +0.15, n = 82,198	−0.04, interval −0.14 to +0.09, n = 83,008	+0.16, interval +0.10 to +0.22, n = 83,008
Superfund proximity	−0.21, interval −0.30 to −0.10, n = 82,198	+0.13, interval +0.05 to +0.21, n = 82,998	+0.08, interval +0.01 to +0.15, n = 82,998
Risk-management facilities	+0.10, interval 0.00 to +0.22, n = 82,198	+0.25, interval +0.10 to +0.32, n = 82,998	+0.14, interval +0.09 to +0.20, n = 82,998
Hazardous-waste facilities	−0.36, interval −0.50 to −0.15, n = 82,198	+0.27, interval +0.13 to +0.41, n = 82,998	+0.13, interval +0.01 to +0.26, n = 82,998
Wastewater discharge	−0.09, interval −0.17 to +0.04, n = 82,198	+0.07, interval −0.07 to +0.21, n = 82,998	+0.06, interval +0.01 to +0.13, n = 82,998
Lead paint (housing built before 1960)	+0.16, interval +0.04 to +0.34, n = 82,198	+0.02, interval −0.09 to +0.13, n = 83,008	+0.06, interval 0.00 to +0.13, n = 83,008
Underground storage tanks	+0.25, interval −0.04 to +0.46, n = 82,198	−0.04, interval −0.29 to +0.32, n = 83,008	+0.27, interval +0.22 to +0.32, n = 83,008
Drinking-water violations	+0.06, interval −0.11 to +0.21, n = 80,047	−0.02, interval −0.14 to +0.17, n = 80,758	−0.02, interval −0.11 to +0.06, n = 80,758

Show data

Data for Pollution percentiles correlate negatively with deprivation for 4 of 6 air indicators and positively with Hispanic share for 5
against	conf_high	conf_low	county_high	county_low	group	indicator	n	r	state_high	state_low
Area Deprivation Index	0.101	−0.378	−0.060	−0.289	Air	PM2.5	82,198	−0.176	0.101	−0.378
Hispanic or Latino share	0.422	0.069	0.396	0.217	Air	PM2.5	83,008	0.315	0.422	0.069
Black alone share	0.240	−0.075	0.118	0	Air	PM2.5	83,008	0.060	0.240	−0.075
Area Deprivation Index	−0.044	−0.361	−0.132	−0.352	Air	Ozone	82,198	−0.244	−0.044	−0.361
Hispanic or Latino share	0.445	0.172	0.418	0.236	Air	Ozone	83,008	0.336	0.445	0.172
Black alone share	0.063	−0.167	−0.010	−0.122	Air	Ozone	83,008	−0.066	0.063	−0.167
Area Deprivation Index	−0.254	−0.455	−0.301	−0.452	Air	Diesel PM	82,198	−0.377	−0.254	−0.455
Hispanic or Latino share	0.457	0.173	0.409	0.258	Air	Diesel PM	82,998	0.340	0.457	0.173
Black alone share	0.277	0.087	0.227	0.131	Air	Diesel PM	82,998	0.179	0.277	0.087
Area Deprivation Index	−0.047	−0.286	−0.103	−0.266	Air	Nitrogen dioxide	82,193	−0.182	−0.047	−0.286
Hispanic or Latino share	0.455	0.340	0.455	0.345	Air	Nitrogen dioxide	83,003	0.402	0.447	0.340
Black alone share	0.245	0.058	0.189	0.098	Air	Nitrogen dioxide	83,003	0.145	0.245	0.058
Area Deprivation Index	−0.247	−0.516	−0.331	−0.492	Air	Traffic proximity	82,198	−0.413	−0.247	−0.516
Hispanic or Latino share	0.419	0.219	0.401	0.267	Air	Traffic proximity	82,998	0.333	0.419	0.219
Black alone share	0.289	0.057	0.220	0.116	Air	Traffic proximity	82,998	0.167	0.289	0.057
Area Deprivation Index	0.147	−0.063	0.130	−0.001	Air	Toxic releases to air	82,198	0.064	0.147	−0.063
Hispanic or Latino share	0.087	−0.135	0.054	−0.135	Air	Toxic releases to air	83,008	−0.037	0.087	−0.109
Black alone share	0.221	0.101	0.202	0.115	Air	Toxic releases to air	83,008	0.162	0.221	0.101
Area Deprivation Index	−0.102	−0.302	−0.141	−0.273	Proximity and discharge	Superfund proximity	82,198	−0.210	−0.102	−0.302
Hispanic or Latino share	0.213	0.048	0.200	0.062	Proximity and discharge	Superfund proximity	82,998	0.134	0.213	0.048
Black alone share	0.152	0.006	0.123	0.027	Proximity and discharge	Superfund proximity	82,998	0.077	0.152	0.006
Area Deprivation Index	0.219	0.001	0.177	0.016	Proximity and discharge	Risk-management facilities	82,198	0.100	0.219	0.001
Hispanic or Latino share	0.324	0.103	0.315	0.174	Proximity and discharge	Risk-management facilities	82,998	0.248	0.324	0.103
Black alone share	0.200	0.090	0.185	0.092	Proximity and discharge	Risk-management facilities	82,998	0.142	0.200	0.090
Area Deprivation Index	−0.154	−0.497	−0.266	−0.450	Proximity and discharge	Hazardous-waste facilities	82,198	−0.363	−0.154	−0.497
Hispanic or Latino share	0.405	0.128	0.358	0.192	Proximity and discharge	Hazardous-waste facilities	82,998	0.274	0.405	0.128
Black alone share	0.259	0.015	0.185	0.079	Proximity and discharge	Hazardous-waste facilities	82,998	0.131	0.259	0.015
Area Deprivation Index	0.035	−0.172	−0.014	−0.170	Proximity and discharge	Wastewater discharge	82,198	−0.092	0.035	−0.172
Hispanic or Latino share	0.207	−0.070	0.164	−0.036	Proximity and discharge	Wastewater discharge	82,998	0.073	0.207	−0.070
Black alone share	0.126	0.007	0.112	0.011	Proximity and discharge	Wastewater discharge	82,998	0.061	0.126	0.007
Area Deprivation Index	0.342	0.040	0.258	0.060	Old housing and infrastructure	Lead paint (housing built before 1960)	82,198	0.160	0.342	0.040
Hispanic or Latino share	0.131	−0.087	0.107	−0.058	Old housing and infrastructure	Lead paint (housing built before 1960)	83,008	0.019	0.131	−0.087
Black alone share	0.126	0.002	0.113	0.007	Old housing and infrastructure	Lead paint (housing built before 1960)	83,008	0.062	0.126	0.002
Area Deprivation Index	0.461	−0.039	0.365	0.126	Old housing and infrastructure	Underground storage tanks	82,198	0.253	0.461	−0.039
Hispanic or Latino share	0.320	−0.294	0.104	−0.177	Old housing and infrastructure	Underground storage tanks	83,008	−0.037	0.320	−0.294
Black alone share	0.317	0.216	0.298	0.243	Old housing and infrastructure	Underground storage tanks	83,008	0.271	0.317	0.216
Area Deprivation Index	0.209	−0.109	0.142	−0.024	Old housing and infrastructure	Drinking-water violations	80,047	0.063	0.209	−0.109
Hispanic or Latino share	0.166	−0.142	0.077	−0.098	Old housing and infrastructure	Drinking-water violations	80,758	−0.017	0.166	−0.142
Black alone share	0.062	−0.107	0.027	−0.065	Old housing and infrastructure	Drinking-water violations	80,758	−0.020	0.062	−0.107

In the air rows the deprivation column is negative for ozone, diesel PM, NO₂ and traffic, and the PM₂.₅ and toxic-release cells include zero. The Hispanic column is positive for PM₂.₅, ozone, diesel PM, NO₂ and traffic, and the Black column is positive for diesel PM, NO₂, traffic and toxic releases. Among the old-housing and infrastructure rows only lead paint is clearly positive with deprivation. Each cell prints the correlation, and cells whose interval includes zero are faded. The interval is the wider of the 95% interval from a bootstrap over counties and the one from a bootstrap over states.Note: Indicators are tract-level national percentiles; several are modeled estimates. Correlations use all tracts that have both values, so the count varies by cell. The data table gives the county and state intervals next to the wider one that the figure uses.Note: Hispanic or Latino residents can be of any race and Black alone residents can be of any ethnicity, so the two columns overlap.Source: EPA EJScreen 2024, Area Deprivation Index 2023 and ACS 2019-2023; population-weighted correlations, modeled estimates

Modeled asthma does not follow the urban-industrial component

If the urban-industrial component tracked illness, modeled asthma would rise with it. CDC PLACES estimates the prevalence of current asthma among adults for every tract from a model fitted to the 2022 Behavioral Risk Factor Surveillance System, with age, sex, race and ethnicity, education and county poverty as inputs.[5] A tract's value therefore reflects who lives there and where, and a correlation of PLACES values with the ADI or with race shares is partly built into the model. The tract file has no intervals, and Chapter 5.1, Health behaviors and the supply of care discusses the same limit for other PLACES measures.

Figure 8 ranks tracts in twentieths on each of three measures. Modeled asthma is highest in the lowest twentieth on the urban-industrial component (11.2% (95% CI 11.1% to 11.3%)) and lowest in the highest (9.9% (95% CI 9.7% to 10.5%)). It rises from 9.7% (95% CI 9.5% to 9.9%) to 11.0% (95% CI 10.6% to 11.5%) across lead-paint twentieths, and from 9.0% (95% CI 8.9% to 9.3%) to 12.7% (95% CI 12.4% to 12.9%) across deprivation twentieths. Across tracts the urban-industrial component correlates −0.10 (95% CI −0.18 to 0.00) with modeled asthma and −0.37 (95% CI −0.42 to −0.32) with modeled COPD, the lead-paint percentile correlates +0.27 (95% CI 0.20 to 0.34) with asthma, and the ADI rank +0.53 (95% CI 0.48 to 0.58). The deprivation line is the one most closely tied to the model's inputs, and the flat-to-falling urban-industrial line says nothing about illness itself, because the values are modeled.

Figure 8

Modeled asthma prevalence is highest in the lowest twentieth of tracts on the urban-industrial component (11.2%) and rises with the lead-paint percentile and with deprivation

Modeled prevalence of current asthma among adults by twentieth of tracts on three measures, population-weighted, 80,754 tracts with all 13 percentiles

Adults with current asthma (%, modeled)

Show data

Data for Modeled asthma prevalence is highest in the lowest twentieth of tracts on the urban-industrial component (11.2%) and rises with the lead-paint percentile and with deprivation
conf_high	conf_low	estimate	n_units	ranked_on	twentieth
11.28	11.08	11.17	4,033	Urban-industrial component	1
10.93	10.79	10.86	4,034	Urban-industrial component	2
10.80	10.66	10.73	4,034	Urban-industrial component	3
10.71	10.56	10.64	4,033	Urban-industrial component	4
10.64	10.47	10.56	4,034	Urban-industrial component	5
10.62	10.45	10.53	4,034	Urban-industrial component	6
10.57	10.38	10.47	4,033	Urban-industrial component	7
10.50	10.29	10.40	4,034	Urban-industrial component	8
10.52	10.29	10.41	4,034	Urban-industrial component	9
10.49	10.27	10.38	4,034	Urban-industrial component	10
10.50	10.24	10.37	4,033	Urban-industrial component	11
10.54	10.24	10.39	4,034	Urban-industrial component	12
10.60	10.24	10.41	4,034	Urban-industrial component	13
10.60	10.26	10.42	4,033	Urban-industrial component	14
10.71	10.32	10.50	4,034	Urban-industrial component	15
10.85	10.36	10.59	4,034	Urban-industrial component	16
11.04	10.45	10.73	4,033	Urban-industrial component	17
11.03	10.24	10.61	4,034	Urban-industrial component	18
11.18	10.05	10.53	4,034	Urban-industrial component	19
10.47	9.68	9.87	4,034	Urban-industrial component	20
9.89	9.55	9.72	4,736	Lead-paint percentile	1
9.88	9.61	9.74	3,292	Lead-paint percentile	2
10.01	9.76	9.88	4,082	Lead-paint percentile	3
10.17	9.93	10.05	4,012	Lead-paint percentile	4
10.33	10.11	10.22	4,027	Lead-paint percentile	5
10.47	10.27	10.38	3,987	Lead-paint percentile	6
10.66	10.46	10.56	3,953	Lead-paint percentile	7
10.75	10.55	10.65	3,976	Lead-paint percentile	8
10.81	10.60	10.71	4,014	Lead-paint percentile	9
10.86	10.68	10.77	4,007	Lead-paint percentile	10
10.93	10.71	10.82	4,043	Lead-paint percentile	11
10.96	10.72	10.84	4,046	Lead-paint percentile	12
10.93	10.66	10.80	4,076	Lead-paint percentile	13
10.95	10.67	10.81	4,108	Lead-paint percentile	14
10.96	10.59	10.77	4,123	Lead-paint percentile	15
11.01	10.63	10.81	4,180	Lead-paint percentile	16
11.05	10.58	10.80	4,203	Lead-paint percentile	17
11.18	10.61	10.88	3,379	Lead-paint percentile	18
11.26	10.60	10.90	4,214	Lead-paint percentile	19
11.47	10.59	10.99	4,216	Lead-paint percentile	20
9.29	8.87	9.05	4,164	ADI national rank	1
9.66	9.24	9.43	3,873	ADI national rank	2
9.88	9.57	9.72	3,982	ADI national rank	3
10.08	9.77	9.93	4,135	ADI national rank	4
10.24	9.91	10.07	3,814	ADI national rank	5
10.40	9.99	10.18	4,039	ADI national rank	6
10.35	10.04	10.20	3,933	ADI national rank	7
10.42	10.15	10.29	4,053	ADI national rank	8
10.49	10.25	10.37	4,132	ADI national rank	9
10.58	10.33	10.46	3,708	ADI national rank	10
10.65	10.42	10.53	4,205	ADI national rank	11
10.78	10.55	10.67	3,968	ADI national rank	12
10.83	10.61	10.72	4,078	ADI national rank	13
10.97	10.79	10.87	3,828	ADI national rank	14
11.12	10.92	11.02	4,143	ADI national rank	15
11.22	11.03	11.12	3,897	ADI national rank	16
11.36	11.17	11.27	4,080	ADI national rank	17
11.58	11.37	11.47	4,042	ADI national rank	18
11.92	11.65	11.78	4,017	ADI national rank	19
12.92	12.44	12.68	3,947	ADI national rank	20

The urban-industrial line starts highest and ends lowest, at 9.9% in the top twentieth, while the lead-paint line rises from 9.7% to 11.0% and the deprivation line from 9.0% to 12.7%. CDC PLACES prevalence is a model-based estimate that uses the age, race, ethnicity and education of a tract's residents, so the deprivation line is partly mechanical. Bands are 95% intervals from a bootstrap over counties.Note: PLACES estimates for tracts come from a model fitted to survey responses, with age, sex, race and ethnicity, education and county poverty as inputs, applied to each tract's population; they are not counts of cases. The lead-paint indicator and the ADI are built from the same Census survey as the population inputs, which makes their lines partly mechanical.Note: Each twentieth holds equal numbers of tracts on its own measure, so the three lines do not describe the same tracts. The tract file carries no intervals for PLACES values.Source: CDC PLACES 2024 release (modeled estimates from BRFSS 2022), EPA EJScreen 2024 and ADI 2023; population-weighted means

Tracts graded D in the 1930s have higher traffic, NO₂ and diesel percentiles

Between 1935 and 1940 the Home Owners' Loan Corporation drew residential security maps for cities and graded neighborhoods from A, labeled best, to D, labeled hazardous.[6] The grades in general are the subject of Chapter 4.1, Home value, income and redlining, and this chapter uses them to look at air only. Each tract takes the grade that covers most of its area, among tracts where graded polygons cover at least 50% of the tract. 10,950 tracts qualify, holding 37.9 million residents in 215 counties.

Pooled across those counties, the traffic percentile steps up from 78 (95% CI 75 to 80) in grade A tracts to 86 (95% CI 83 to 89) in grade D tracts. NO₂ goes from 73 (95% CI 70 to 76) to 86 (95% CI 83 to 89), diesel PM from 73 (95% CI 69 to 77) to 86 (95% CI 81 to 89), and the stacked-hazard score from 2.3 (95% CI 2.0 to 2.5) to 3.2 (95% CI 2.8 to 3.5). PM₂.₅ moves less, from 56 (95% CI 49 to 63) to 60 (95% CI 51 to 69), and the two pooled intervals overlap. Pooled gaps mix the differences between grades with the differences between the cities that were graded.

Comparing tracts within the same county separates the two. In the 199 counties that hold more than one grade, grade D tracts rank 3.9 (95% CI 2.7 to 5.1) points above grade A tracts on the traffic percentile, 9.7 (95% CI 7.2 to 12.2) on NO₂, 5.1 (95% CI 3.9 to 6.5) on diesel PM and 1.1 (95% CI 0.2 to 2.0) on PM₂.₅, and their stacked-hazard score is 0.6 (95% CI 0.4 to 0.7) higher (Figure 9). Each within-county gap is smaller than the pooled gap, and the PM₂.₅ gap is the smallest with an interval that excludes zero. A comparison of the same grades across U.S. cities with modeled NO₂ and PM₂.₅ for 2010 also finds higher NO₂ in grade D than in grade A neighborhoods, and a smaller difference for PM₂.₅.[7] The graded tracts are dense urban cores, so even grade A tracts rank above the national middle for traffic and NO₂.

Figure 9

Within the same county, tracts graded D in the 1930s rank 3.9 points above grade A tracts on the traffic percentile and 9.7 points on NO₂, and the stacked-hazard score is 0.6 higher

Difference from grade A in the same county in the mean national percentile of four air indicators and in the mean stacked-hazard score, by dominant HOLC grade of the tract, population-weighted

Traffic proximity (percentile)

NO₂ (percentile)

Diesel PM (percentile)

PM₂.₅ (percentile)

Stacked air hazards (0 to 5)

Difference from grade A in the same county (units named in each panel title)

Show data

Data for Within the same county, tracts graded D in the 1930s rank 3.9 points above grade A tracts on the traffic percentile and 9.7 points on NO₂, and the stacked-hazard score is 0.6 higher
conf_high	conf_low	contrast	estimate	grade	label	measure	n_units
n/a	n/a	within	0	A	Grade A	Traffic proximity (percentile)	672
1.78	−0.180	within	0.810	B	Grade B	Traffic proximity (percentile)	2,246
3.15	0.860	within	2.06	C	Grade C	Traffic proximity (percentile)	5,054
5.12	2.67	within	3.91	D	Grade D	Traffic proximity (percentile)	2,978
80.28	74.52	pooled	77.55	A	Grade A	Traffic proximity (percentile)	672
84.21	78.81	pooled	81.74	B	Grade B	Traffic proximity (percentile)	2,246
87.02	80.11	pooled	84.16	C	Grade C	Traffic proximity (percentile)	5,054
88.62	82.79	pooled	86.34	D	Grade D	Traffic proximity (percentile)	2,978
n/a	n/a	within	0	A	Grade A	NO₂ (percentile)	672
6.41	2.47	within	4.46	B	Grade B	NO₂ (percentile)	2,246
8.76	4.48	within	6.67	C	Grade C	NO₂ (percentile)	5,054
12.20	7.23	within	9.74	D	Grade D	NO₂ (percentile)	2,978
76.36	69.73	pooled	73.03	A	Grade A	NO₂ (percentile)	672
81.99	74.96	pooled	78.72	B	Grade B	NO₂ (percentile)	2,246
85.72	79.03	pooled	82.73	C	Grade C	NO₂ (percentile)	5,054
89.21	82.52	pooled	86.36	D	Grade D	NO₂ (percentile)	2,978
n/a	n/a	within	0	A	Grade A	Diesel PM (percentile)	672
2.19	0.240	within	1.24	B	Grade B	Diesel PM (percentile)	2,246
3.57	1.71	within	2.70	C	Grade C	Diesel PM (percentile)	5,054
6.45	3.92	within	5.14	D	Grade D	Diesel PM (percentile)	2,978
76.57	68.93	pooled	73.05	A	Grade A	Diesel PM (percentile)	672
81.66	72.33	pooled	77.41	B	Grade B	Diesel PM (percentile)	2,246
84.97	75.26	pooled	81.17	C	Grade C	Diesel PM (percentile)	5,054
89.44	80.50	pooled	86.02	D	Grade D	Diesel PM (percentile)	2,978
n/a	n/a	within	0	A	Grade A	PM₂.₅ (percentile)	672
0.880	−0.010	within	0.420	B	Grade B	PM₂.₅ (percentile)	2,246
1.28	0.030	within	0.670	C	Grade C	PM₂.₅ (percentile)	5,054
2.02	0.250	within	1.13	D	Grade D	PM₂.₅ (percentile)	2,978
62.68	49.05	pooled	55.85	A	Grade A	PM₂.₅ (percentile)	672
61.24	47.08	pooled	54.02	B	Grade B	PM₂.₅ (percentile)	2,246
68.58	46.88	pooled	58.11	C	Grade C	PM₂.₅ (percentile)	5,054
69.39	51.02	pooled	60.02	D	Grade D	PM₂.₅ (percentile)	2,978
n/a	n/a	within	0	A	Grade A	Stacked air hazards (0 to 5)	672
0.340	0.050	within	0.200	B	Grade B	Stacked air hazards (0 to 5)	2,246
0.500	0.240	within	0.380	C	Grade C	Stacked air hazards (0 to 5)	5,054
0.720	0.420	within	0.580	D	Grade D	Stacked air hazards (0 to 5)	2,978
2.52	2.04	pooled	2.29	A	Grade A	Stacked air hazards (0 to 5)	672
2.87	2.29	pooled	2.60	B	Grade B	Stacked air hazards (0 to 5)	2,246
3.35	2.57	pooled	3	C	Grade C	Stacked air hazards (0 to 5)	5,054
3.50	2.85	pooled	3.19	D	Grade D	Stacked air hazards (0 to 5)	2,978

Within a county, dots move right from grade A, the baseline at zero, to grade D for every measure, and the PM₂.₅ differences are the smallest. The second view shows the grade means pooled across all graded tracts, where the traffic percentile goes from 78 in grade A to 86 in grade D, larger gaps that also reflect differences between the cities that were graded. Whiskers are 95% intervals from a bootstrap over counties. Each tract takes the 1930s grade that covers most of its area, and tracts are kept when graded areas cover at least 50% of them.Note: Differences in the first view come from a population-weighted regression of the tract value on grade with county fixed effects, so only counties that hold more than one grade inform them. Grade A is the baseline and has no interval. The pooled view compares tracts across 215 counties.Note: The HOLC maps cover the dense urban cores of cities, so even grade A tracts sit above the national median for traffic and NO₂. The comparison is among graded tracts and does not compare graded with ungraded places.Note: A tract graded D today holds residents of today and not those of the 1930s. Grades A to D follow the original map labels, and areas graded E or F or left blank are set aside. Other outcomes by grade, including the Area Deprivation Index, are in the chapter on home value, income and redlining.Source: Mapping Inequality HOLC maps (1935-1940) matched to 2020 tracts, EPA EJScreen 2024 and ACS 2019-2023; population-weighted means, modeled estimates

Causal evidenceBoundary design; difference-in-differences

Whether redlining and traffic shape the air a neighborhood receives

The comparisons above describe tracts and do not test whether a grade changed the air. Two studies use designs that bear on parts of the question, and a third, in the next box, follows hazardous-waste sites over time.

Salazar-Miranda and coauthors compared neighborhoods on either side of HOLC boundaries within cities, a boundary design. The lower-graded side has higher current and projected heat and flood risk, less tree canopy and less permeable ground.[8] Their outcomes are climate risks and not air pollution, so they do not test whether the grade changed air quality.

Currie and Walker compared births to mothers living near toll plazas with births to mothers living farther from them, before and after E-ZPass reduced congestion at the plazas, a difference-in-differences design. Prematurity and low birth weight fell among mothers near the plazas relative to mothers farther away, which suggests that traffic congestion near a road harms infants.[9] It concerns plazas that adopted electronic tolling and does not say how much of a neighborhood's air pollution comes from a 1930s grade.

What these data cannot show

Percentiles rank tracts and do not give a level of exposure, and a percentile cannot say whether a concentration meets a health guideline. The indicators are EPA model output for 2020 and 2021 with their own error, assigned to tracts from finer geography. The negative link between deprivation and pollution percentiles depends on the measure of deprivation and on the comparison, and each comparison answers a different question. The ADI includes home values and rents, the poverty rate does not, and a within-county comparison holds the metropolitan area fixed while it discards differences between areas.

Everything here is ecological. The shares of Hispanic and Black residents are ACS estimates for 2019 to 2023 without margins of error, the two groups overlap, and a tract average does not describe any one of its residents. Counties in one state share traffic, industry and climate, so the county bootstrap gives intervals narrower than one that redraws whole states: with states redrawn the correlation of the traffic percentile with the ADI rank is −0.41 (95% CI −0.52 to −0.25), which still excludes zero, and that of the PM₂.₅ percentile is −0.18 (95% CI −0.38 to 0.10), which includes it.

The principal components are a description. They use national percentiles, which discard the differences in concentration, and they cover only the tracts with all indicators, so Alaska, Hawaii and Puerto Rico are absent. County scores are means of modeled tract values. PLACES values are modeled from survey responses and demographic inputs, they are crude prevalence among adults, and asthma and COPD depend on age and smoking in ways the tract file does not separate. The HOLC comparison is among graded urban cores, matches a 1930s polygon to a 2020 tract by area, and describes the residents of today. It cannot separate what the grades did from the features of the neighborhoods that surveyors graded, and the box above shows what the available designs do and do not cover.

Data sources

EPA EJScreenEPA EJScreen 2024, tract file (Zenodo archive, release 2.32): national percentiles of 13 indicators on 2020 tracts; air quality modeled for 2020, toxic releases for 2021Years: 2024 (indicators); ACS 2018-2022 (demographics) · Weights: none (modeled indicators); tract population weights from the ACS · License: Public domain (U.S. EPA); archived copy on Zenodo
Area Deprivation IndexNeighborhood Atlas Area Deprivation Index 2023, national rank, tract valuesYears: 2023 release (version 4.0.1; built on ACS 2019-2023) · Weights: ACS 2019-2023 tract population · License: Free for non-commercial use; cite the Neighborhood Atlas and Kind and Buckingham
American Community SurveyACS 2019-2023 tract population, race-alone and Hispanic or Latino shares, poverty rateYears: 2019-2023 (5-year) · Weights: Census Bureau survey weights applied upstream · License: Public domain (U.S. Census Bureau)
CDC PLACESCDC PLACES 2024 release, tract estimates of current asthma and COPD among adults (BRFSS 2022)Years: 2024 release (tracts); 2025 release (ZCTAs, 2023 data year) · Weights: none (model-based estimates); ACS tract population · License: Public domain (CDC, U.S. Department of Health and Human Services)
Mapping Inequality (HOLC maps)Mapping Inequality digitized HOLC maps (1935 to 1940), grades A to D, matched to 2020 tracts by areaYears: 1930s-1940s maps (2023 digitization) · Weights: none (map polygons) · License: CC BY-NC-SA 4.0 (cite Nelson and coauthors)

All data sources

Methods

This chapter compares census tracts on the national percentiles of EPA's EJScreen 2024 air indicators, set against the Area Deprivation Index, race and ethnicity, rural or metropolitan status, PLACES asthma and COPD, and the grades of the 1930s HOLC maps. The indicators are EPA model output for 2020 and 2021, not monitor readings, and each percentile is the share of tracts with a lower value. The tracts in Alaska, Hawaii and Puerto Rico that lack PM2.5 and ozone are left out, and the HOLC comparisons cover only tracts that the digitized maps grade. Estimates are modeled and carry no sampling error beyond the bootstrap intervals, which describe which places are in the file.

Where the air indicators come from

Air indicators are the national percentiles in EPA's EJScreen 2024 tract file, as archived on Zenodo (release 2.32, after EPA removed the tool in 2025). EPA's documentation describes percentiles ranked across block groups; the percentiles in this file rank tracts (their unweighted mean is close to 50), so a percentile is the share of tracts with a lower value.

EPA's documentation dates PM2.5, ozone, NO2 and diesel PM to 2020, toxic releases to air to 2021 and traffic to 2020 traffic counts; PM2.5 is in micrograms per cubic meter, ozone and NO2 in parts per billion, diesel PM in micrograms per cubic meter, and traffic and toxic releases are indices. The indicators are EPA model output, not monitor readings.

Figure explained here: Figure 1

Which tracts are included and how they are weighted

The tract file is on 2020 tracts, like the tract atlas and PLACES, and joins on the 11-character GEOID without a crosswalk (Connecticut carries planning-region codes in all of them); the 686 territory rows of EJScreen are not on the spine and drop out.

PM2.5 and ozone are missing for all tracts in Alaska, Hawaii and Puerto Rico (the tracts with residents in those places are left out of the analysis), and diesel PM is missing for Alaska.

A tract is kept when its ACS 2019-2023 population is positive, and every mean and correlation weights tracts by that population.

How the groups and exposure measures are defined

Deciles and twentieths hold nearly equal numbers of tracts, and tracts tied on the sorting variable share a bin, so the lowest Black-share decile holds all tracts with a share of zero.

Race and ethnicity shares come from the ACS tract file: Hispanic or Latino of any race, and Black or African American alone of any ethnicity, so the groups overlap.

The stacked score counts how many of PM2.5, diesel PM, NO2, traffic proximity and toxic releases to air have a national percentile of 75 or more. Ozone is left out because it forms in the air from other pollutants and does not mark a local source; a count of six that adds it keeps the same ordering of the three sorting variables.

Rural tracts are USDA RUCA codes 7 to 10 and metropolitan tracts codes 1 to 3, from the staged geography file (2010 revised codes mapped to 2020 tracts by largest land overlap); Connecticut has no code.

PLACES values are model-based estimates of crude prevalence among adults.

Figures explained here: Figure 3, Figure 4, Figure 2 and Figure 8

How redlining grades are matched to tracts

HOLC polygons graded A to D are intersected with 2020 tracts in an equal-area projection; a tract takes the grade covering most of its area when graded polygons cover at least half of it.

The pooled grade means compare tracts across counties; the within-county contrasts come from a population-weighted regression of the tract value on grade with county fixed effects.

Figure explained here: Figure 9

How intervals and correlations are computed

Intervals for means and correlations are 95% percentile intervals from 2,000 draws of whole counties with replacement (seed 12345); Connecticut counts as its nine planning regions.

Three kinds of interval differ. The coefficients of the within-county joint models are weighted least squares with errors clustered on county and t intervals. Two correlations in the limits section (traffic and PM2.5 against the ADI rank) and the old-housing, air and race counts of the correlation strip redraw whole states instead, with the same number of draws and seed; the strip shades a cell by the wider of its county and state intervals, and a count requires both intervals to exclude zero. The county-level correlation of mean traffic percentile with mean ADI rank bootstraps counties, and the grade contrasts within counties bootstrap counties as well.

Correlations are population-weighted Pearson correlations. Within-county correlations subtract each county's population-weighted mean from both variables and use counties with at least 5 tracts that have both values.

Estimates are modeled and carry no sampling error beyond these bootstraps, which describe which places are in the file.

Figure explained here: Figure 7

How the principal components and the map are built

The PCA standardizes the 13 percentiles to mean zero and standard deviation one across the 80,754 tracts that have all 13, counts each tract once, and takes the eigenvectors of the correlation matrix; the intervals redraw whole counties, rebuild the matrix and match eigenvectors to the full-sample components. County scores are the population-weighted means of the tract scores on the first component; a county is mapped when at least 50% of its residents live in tracts with all 13 indicators.

Puerto Rico is not drawn on maps.

Figures explained here: Figure 5 and Figure 6

References

Fitzgerald, William; Gehrke, Gretchen (Environmental Data & Governance Initiative) (2025). EPA Environmental Justice Screening Tool (EJScreen) data, 2015-2024. Zenodo, version 1, January 29, 2025; the record includes EPA's EJScreen Technical Documentation, version 2.3, revised July 31, 2024. doi:10.5281/zenodo.14767363
Singh, Gopal K. (2003). Area deprivation and widening inequalities in US mortality, 1969-1998. American Journal of Public Health 93(7): 1137-1143. doi:10.2105/ajph.93.7.1137
Bullard, Robert D. (1990). Dumping in Dixie: Race, Class, and Environmental Quality. Westview Press, Boulder, CO.
Tessum, Christopher W.; Apte, Joshua S.; Goodkind, Andrew L.; Muller, Nicholas Z.; Mullins, Kimberley A.; Paolella, David A.; Polasky, Stephen; Springer, Nathaniel P.; Thakrar, Sumil K.; Marshall, Julian D.; Hill, Jason D. (2019). Inequity in consumption of goods and services adds to racial-ethnic disparities in air pollution exposure. Proceedings of the National Academy of Sciences 116(13): 6001-6006. doi:10.1073/pnas.1818859116
Greenlund, Kurt J.; Lu, Hua; Wang, Yan; Matthews, Kevin A.; LeClercq, Jennifer M.; Lee, Benjamin; Carlson, Susan A. (2022). PLACES: Local Data for Better Health. Preventing Chronic Disease 19: 210459. doi:10.5888/pcd19.210459
Nelson, Robert K.; Winling, LaDale; Marciano, Richard; Connolly, Nathan; et al. (2023). Mapping Inequality: Redlining in New Deal America. American Panorama, ed. Robert K. Nelson and Edward L. Ayers, Digital Scholarship Lab, University of Richmond. dsl.richmond.edu/panorama/redlining/
Lane, Haley M.; Morello-Frosch, Rachel; Marshall, Julian D.; Apte, Joshua S. (2022). Historical Redlining Is Associated with Present-Day Air Pollution Disparities in U.S. Cities. Environmental Science & Technology Letters 9(4): 345-350. doi:10.1021/acs.estlett.1c01012
Salazar-Miranda, Arianna; Conzelmann, Claire; Phan, Toan; Hoffman, Jeremy (2024). Long-term effects of redlining on climate risk exposure. Nature Cities 1(6): 436-444. doi:10.1038/s44284-024-00076-y
Currie, Janet; Walker, Reed (2011). Traffic Congestion and Infant Health: Evidence from E-ZPass. American Economic Journal: Applied Economics 3(1): 65-90. doi:10.1257/app.3.1.65
Mohai, Paul; Saha, Robin (2015). Which came first, people or pollution? Assessing the disparate siting and post-siting demographic change hypotheses of environmental injustice. Environmental Research Letters 10(11): 115008. doi:10.1088/1748-9326/10/11/115008

How to cite

@misc{singh2026airpollution,
  author = {Singh, Vishal},
  title = {Air pollution},
  year = {2026},
  url = {https://vishalsingh.org/zipcode/air-pollution},
  note = {In ZIP Code Destiny. Published 9 October 2026.}
}

What the pollution indicators measure

Pollution percentiles fall as deprivation rises

The fall persists among metropolitan tracts for traffic and diesel PM, and the sign reverses within counties

Pollution percentiles rise with the Hispanic and Black share of a tract

Hazards stack in the same tracts

Two components summarize the indicators

The urban-industrial component is highest in a few dense counties

Most air indicators correlate negatively with deprivation and positively with Hispanic and Black share

Modeled asthma does not follow the urban-industrial component

Tracts graded D in the 1930s have higher traffic, NO₂ and diesel percentiles

What these data cannot show

Data sources

Methods

Where the air indicators come from

Which tracts are included and how they are weighted

How the groups and exposure measures are defined

How redlining grades are matched to tracts

How intervals and correlations are computed

How the principal components and the map are built

References

How to cite

Reading paths