Chapter 3.3 · Two ladders, one street

Race, income and mobility

White children from low-income families reached a higher adult income rank than Black children raised in the same tract in 92.5% of the 30,997 tracts where both are estimated, and across the 1,265 counties with life expectancy estimates for both groups, weighted by population, non-Hispanic white life expectancy exceeds non-Hispanic Black by 4.4 years.

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

White children from low-income families reached a higher adult income rank than Black children from low-income families raised in the same census tract in 92.5% of the 30,997 tracts where the Opportunity Atlas estimates both groups.[1] The mean rank, weighted by tract population, was 44.9 for white children and 34.6 for Black children, a gap of 10.3 (95% CI 10.0 to 10.6) percentile points. Both groups were born in 1978 to 1983 to parents at the 25th percentile of the national income distribution, and the rank is the position they reached in the national distribution of household income in their thirties. The tracts hold 144.8 million people, 46.9% of the 2010 population.

Comparing two groups inside one tract removes the differences between neighborhoods that a county or ZIP code average mixes in. The chapter applies the same logic to parental income and to life expectancy within counties, and it asks what predicts upward mobility across counties. Every Atlas figure describes children who grew up in the 1980s and 1990s, and none describes the same tracts today.

Upward mobility across counties

The Atlas follows the children of each tract from childhood into adult tax and census records, so it can map how far children from a given parental income level rise.[1] For parents at the 25th percentile, the mean adult rank in a county averages 41.3 across counties weighted by population. It runs from 34.9 in the county at the fifth percentile to 48.4 in the county at the ninety-fifth. Among the states, means are highest in North Dakota (53.3) and lowest in South Carolina (35.9), and the highest state means lie in the Plains and the Mountain West while the lowest lie in the Southeast (Figure 1). The District of Columbia, a single county that is left out of that ranking, has a lower value than any state, 34.9. The map averages over every child in a county, and the next section separates two groups of them.

Figure 1

Children from low-income families reach the highest adult income ranks in the Plains and Mountain West and the lowest in the Southeast

Mean adult household income rank (percentile) of children born 1978 to 1983 to parents at the 25th income percentile, by county of childhood

Loading county values

Mean adult household income rank of children from low-income families (percentile)

Lower adult income rankHigher adult income rank

Values outside 33.3 to 49.3 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.

Teal counties, where children from low-income families reach higher adult ranks, cluster in the Plains and Mountain West, and brown counties cluster in the Southeast. The map plots each county's mean rank on a scale centered on the population-weighted county mean of 41.3; the 5 gray counties have no estimate.Note: Each gray county has the reason for its missing value in its tooltip. Puerto Rico has estimates in the Atlas but is not drawn.Note: The county estimates are model-based and carry privacy noise, so single small counties are noisy. The Atlas lists Connecticut under its legacy counties, which are the eight drawn here, and Valdez-Cordova Census Area (split in 2019) is drawn as its two successors with the same value.Note: Alaska is drawn. The Atlas estimates cover 1978 to 1983 birth cohorts and describe where children grew up, not the counties today.Source: Opportunity Atlas county outcomes, 1978-1983 birth cohorts; modeled estimates, not weighted

Black and white children from the same tract

Because the Atlas reports outcomes by race, it can place children of two groups in one tract, with parents at the same income percentile, and compare them. Figure 2 plots one dot per tract. If a tract gave children of both groups the same outcome, its dot would sit on the diagonal. The dot sits below the diagonal in 92.5% of the tracts. A tract has both estimates only where the Atlas can measure enough children of each group, so these tracts are those with sizeable Black and white populations and not a sample of all tracts. The Atlas keeps Hispanic children, of any race, apart from non-Hispanic white, Black and Asian children, and the white, Black and Asian children in this chapter are non-Hispanic.[2]

The gap is wider where white children do best, though by less than a tract-by-tract comparison suggests. Averaging the tracts within each county damps the noise in the tract estimates, and 336 counties have at least 20 tracts with both estimates. In the lowest-ranked third of these counties by white rank, white children ranked 8.4 (95% CI 8.1 to 8.7) points above Black children, and in the highest-ranked third they ranked 11.6 (95% CI 10.8 to 12.1) points above. Across the counties, each point of difference in the white rank goes with a difference of 0.60 (95% CI 0.49 to 0.71) points in the Black rank, so where white children rank higher, Black children rank higher too, by a smaller amount.

Sorting tracts on the white rank itself gives a larger contrast, 5.6 points in the lowest-ranked third and 14.8 in the highest, and a slope of 0.33. Noise in the white estimate both sorts the tracts and enters the white minus Black difference, so the tract figures overstate the contrast.

Figure 2

White children from low-income families reached a higher adult income rank than Black children from the same tract in 92.5% of 30,997 tracts

Adult household income rank (percentile) of Black children against white children raised in the same census tract by parents at the 25th income percentile, one dot per tract

Loading data

Most dots lie below the diagonal, which marks equal outcomes, and the markers fall further below it where white children do better. Dots are a random sample of 5,000 tracts drawn with a fixed seed from the 30,997 that have both estimates, and 35 tracts fall outside the plotted range; markers are population-weighted mean Black ranks in tenths of 336 counties ranked by their mean white rank, with 95% intervals from a bootstrap over states.Note: The dots are a random sample of 5,000 tracts inside the plotted range (seed 12345); every estimate in the text uses all 30,997 tracts. Tracts have a Black and a white estimate only where the Atlas has enough children of both groups, so they are not a sample of all tracts.Note: The markers average the tracts of each county that has at least 20 tracts with both estimates, because sorting tracts on the noisy white rank would flatten a line of tract means and widen the apparent contrast. Each tract is weighted by its 2010 census population, the nearest available stand-in for the number of children. Ranks are percentiles of the national distribution of household income for children born in 1978 to 1983, measured in 2014 to 2015. The Opportunity Atlas adds noise to its estimates to protect privacy and the staged values are never clipped, so a few fall outside 0 to 100.Source: Opportunity Atlas tract outcomes, 2010 tracts, 1978-1983 birth cohorts; population-weighted (2010 tract population), modeled estimates

Each tract's estimate for a group rests on a modest number of children, and noise is added to protect privacy. Noise of that kind spreads dots in both directions around the diagonal. It would pull the share of tracts below the diagonal toward an even split, not away from one.

The comparison is not limited to two groups. In tracts where white and Hispanic children are both estimated, Hispanic children ranked 2.6 (95% CI 2.3 to 2.9) points below white children, and in tracts where white and Asian children are both estimated, Asian children ranked 9.4 (95% CI 8.9 to 9.8) points above them (Figure 3). Each comparison uses its own set of tracts, so the three gaps are not measured on one set of tracts. The Atlas also measures incarceration on Census Day 2010, and Chapter 3.1, School discipline and incarceration reports the same-tract comparison for that outcome.

Figure 3

Black children ranked 10.3 points below white children in the same tracts, Hispanic children 2.6 points below and Asian children 9.4 points above

Mean adult household income rank of each group minus that of white children raised in the same tract by parents at the 25th income percentile, tracts where both are estimated

Group's mean rank minus white children's mean rank (percentile points)

Show data

Data for Black children ranked 10.3 points below white children in the same tracts, Hispanic children 2.6 points below and Asian children 9.4 points above
conf_high	conf_low	estimate	group	n
−10.03	−10.60	−10.31	Black children	30,997
−2.32	−2.92	−2.61	Hispanic children	35,579
9.83	8.89	9.37	Asian children	15,078

Black children's dot lies furthest left, Hispanic children's near zero and Asian children's to the right of zero, and none of the three intervals reaches zero. Each dot is a population-weighted mean over the tracts where the group and white children are both estimated, with a 95% interval from a bootstrap over counties.Note: Each row uses its own set of tracts, the tracts where that group and white children both have an estimate, so the rows are not a single sample. The number of tracts is in each row's tooltip. Ranks are percentiles of the national distribution of household income for children born in 1978 to 1983, measured in 2014 to 2015. The Opportunity Atlas adds noise to its estimates to protect privacy and the staged values are never clipped, so a few fall outside 0 to 100.Source: Opportunity Atlas tract outcomes, 2010 tracts, 1978-1983 birth cohorts; population-weighted (2010 tract population), modeled estimates

The advantage of a higher-income family, tract by tract

The Atlas also reports adult ranks for children whose parents were at the 75th percentile, with races pooled. The difference between that rank and the rank of children at the 25th percentile is a tract-level measure of the advantage of a higher-income family, which this chapter calls the premium. Across 71,924 tracts the mean rank was 43.2 for children at the 25th percentile and 56.8 for children at the 75th, a premium of 13.6 (95% CI 13.2 to 13.9) points. The premium varies across tracts, from 5.8 at the fifth percentile of tracts to 21.3 at the ninety-fifth, and sampling noise in the tract estimates widens that range.

The premium is smaller where children from low-income families do better. Across tracts, the rank of children at the 75th percentile rises 0.69 (95% CI 0.66 to 0.72) points for each point of the rank of children at the 25th, a line flatter than the diagonal in Figure 4, so the two ladders sit closer together in tracts where the lower ladder is higher. Sampling noise in the rank of children at the 25th percentile pulls a tract slope toward zero, so the slope is also estimated on county averages. It is 0.70 (95% CI 0.60 to 0.82) for counties with at least 10 tracts and 0.81 (95% CI 0.69 to 0.91) for counties with at least 100 tracts, higher than the tract slope and still flatter than the diagonal.

Figure 4

Children raised at the 75th income percentile reach adult ranks 13.6 points above those raised at the 25th in the same tract, a difference that is smaller where the latter do better

Adult household income rank (percentile) of children raised at the 75th income percentile against those raised at the 25th, one dot per tract, races pooled

Loading data

Dots lie above the diagonal, which would mean no advantage, and the fitted line is flatter than the diagonal, so the gap between the two ladders is smaller in tracts where children from low-income families do better. Dots are a random sample of 5,000 of 71,924 tracts, and 15 tracts fall outside the plotted range; markers are population-weighted decile means with 95% intervals from a bootstrap over counties.Note: The dots are a random sample of 5,000 tracts inside the plotted range (seed 12345); the means, slope and intervals use all 71,924 tracts with both estimates.Note: The fitted line is the population-weighted least-squares line through the tracts, with slope 0.69. Sampling noise in the 25th-percentile rank pulls a tract slope toward zero. Averaging tracts to counties damps that noise: the slope is 0.70 for counties with at least 10 tracts and 0.81 for counties with at least 100, and all of them are flatter than the diagonal. Ranks are percentiles of the national distribution of household income for children born in 1978 to 1983, measured in 2014 to 2015. The Opportunity Atlas adds noise to its estimates to protect privacy and the staged values are never clipped, so a few fall outside 0 to 100.Source: Opportunity Atlas tract outcomes, 2010 tracts, 1978-1983 birth cohorts; population-weighted (2010 tract population), modeled estimates

Among the 148 counties with at least 100 tracts, the premium is smallest in the Bronx (8.0), and the 8 smallest premiums are all in New York City or California, among them San Francisco (8.9). It is largest in Hamilton County, Ohio (17.3), and Luzerne County, Pennsylvania (16.9). A small premium is not a sign of a better place for children from low-income families. In the Bronx, children at the 25th percentile reached a mean rank of 39.9 and children at the 75th percentile reached 47.9, both below the mean for all tracts. In San Francisco the ranks were 50.3 and 59.2, both above it.

Friendships across incomes and upward mobility

Across counties, upward mobility moves with who a place's residents befriend. The Social Capital Atlas builds economic connectedness from the friendships of Facebook users. It compares the share of high-status friends among people with below-median status to the share expected if friendships ignored status.[3] Across 2,942 counties, economic connectedness correlates +0.69 (95% CI 0.64 to 0.74) with the mean adult rank of children from low-income families, weighted by population.

That correlation is larger than the correlation of any civic or cohesion measure with mobility (Figure 5). The density of civic organizations correlates +0.06 (95% CI −0.01 to 0.14) with it, the volunteering rate +0.19 (95% CI 0.12 to 0.25) and the network support ratio −0.25 (95% CI −0.33 to −0.17). Civic participation is the subject of Chapter 7.1, Civic life, loneliness and partnership.

Figure 5

Across counties, economic connectedness correlates +0.69 with upward mobility, more than any civic or cohesion measure

Population-weighted correlation of each Social Capital Atlas county measure with the mean adult income rank of children from low-income families

Contact across incomes
Cohesion
Civic life

Show data

Data for Across counties, economic connectedness correlates +0.69 with upward mobility, more than any civic or cohesion measure
conf_high	conf_low	estimate	group	measure	n
0.741	0.644	0.693	Contact across incomes	Economic connectedness	2,942
0.734	0.650	0.695	Contact across incomes	Exposure to high-income friends	2,939
−0.076	−0.316	−0.192	Contact across incomes	Friending bias	2,939
0.117	−0.012	0.052	Cohesion	Network clustering	2,983
−0.168	−0.332	−0.254	Cohesion	Support ratio	2,983
0.253	0.121	0.187	Civic life	Volunteering rate	2,983
0.138	−0.011	0.062	Civic life	Civic organizations	2,983

The two measures of contact across incomes, economic connectedness and exposure to high-income friends, sit furthest to the right, and the civic and cohesion measures sit close to zero or below it. Each dot is a population-weighted correlation across counties, with a 95% interval from a bootstrap over counties; the number of counties is in the tooltip.Note: Economic connectedness is built from Facebook friendships among active users in 2022, while the mobility ranks describe children who grew up in the 1980s and 1990s, so the correlation links a recent network measure to an earlier cohort.Note: Intervals resample whole counties. Counties in one state tend to resemble each other, and resampling whole states gives a wider interval for every measure.Note: Counties with a missing measure drop out of that row only, so the number of counties differs by row.Source: Social Capital Atlas county measures, 2022, and Opportunity Atlas county outcomes, 1978-1983 birth cohorts; population-weighted (2018 county population)

Connectedness also moves with income. Across 18,896 ZIP code areas (ZCTAs), economic connectedness correlates +0.82 (95% CI 0.81 to 0.83) with the area's rank in median household income. Its mean is 0.57 (95% CI 0.56 to 0.58) in the lowest tenth of areas by median income and 1.19 (95% CI 1.16 to 1.21) in the highest tenth. Because connectedness is higher where incomes are higher, the county correlation could partly reflect income.

With the log of median household income, the Black share of residents, the share with a college degree and the poverty rate in the same weighted regression, the standardized coefficient on economic connectedness is 0.56 (95% CI 0.42 to 0.70) and the coefficient on the log of income is 0.46 (95% CI 0.26 to 0.67). The difference between them is 0.10 (95% CI −0.16 to 0.36), and its interval includes zero, so the data do not rank the two. The coefficient on connectedness stays positive with income in the model.

Economic connectedness is measured in 2022, long after the 1978 to 1983 cohorts grew up. The correlation links a recent network measure to an earlier cohort, and the data cannot show that friendships across incomes change a child's later income.

Life expectancy of Black and white residents of the same county

The comparison inside one place extends to adult health. In County Health Rankings, 1,265 counties publish a life expectancy for both non-Hispanic white and non-Hispanic Black residents, for deaths in 2019 to 2021.[4] White life expectancy exceeds Black by 4.4 (95% CI 4.0 to 4.8) years when counties count in proportion to their population, and by 3.5 (95% CI 3.3 to 3.7) years when each county counts once. The population-weighted means are 78.1 years for white residents and 73.7 years for Black residents. White life expectancy is the higher in 89% of these counties, and the counties hold 98% of the non-Hispanic Black residents of all counties in the file.

Figure 6 shows the 20 most populous counties in that sample. The gap is positive in each, from 1.1 (95% CI 0.7 to 1.5) years in Broward County, Florida, to 9.9 (95% CI 9.7 to 10.2) years in Cook County, Illinois. Among all 1,265 counties, the widest gap is in the District of Columbia, at 17.7 (95% CI 17.0 to 18.3) years.

Figure 6

In each of the 20 most populous counties, white residents live longer than Black residents, by 1.1 to 9.9 years

Life expectancy at birth in years, non-Hispanic Black and non-Hispanic white residents, for the most populous counties with both estimates, deaths in 2019 to 2021

Non-Hispanic Black residents
Non-Hispanic white residents

Show data

Data for In each of the 20 most populous counties, white residents live longer than Black residents, by 1.1 to 9.9 years
county	hi_black	hi_white	le_black	le_white	lo_black	lo_white	population
Cook County, IL	70.41	80.28	70.23	80.17	70.04	80.06	5109292
Alameda County, CA	73.56	81.60	73.07	81.37	72.59	81.15	1628997
Wayne County, MI	69.01	76.61	68.77	76.43	68.53	76.25	1757043
Los Angeles County, CA	73.40	79.93	73.19	79.84	72.97	79.74	9721138
Maricopa County, AZ	73.51	79.17	73.08	79.06	72.65	78.95	4551524
King County, WA	76.03	81.56	75.48	81.41	74.92	81.26	2266789
Miami-Dade County, FL	73.69	78.80	73.37	78.53	73.06	78.25	2673837
Harris County, TX	72.89	77.81	72.68	77.67	72.47	77.52	4780913
San Diego County, CA	76.09	80.64	75.56	80.51	75.04	80.37	3276208
Dallas County, TX	72.92	77.72	72.67	77.53	72.42	77.34	2600840
Kings County, NY	77.81	82.14	77.57	81.97	77.33	81.79	2590516
Clark County, NV	72.18	76.33	71.80	76.15	71.42	75.98	2322985
Santa Clara County, CA	79.30	82.61	78.26	82.40	77.21	82.18	1870945
San Bernardino County, CA	71.37	74.84	70.92	74.61	70.46	74.39	2193656
Riverside County, CA	74.90	77.92	74.39	77.72	73.88	77.53	2473902
Tarrant County, TX	74.29	77.28	73.95	77.11	73.61	76.94	2154595
Orange County, CA	78.78	81.01	77.87	80.86	76.96	80.72	3151184
Queens County, NY	77.92	80.61	77.62	80.39	77.31	80.17	2278029
Bexar County, TX	74.40	76.82	73.89	76.59	73.38	76.36	2059530
Broward County, FL	77.48	78.50	77.18	78.28	76.88	78.05	1947026

Every white dot lies to the right of the Black dot, and the last column gives the gap. The plot shows life expectancy for each group with the 95% interval published by County Health Rankings; rows run from the widest gap to the narrowest.Note: County Health Rankings computes life expectancy from NCHS death counts for racial and ethnic groups that do not overlap, so the Black and white groups here are non-Hispanic and exclude Hispanic residents of any race. Counties with fewer deaths have wider intervals, and the intervals in this figure are County Health Rankings' own.Note: The counties are the most populous ones among those with both estimates, which is the sample behind the headline gap.Source: County Health Rankings 2024 (NCHS deaths 2019-2021); full counts of deaths, county life expectancy by race

In 140 counties the estimate for Black residents is the higher. County Health Rankings publishes an interval for each estimate, and the two intervals separate in that direction in only 9 counties, against 677 counties where the interval for white residents lies entirely above the interval for Black residents. The counties with a higher Black estimate have a median population of 59,039, below the 82,911 median of the sample, which is consistent with noisier estimates in smaller counties.

The gap extends to premature mortality, which County Health Rankings measures as age-adjusted deaths of residents under age 75 per hundred thousand residents. Across 1,242 of these counties that report both rates, the population-weighted mean for Black residents is 1.5 (95% CI 1.5 to 1.6) times the mean for white residents. This is a death rate and not the years of potential life lost. Infant mortality by race is covered in Chapter 1.2, Infant mortality.

The gap is wider where residents are more segregated. The population-weighted correlation between the gap and the Black-white index of dissimilarity, which gives the share of either group that would have to move for the two to be spread evenly across tracts, is +0.30 (95% CI 0.14 to 0.45) across 1,260 counties. Within states, each additional point of the index goes with 0.06 (95% CI 0.00 to 0.12) more years of gap (state fixed effects), and the lower end of that interval is close to zero.

Race, place and their interaction

Life expectancy differs across counties and between the two groups, and a decomposition says how much of the variation each source holds. The cells are the county-by-race life expectancies, a white and a Black estimate for each of the 1,265 counties. The variance across the cells splits into three parts. Place is the spread of the county averages, race is the average white-Black gap, and the interaction is the spread of the gap across counties. Weighted by county population, place accounts for 53% (95% CI 48% to 58%) of the variance, race for 31% (95% CI 26% to 37%) and the interaction for 16% (95% CI 13% to 19%). When each county counts once, the shares are 64% (95% CI 62% to 66%) for place, 19% (95% CI 17% to 22%) for race and 17% (95% CI 15% to 19%) for the interaction. The race share is larger when large counties carry more weight, because the average gap is larger there, so the chapter reports both.

Figure 7

Place accounts for 53% of the variance in county life expectancy by race, race for 31% and their interaction for 16%

Shares of the variance of county-by-race life expectancy (a non-Hispanic white and a non-Hispanic Black estimate for each county), percent, deaths in 2019 to 2021

Place
Race
Race-by-place interaction

Show data

Data for Place accounts for 53% of the variance in county life expectancy by race, race for 31% and their interaction for 16%
Weighting	Component	Share of variance (%)	Lower limit of the 95% interval	Upper limit of the 95% interval
Weighted by county population	Place	53.0	47.5	58.5
Weighted by county population	Race	31.3	25.7	37.1
Weighted by county population	Race-by-place interaction	15.7	13.0	18.8
Each county counted once	Place	64.1	61.8	66.5
Each county counted once	Race	19.1	16.7	21.7
Each county counted once	Race-by-place interaction	16.8	15.0	18.6

The place segment is the longest in both bars, and the race segment is shorter when each county counts once. Each bar splits the variance of life expectancy across county-by-race cells into the spread of county averages (place), the average white-Black gap (race) and the spread of the gap across counties (interaction).Note: Shares are exact for the county estimates in the file and sum to 100 in each bar. The 95% intervals in the table come from a bootstrap over counties.Note: The estimates carry sampling noise that is largest in small counties; the noise sits inside the place and interaction shares and is not removed.Note: Both groups of a county get the same weight, which keeps the split exact: the interaction equals the variance of the county gaps divided by four.Source: County Health Rankings 2024 (NCHS deaths 2019-2021); full counts of deaths, weighted by county population in the first bar and unweighted in the second

Weighted by county population, the 90th percentile of Black life expectancy across counties is 78.2 years, above the 10th percentile of white life expectancy, 74.4 years, so the two groups' distributions across places overlap, and the white estimate is still the higher in 89% of counties.

Causal evidenceRandomized housing-voucher experiment; movers design

Whether place changes a child's adult outcomes

The comparisons above describe differences across places and groups. Two designs test whether place changes a child's own outcome.

Moving to Opportunity was a randomized housing-voucher experiment. Families in high-poverty public housing were randomly assigned to be offered a voucher that could be used only in a low-poverty neighborhood, to be offered an unrestricted voucher, or to stay in public housing. Chetty, Hendren and Katz followed the children into adulthood and found that those who were young when their families moved earned more and attended college more often as adults, while the estimated effects were negative for children who were teenagers at the move.[5]

Chetty and Hendren used a movers design. They compared children whose families moved between counties and commuting zones when the children were at different ages, including siblings in the same family. Because the age at the move varies while the family stays the same, the design isolates the years a child spent in the destination. Children's adult outcomes moved toward those of children raised in the destination in proportion to the years spent there.[6] A companion paper applies the design county by county,[7] and the Opportunity Atlas authors combine the experimental and movers estimates and report that 60% of the variation in outcomes across neighborhoods is driven by causal effects of the neighborhoods.[1]

Neither design tests whether the Black-white gap inside a tract would close if the tract changed. Both estimate how much a move changes a child's own outcome, and neither compares the gap before and after.

What these data cannot show

The patterns here describe tracts, ZIP code areas and counties. The average life expectancy of Black residents of Cook County says little about how long any one Black resident of Cook County will live, because the data are averages for places and groups (ecological inference).

A same-tract comparison removes the differences between neighborhoods but not all differences. Within a tract, families at one income percentile can still differ in wealth, in the exact block, in schooling and in how institutions treat them. The gap is a residual, and the data describe it without explaining it. Chetty, Hendren, Jones and Porter find that the gap in individual adult income between Black and white children raised at the same parental income is concentrated among boys and men, and that it is smaller in tracts with lower poverty and a higher share of Black fathers present, a comparison across places and not an experiment.[2] The ranks here pool boys and girls and measure household income, which includes a spouse's income, because race by gender is not available in the staged data (see Chapter 3.5, Boys and mobility (planned)).

The Atlas estimates are for children born in 1978 to 1983 and observed in 2014 to 2015. Each tract estimate rests on a modest number of children and carries privacy noise, which is not clipped. Tracts weigh in by their 2010 population because the file has no child counts, and the tracts with both estimates are those with enough children of both groups.

The life-expectancy estimates are computed from death counts that are small in small counties, so the share of counties with a reversed gap and the shares in the decomposition include sampling noise. County Health Rankings counts Black and white residents who are not Hispanic, and the index of dissimilarity is measured for 2018 to 2022. The connectedness measures come from active Facebook users and describe places in 2022, and the regression cannot separate connectedness from income more finely than its interval allows. Puerto Rico has Atlas estimates but is not drawn on the maps, and Connecticut is missing from the regression because its tract codes differ across the files.

Data sources

Opportunity AtlasTract and county outcomes of children born 1978 to 1983, native 2010 tracts, ranks as percentiles; income measured in 2014 to 2015Years: 1978-1983 birth cohorts (2018 release) · Weights: 2010 census tract population for tract comparisons; county population for county summaries · License: Public use (Opportunity Insights and the U.S. Census Bureau)
Social Capital AtlasEconomic connectedness and network measures, county and ZIP code, 2022Years: 2022 (friendship snapshot) · Weights: 2018 population (Social Capital Atlas) · License: Open for research use (Opportunity Insights)
County Health RankingsCounty Health Rankings 2024: life expectancy by race (deaths 2019 to 2021), premature death by race, Black-white residential segregationYears: 2024 release · Weights: county total population; each county once where stated · License: CC BY 4.0 (University of Wisconsin Population Health Institute)
American Community SurveyACS 2019 to 2023 tract file, aggregated to counties for the connectedness regressionYears: 2019-2023 (5-year) · Weights: 2019-2023 tract population · License: Public domain (U.S. Census Bureau)
USDA Food Access Research Atlas2010 census population of each 2010 tract, from the Food Access Research Atlas file; the food-access flags are not usedYears: 2019 release (population from the 2010 census) · Weights: none (census counts); used as the weight for tract comparisons · License: Public domain (U.S. Department of Agriculture, Economic Research Service)

All data sources

Methods

The tract comparisons use Opportunity Atlas estimates on native 2010 tracts for children born in 1978 to 1983 and observed in 2014 to 2015, and a tract enters a comparison only when both estimates exist. Atlas privacy noise is not clipped, and tracts are weighted by their 2010 census population because the Atlas has no child counts. The county comparisons of life expectancy by race use County Health Rankings estimates for deaths in 2019 to 2021, for counties with both a non-Hispanic white and a non-Hispanic Black value, so the mobility outcomes and the life-expectancy outcomes refer to different periods. The connectedness regression drops Connecticut.

Which tracts and children the Atlas comparisons cover

Tract comparisons use the Opportunity Atlas on native 2010 tracts. Ranks are percentiles of the national household income distribution for children born in 1978 to 1983, observed in 2014 to 2015; race-specific ranks are staged for parents at the 25th percentile with genders pooled, and the 75th-percentile rank only with races pooled. Atlas privacy noise is not clipped.

A tract enters a comparison when both estimates exist and its 2010 population is positive.

Figures explained here: Figure 1, Figure 2, Figure 3 and Figure 4

How life expectancy by race and the variance split are defined

Life expectancy by race is the County Health Rankings 2024 estimate, computed from NCHS death counts for deaths in 2019 to 2021, for counties with both a non-Hispanic white and a non-Hispanic Black value; the white-Black gap within counties, weighted and unweighted, comes from the book's shared routine.

The variance split treats the white and Black life expectancy of each county as two cells, gives both the same weight (county population, or one), and divides the weighted variance of the cells into the variance of county averages (place), the squared difference between the group means (race) and the remainder, which equals the weighted variance of the county gaps divided by four (interaction); the shares sum to 100.

Figures explained here: Figure 6 and Figure 7

How tracts and counties are weighted and joined

Each tract is weighted by its 2010 census population, read from the USDA Food Access Research Atlas file, which carries the 2010 census count of every 2010 tract (the Atlas has no child counts); counting each tract once gives similar results, and the share of tracts in which white children rank higher counts each tract once.

County mobility ranks, the Social Capital Atlas measures and ACS controls are joined on county codes; correlations are weighted by the Atlas's 2018 population, and the regression is a weighted fit of standardized variables with errors clustered on state (Connecticut drops out because the ACS tract file uses planning-region codes).

Figures explained here: Figure 2, Figure 3, Figure 4 and Figure 5

How the gap across the white rank is estimated

The contrast of the white-Black gap across the white rank is estimated at county grain, on counties with at least 20 tracts that have both estimates, because sorting tracts on a noisy rank builds the contrast in; the county slopes of the premium use counties with at least 10 and at least 100 tracts, with state-bootstrap intervals.

Figures explained here: Figure 2 and Figure 4

How uncertainty is measured

Intervals are 95% percentile intervals from 2,000 draws of whole counties with replacement (seed 12345), computed by the same routine for means, shares, slopes and correlations; state-level quantities use states as the resampling unit. Scatter plots show a random sample of tracts inside the plotted range drawn with the same seed; every estimate uses all tracts.

Intervals on the shares resample counties. Counties with a missing estimate are not drawn and are not treated as zero.

References

Chetty, Raj; Friedman, John N.; Hendren, Nathaniel; Jones, Maggie R.; Porter, Sonya R. (2018). The Opportunity Atlas: Mapping the Childhood Roots of Social Mobility. NBER Working Paper 25147. doi:10.3386/w25147
Chetty, Raj; Hendren, Nathaniel; Jones, Maggie R.; Porter, Sonya R. (2020). Race and Economic Opportunity in the United States: An Intergenerational Perspective. The Quarterly Journal of Economics 135(2): 711-783. doi:10.1093/qje/qjz042
Chetty, Raj; Jackson, Matthew O.; Kuchler, Theresa; Stroebel, Johannes; et al. (2022). Social capital I: measurement and associations with economic mobility. Nature 608: 108-121. doi:10.1038/s41586-022-04996-4
University of Wisconsin Population Health Institute (2024). County Health Rankings & Roadmaps Technical Documentation, Version 2. University of Wisconsin Population Health Institute (updated March 18, 2024). www.countyhealthrankings.org/sites/default/files/media/document/2024%20CHRR%20Technical%20Document_1.pdf
Chetty, Raj; Hendren, Nathaniel; Katz, Lawrence F. (2016). The Effects of Exposure to Better Neighborhoods on Children: New Evidence from the Moving to Opportunity Experiment. American Economic Review 106(4): 855-902. doi:10.1257/aer.20150572
Chetty, Raj; Hendren, Nathaniel (2018). The Impacts of Neighborhoods on Intergenerational Mobility I: Childhood Exposure Effects. The Quarterly Journal of Economics 133(3): 1107-1162. doi:10.1093/qje/qjy007
Chetty, Raj; Hendren, Nathaniel (2018). The Impacts of Neighborhoods on Intergenerational Mobility II: County-Level Estimates. The Quarterly Journal of Economics 133(3): 1163-1228. doi:10.1093/qje/qjy006

How to cite

@misc{singh2026raceincomeandmobility,
  author = {Singh, Vishal},
  title = {Race, income and mobility},
  year = {2026},
  url = {https://vishalsingh.org/zipcode/race-income-and-mobility},
  note = {In ZIP Code Destiny. Published 9 October 2026.}
}

Upward mobility across counties

Black and white children from the same tract

The advantage of a higher-income family, tract by tract

Friendships across incomes and upward mobility

Life expectancy of Black and white residents of the same county

Race, place and their interaction

What these data cannot show

Data sources

Methods

Which tracts and children the Atlas comparisons cover

How life expectancy by race and the variance split are defined

How tracts and counties are weighted and joined

How the gap across the white rank is estimated

How uncertainty is measured

References

How to cite

Reading paths