Chapter 9.1 · The eight Americas

Eight types of neighborhood

K-means sorts 77,745 census tracts into eight types whose mean life expectancy for deaths in 2010 to 2015, itself one of the 13 inputs, runs from 72.3 years to 82.3 years, and a silhouette of 0.15 says the types are not sharply separated.

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

K-means sorts the 77,745 census tracts that hold 311.7 million residents into eight types. Mean life expectancy for deaths in 2010 to 2015 differs by 10.0 (95% CI 9.8 to 10.2) years between the types, from 72.3 (95% CI 72.2 to 72.5) years in the high-poverty type to 82.3 (95% CI 82.2 to 82.5) years among high-income owners, but life expectancy is one of the 13 inputs, so part of that spread is built in. A fit that leaves out life expectancy, the adult income rank, diabetes and smoking still separates its groups by 7.9 (95% CI 7.5 to 8.3) years of life expectancy, and it agrees with the typology at 0.55 on a scale where one means identical partitions and zero means chance. The names are this book's labels for groups of tracts that look alike on the inputs. The groups are not sharply separated: the silhouette of the typology, a score of how well tracts fit their own type against the nearest other, is 0.15, against 0.10 for simulated data that contain no clusters. The overlap suggests that the types are regions of a continuum, and this chapter builds the types, maps them by county, compares them on eight measures and draws the overlap.

Three other chapters hold related material. The split of life-expectancy variance among states, counties and tracts, and a principal-components analysis of nine measures, are in Chapter 10.1, Gradients, slopes and variance. The parts of a deprivation index are in Chapter 10.2, The anatomy of a deprivation index. A second clustering, built from deprivation measures with outcomes left out, is in Chapter 10.4, Families of places. The types here differ from those families because life expectancy, mobility, diabetes and smoking are inputs to the types, so the types describe bundles of conditions and outcomes and do not test whether either predicts the other.

How the types were built

The typology starts from the 85,396 tracts of the 2020 census. Puerto Rico's 981 tracts have no life expectancy or PLACES estimate and are out. A further 5,451 tracts lack at least one input, 1,209 are institutional tracts with a large share of residents in group quarters such as prisons or dormitories, where life expectancy reflects the institution, and 10 have 200 or fewer residents. What remains is 77,745 tracts, which hold 93.8% of the residents of the states and the District of Columbia (Figure 1). Utah and the District of Columbia are included.

Figure 1

The typology covers 77,745 of the 85,396 census tracts

Census tracts at each step from the 2020 census to the tracts with every input, counted at the first step that removes them

Step	Tracts
Tracts of the 2020 census	85,396
Puerto Rico (no life expectancy or PLACES estimates)	981
Other tracts missing at least one input or the population	5,451
Institutional tracts (25 percent or more of residents in group quarters)	1,209
Tracts with 200 or fewer residents	10
Tracts in the typology	77,745

Show data

Data for The typology covers 77,745 of the 85,396 census tracts
step	tracts
Tracts of the 2020 census	85396
Puerto Rico (no life expectancy or PLACES estimates)	981
Other tracts missing at least one input or the population	5451
Institutional tracts (25 percent or more of residents in group quarters)	1209
Tracts with 200 or fewer residents	10
Tracts in the typology	77745

Most tracts enter the typology. The largest exclusion is the 5,451 tracts outside Puerto Rico that lack at least one input, and the rows are counts of tracts, not of residents.Note: Puerto Rico's tracts all lack life expectancy and PLACES estimates, so none can enter. Institutional tracts have life expectancy estimates that reflect group quarters such as prisons or dormitories. Tracts with 200 or fewer residents are left out.Source: 2020 census tracts with ACS 2019-2023, USALEEP 2010-2015, Opportunity Atlas and PLACES inputs; counts of tracts, full count

The 13 inputs are the log of median household income, the poverty rate, the share of adults with a college degree, the Black share and the Hispanic share of residents, the share aged 65 and over, the single-parent share of households, the homeownership rate, life expectancy at birth, the adult income rank of children raised in low-income families, the diabetes rate, the smoking rate and the log of population density. Life expectancy is the USALEEP estimate for deaths in 2010 to 2015, the income rank is the Opportunity Atlas rank for children born in 1978 to 1983, the diabetes and smoking rates are modeled PLACES estimates, and the rest come from the American Community Survey (ACS) for 2019 to 2023 and from tract land area. A tract's type therefore blends measures from several decades. Life expectancy, the income rank, diabetes and smoking are outcomes of the kind the chapter later compares, which is why the comparison separates measures that went in from measures that did not.

Each input is standardized to mean zero and standard deviation one across tracts, and each tract counts once. K-means then places eight centers, assigns every tract to the nearest and repeats until the assignments settle, from many random starts. Eight is a choice that the data do not select, since the mean silhouette is highest for a fit with 5 types. Refits on bootstrap resamples of tracts agree with the typology at a median of 0.95 on that scale, and a refit that weights each tract by its population agrees at 0.81, so the partition is one of several that the data support (Chapter 10.4, Families of places sets out the method and its stability checks for a related clustering).

Types are numbered by the population-weighted mean of the first principal component of the inputs, a combined advantage score, from most to least advantaged, so the numbering does not depend on the starting seed. The names are the book's labels for the profiles in Figure 2, written into the build with a check that each profile still fits its name. A name describes what the inputs measure, such as density, income, poverty, education, tenure and age. The Black and Hispanic shares are inputs, and they help separate the last three types from one another, which the names do not show. The names label tracts and not the people who live in them. The chapter's label echoes the Eight Americas of Murray and colleagues, who divided the population by race and county to compare mortality;[1] these types group tracts by a profile and share only the count of groups.

The eight types

Figure 2 profiles each type. High-income owners hold 57.9 million residents, with tract median household incomes averaging 147 thousand dollars, a college share of 59.9% and a smoking rate of 9.3%. The high-poverty type holds 14.3 million, with a poverty rate of 34.0%, a Black share of 60.8%, a diabetes rate of 19.6% and a smoking rate of 24.4%. Middle-income owners are the most populous type, with 74.1 million residents and a life expectancy of 79.2 (95% CI 79.0 to 79.3) years, and they sit closest to the middle of the advantage score.

Figure 2

Life expectancy, one of the 13 inputs, averages 82.3 years among high-income owners and 72.3 years in the high-poverty type, with the other six types between

Population-weighted mean of tract values by neighborhood type, types numbered from most to least advantaged on the first principal component

Type	Residents (millions)	Tracts	Life expectancy (years)	Adult income rank, children from low-income families (percentile)	Household income (thousands of dollars)	Poverty (%)	College degree (%)	Black (%)	Hispanic (%)	Density (residents per square mile)	Smoking (%)	ADI national rank
High-income owners	57.9	12,834	82.3	50.5	147	4.4	59.9	4.1	9.9	2,102.4	9.3	19.3
Dense, college-educated renters	25.9	6,559	80.3	44.2	95	11.7	56.1	9.9	18.0	8,120.9	10.4	26.3
Low-density older owners	36.6	10,538	80.2	45.9	83	8.0	30.8	2.2	6.6	124.6	14.4	51.0
Middle-income owners	74.1	16,095	79.2	43.8	92	8.2	33.2	9.4	18.9	2,047.2	13.5	42.6
Low-density, lower-income owners	37.4	10,850	75.9	39.9	60	15.9	17.4	7.2	7.5	112.9	20.0	76.0
Dense, lower-income	31.6	7,273	78.1	40.5	63	20.0	16.2	7.3	70.6	5,953.0	16.3	51.2
Mid-density, lower-income	33.9	8,829	75.7	38.0	58	19.0	24.3	30.3	16.3	3,643.8	17.9	64.6
High-poverty	14.3	4,767	72.3	32.4	37	34.0	14.1	60.8	13.7	2,773.4	24.4	83.6

Show data

Data for Life expectancy, one of the 13 inputs, averages 82.3 years among high-income owners and 72.3 years in the high-poverty type, with the other six types between
adi	blk	col	density	his	income	kfr	le	pov	residents	smk	tracts	type
19.29	4.08	59.86	2102.4	9.87	146.8	50.54	82.33	4.40	57.95	9.26	12834	High-income owners
26.28	9.95	56.13	8120.9	18.03	94.63	44.24	80.32	11.72	25.85	10.43	6559	Dense, college-educated renters
50.97	2.17	30.79	124.6	6.57	83.12	45.86	80.19	8.01	36.62	14.37	10538	Low-density older owners
42.60	9.39	33.20	2047.2	18.85	92.47	43.75	79.16	8.21	74.09	13.53	16095	Middle-income owners
75.95	7.20	17.40	112.9	7.50	59.57	39.87	75.90	15.95	37.37	20.02	10850	Low-density, lower-income owners
51.24	7.31	16.24	5953.0	70.65	62.57	40.49	78.09	19.97	31.62	16.33	7273	Dense, lower-income
64.58	30.31	24.31	3643.8	16.30	58.23	37.95	75.67	18.96	33.93	17.94	8829	Mid-density, lower-income
83.58	60.81	14.07	2773.4	13.65	37.46	32.39	72.34	33.98	14.31	24.44	4767	High-poverty

Read across a row for one type and down a column for one measure. Life expectancy is highest in the first type and lowest in the last, but it is an input to the types, so its spread across them is partly built in. The order of the types is not an order of life expectancy, since the dense, lower-income type lives longer on average than the low-density, lower-income owners. Each cell is the mean of tract values weighted by residents, and income is the mean of tract median household incomes.Note: The type names are the book's labels for groups of tracts that resemble one another on the inputs, not categories in the data. They describe density, income, poverty, education, tenure and age. The Black and Hispanic shares are inputs and help separate the last three types, which the names do not carry.Note: Intervals for the measures in the outcomes figure are bootstrap intervals over counties; these cells describe the tracts in each type.Note: Life expectancy is for deaths in 2010 to 2015 on tracts of 2020 (area-weighted), mobility is for children born 1978 to 1983, the ACS columns cover 2019 to 2023, smoking is a modeled PLACES estimate, and the ADI is the 2023 national rank (1 least and 100 most deprived). Density is the weighted geometric mean of tract residents per square mile of land, so a few very dense tracts do not set the value.Note: Black residents are Black alone and may be Hispanic; Hispanic residents are of any race.Source: ACS 2019-2023, USALEEP life expectancy 2010-2015, Opportunity Atlas (children born 1978 to 1983), PLACES 2024 release and ADI 2023; tract values weighted by ACS population; population-weighted

Two types have the lowest density. Low-density older owners hold 36.6 million residents, of whom 26.8% are aged 65 and over, in tracts where the homeownership rate is 83.1%, and their life expectancy is 80.2 (95% CI 80.1 to 80.3) years. Low-density, lower-income owners hold 37.4 million residents, with a smoking rate of 20.0%, a poverty rate of 15.9%, a mean Area Deprivation Index national rank of 76.0 (a higher rank is more deprived) and a life expectancy of 75.9 (95% CI 75.8 to 76.0) years, which is 4.3 (95% CI 4.2 to 4.4) years below the first. A single low-density label would join the two, and the inputs separate them.

Dense, college-educated renters hold 25.9 million residents at a typical density of 8,121 (95% CI 6,005 to 10,567) residents per square mile, the highest of the eight, with the lowest homeownership (36.2%) and a college share of 56.1%, second only to high-income owners. Their life expectancy, 80.3 (95% CI 80.0 to 80.6) years, is about that of low-density older owners. The dense, lower-income type holds 31.6 million residents with a poverty rate of 20.0%, a single-parent share of 15.0% and a life expectancy of 78.1 (95% CI 77.6 to 78.5) years, and the mid-density, lower-income type holds 33.9 million residents with a life expectancy of 75.7 (95% CI 75.5 to 75.9) years. The Hispanic share averages 70.6% in the dense, lower-income type and the Black share averages 30.3% in the mid-density, lower-income type. Both shares are inputs, and the names do not carry them.

Where the types are most common

Coloring each county by its most common type draws a map dominated by two types. Low-density, lower-income owners are the most common type in 1,274 counties and low-density older owners in 970, together most of the 3,087 counties that have a type (Figure 3). The first is the most common type in most counties of Kentucky, Tennessee, West Virginia, Arkansas and Alabama, and the second in most counties of North Dakota, South Dakota, Nebraska, Iowa, Minnesota and Vermont. High-income owners are the most common type in 152 counties, among them most counties of Connecticut and New Jersey. The high-poverty type is the most common type in 85 counties, mostly in Mississippi, Alabama, Georgia, Louisiana and Arkansas.

Figure 3

Low-density, lower-income owners are the most common type in 1,274 counties and low-density older owners in 970, most of the 3,087 counties with a type

Neighborhood type that holds the most residents in each county, tracts typed by k-means on 13 measures, ACS 2019-2023 population

Loading county values

Most common neighborhood type

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.

Two types are the most common type in most counties: low-density, lower-income owners across the South and Appalachia, and low-density older owners across the northern Plains, the upper Midwest, Montana and Wyoming and northern New England. A county's color is the type of its largest group of residents, and the median county's modal type covers 64.4% of its typed residents, so the color hides the mix inside the county. Gray counties have no type, and the tooltip says why.Note: A county gets a type when its typed tracts hold at least 50% of its residents. 32 counties have no typed tract (a tract that is not institutional, has every input and has more than the floor of residents) and 24 have typed tracts that hold a smaller share, so they are gray with the reason in the tooltip. A county with a tie in residents would take the lower-numbered type; none has a tie. 472 counties with a type rest on fewer than 3 typed tracts.Note: Connecticut is drawn as its eight legacy counties, with tracts assigned through the exact tract crosswalk. Puerto Rico has no typed tract and is not drawn.Note: Colors mark the eight types and have no order.Source: k-means types of tracts (ACS 2019-2023, USALEEP 2010-2015, Opportunity Atlas, PLACES 2024 release) summed to counties by ACS population; population-weighted

A county's color is the type of its largest group of residents, and a county's residents are mixed. In the median county the modal type holds 64.4% of the typed residents, and across all counties with a type, 46.6% of typed residents live in a tract of their own county's modal type, which leaves more residents in a tract of another type. Among the 163 counties with at least 100 typed tracts, a median of 3 types each hold at least 10% of the typed residents.

Counties are typed from their tracts. A county gets a type when its typed tracts hold at least 50% of its residents. 32 counties have no tract that is not institutional, has every input and has more than 200 residents, and 24 have typed tracts that hold a smaller share, so they are gray and the tooltip gives the reason. Residents are counted from the ACS, and no county has a tie, but 472 of the counties with a type rest on fewer than 3 typed tracts. Dropping those counties would remove them from the map and leave every other color unchanged. Connecticut is drawn as its eight legacy counties, and Puerto Rico has no typed tract and is not drawn.

Outcomes by type

The comparison separates measures that went into the types from measures that did not (Figure 4). Life expectancy, the adult income rank of children raised in low-income families, diabetes and smoking are inputs, so their order across types is partly built in. Teen births and incarceration among the same Opportunity Atlas cohorts, frequent mental distress and obesity among adults are not inputs. The high-poverty type has the least favorable mean on all eight measures, and the first or the second type has the most favorable.

Figure 4

The high-poverty type has the least favorable mean on all eight measures, four that went into the types and four that did not

Population-weighted mean of tract values by neighborhood type with 95% intervals from a bootstrap over counties, each panel in its own units

Life expectancy and adult income rank are lowest and diabetes and smoking highest in the high-poverty type

Four measures that are inputs to the typology, population-weighted type means with 95% intervals

Life expectancy at birth (years)

Adult income rank of children from low-income families (percentile)

Diabetes among adults (%)

Smoking among adults (%)

Population-weighted mean of tract values (units in each panel title)

Life expectancy and adult income rank are lowest and diabetes and smoking highest in the high-poverty type. The order across the other types differs by measure.

Show data

Data for The high-poverty type has the least favorable mean on all eight measures, four that went into the types and four that did not, Used to form the types
conf_high	conf_low	estimate	label	n	panel	type
82.48	82.17	82.33	High-income owners	12,834	Life expectancy at birth (years)	High-income owners
80.60	80.00	80.32	Dense, college-educated renters	6,559	Life expectancy at birth (years)	Dense, college-educated renters
80.28	80.09	80.19	Low-density older owners	10,538	Life expectancy at birth (years)	Low-density older owners
79.28	79.05	79.16	Middle-income owners	16,095	Life expectancy at birth (years)	Middle-income owners
75.97	75.83	75.90	Low-density, lower-income owners	10,850	Life expectancy at birth (years)	Low-density, lower-income owners
78.50	77.64	78.09	Dense, lower-income	7,273	Life expectancy at birth (years)	Dense, lower-income
75.86	75.50	75.67	Mid-density, lower-income	8,829	Life expectancy at birth (years)	Mid-density, lower-income
72.53	72.15	72.34	High-poverty	4,767	Life expectancy at birth (years)	High-poverty
50.90	50.17	50.54	High-income owners	12,834	Adult income rank of children from low-income families (percentile)	High-income owners
44.82	43.60	44.24	Dense, college-educated renters	6,559	Adult income rank of children from low-income families (percentile)	Dense, college-educated renters
46.08	45.64	45.86	Low-density older owners	10,538	Adult income rank of children from low-income families (percentile)	Low-density older owners
44.00	43.49	43.75	Middle-income owners	16,095	Adult income rank of children from low-income families (percentile)	Middle-income owners
40.03	39.71	39.87	Low-density, lower-income owners	10,850	Adult income rank of children from low-income families (percentile)	Low-density, lower-income owners
40.99	39.94	40.49	Dense, lower-income	7,273	Adult income rank of children from low-income families (percentile)	Dense, lower-income
38.20	37.69	37.95	Mid-density, lower-income	8,829	Adult income rank of children from low-income families (percentile)	Mid-density, lower-income
32.72	32.07	32.39	High-poverty	4,767	Adult income rank of children from low-income families (percentile)	High-poverty
9.66	9.34	9.50	High-income owners	12,834	Diabetes among adults (%)	High-income owners
8.72	8.17	8.45	Dense, college-educated renters	6,559	Diabetes among adults (%)	Dense, college-educated renters
12.44	12.23	12.34	Low-density older owners	10,538	Diabetes among adults (%)	Low-density older owners
11.21	10.94	11.07	Middle-income owners	16,095	Diabetes among adults (%)	Middle-income owners
14.91	14.75	14.83	Low-density, lower-income owners	10,850	Diabetes among adults (%)	Low-density, lower-income owners
15.39	14.43	14.89	Dense, lower-income	7,273	Diabetes among adults (%)	Dense, lower-income
13.57	13.28	13.43	Mid-density, lower-income	8,829	Diabetes among adults (%)	Mid-density, lower-income
19.89	19.31	19.59	High-poverty	4,767	Diabetes among adults (%)	High-poverty
9.46	9.09	9.26	High-income owners	12,834	Smoking among adults (%)	High-income owners
10.71	10.18	10.43	Dense, college-educated renters	6,559	Smoking among adults (%)	Dense, college-educated renters
14.53	14.19	14.37	Low-density older owners	10,538	Smoking among adults (%)	Low-density older owners
13.71	13.36	13.53	Middle-income owners	16,095	Smoking among adults (%)	Middle-income owners
20.13	19.91	20.02	Low-density, lower-income owners	10,850	Smoking among adults (%)	Low-density, lower-income owners
16.94	15.83	16.33	Dense, lower-income	7,273	Smoking among adults (%)	Dense, lower-income
18.19	17.69	17.94	Mid-density, lower-income	8,829	Smoking among adults (%)	Mid-density, lower-income
24.84	24.04	24.44	High-poverty	4,767	Smoking among adults (%)	High-poverty

In the first tab, four measures that were inputs to the types; in the second, four that were not. The high-poverty type has the least favorable value in every panel, and the first or the second type has the most favorable. Dots are type means and bars are 95% intervals, which are short next to the gaps between the extreme types. Rows follow the type numbering, not the size of the value.Note: The first tab shows measures that were inputs to the typology, so their order across types is partly built in. The second tab shows measures that were not inputs, but PLACES estimates are modeled partly from ACS covariates that are inputs, and the teen-birth and incarceration shares come from the same Opportunity Atlas file as an input.Note: Opportunity Atlas measures are for children born 1978 to 1983 raised in families at the 25th income percentile, interpolated from 2010 to 2020 tracts by land area, and tract population stands in for the number of children. Life expectancy is for deaths in 2010 to 2015. PLACES values are modeled crude prevalence for adults, with BRFSS 2022 data.Note: Intervals show how the means would vary with the sample of counties, given the types; they do not include uncertainty about the typology itself (a bootstrap over counties; replicates and seed in the methods note).Source: ACS 2019-2023, USALEEP life expectancy 2010-2015, Opportunity Atlas (children born 1978 to 1983), PLACES 2024 release and ADI 2023; tract values weighted by ACS population; population-weighted, modeled estimates for the PLACES measures

Among the held-out measures, the share of women raised in low-income families who had a teen birth is 13.2% (95% CI 12.6% to 14.0%) among high-income owners and 44.8% (95% CI 44.0% to 45.7%) in the high-poverty type. The share of children from those families who were incarcerated on Census Day 2010 is 1.1% (95% CI 1.1% to 1.2%) in the first type and 4.6% (95% CI 4.4% to 4.8%) in the last. Both measures come from the same Opportunity Atlas file as one of the inputs, and PLACES estimates are modeled partly from ACS covariates that are inputs, so the agreement between the types and these measures is partly mechanical.

Between the two ends the types do not line up in one order. The dense, lower-income type, numbered sixth, has a life expectancy 2.2 (95% CI 1.7 to 2.6) years above the low-density, lower-income owners, numbered fifth, although its poverty rate (20.0%) and single-parent share (15.0%) are higher than theirs (15.9% and 8.3%). The intervals for dense, college-educated renters and low-density older owners overlap on life expectancy, so those two cannot be ranked. The intervals are narrow next to most gaps between types, since each type holds thousands of tracts, and they describe how the means would vary with the sample of counties given the types, not the choice of types.

The label foretells part of a tract's outcome and leaves the rest. The types account for 49% (95% CI 47% to 51%) of the variance of tract life expectancy, for 46% (95% CI 44% to 48%) of the variance of the teen-birth share, which is held out, for 60% (95% CI 58% to 62%) of the variance of frequent mental distress, a modeled PLACES measure whose share is partly mechanical, and for 15% (95% CI 13% to 17%) of the variance of the incarceration share. Tracts of the same type still differ, and they differ most on incarceration.

A fit that leaves out the four outcome inputs is a fairer test of how far the other inputs sort outcomes. It uses the same seed and starts on income, poverty, education, race and ethnicity, age, family structure, tenure and density. Its groups span 7.9 (95% CI 7.5 to 8.3) years of life expectancy, against 10.0 (95% CI 9.8 to 10.2) years between the extreme types of the typology, and they account for 33% (95% CI 31% to 35%) of the variance of tract life expectancy, against 49%, and for 36% (95% CI 34% to 39%) of the variance of teen births, against 46%. The inputs that are not outcomes sort life expectancy and teen births, less sharply than the typology does.

The types are not sharply separated

A silhouette of 0.15 is low, and two comparisons give it a scale. K-means finds eight groups in data that have none: for simulated data with the correlations of the inputs, the silhouette is 0.10 and 49.9% of the variance lies between the groups. The tract data score 0.15 and 56.1%, with 43.9% of the variance within the types, a step above what a cloud without clusters gives. The benchmark is a guide and not a proof, because the simulated margins are symmetric and tract incomes, densities and rates are not. The second comparison is the tracts themselves: 10.2% of a sample of 20,000 tracts have a negative silhouette, which means that on average they sit closer to the tracts of another type than to those of their own.

Figure 5 shows the overlap on the first two principal components, which capture 60.2% of the variance in the inputs. The first, 43.2%, weights income, college share, life expectancy and the adult income rank positively and poverty, smoking, diabetes, single-parent share and Black share negatively, and it is the advantage score that numbers the types. The second, 17.0%, weights density and Hispanic share positively and homeownership and the share of older adults negatively. The plotted sample holds 6,000 tracts, and the ranges from the fifth to the ninety-fifth percentile of tracts overlap for every pair of neighboring types along the advantage score. The overlap suggests that the types are bands of a continuum, and a tract near a border could have been assigned to the neighbor by a small change in the inputs. The advantage score of this figure is a different object from the nine-measure analysis of Chapter 10.1, Gradients, slopes and variance, which uses other measures.

Figure 5

Neighboring types overlap along the first component, and 10.2% of 20,000 sampled tracts sit nearer another type than their own

A random sample of tracts on the first two principal components of the 13 standardized inputs, colored by type, with each type's centroid labeled

Loading data

Look for colors that shade into one another along the horizontal axis, which is an advantage score: neighboring types overlap, which suggests a continuum. The first component captures 43.2% of the variance in the inputs and the second 17.0%. Each dot is a tract in the plotted sample of 6,000, and each labeled dot is a type's population-weighted centroid. The share of tracts nearer another type comes from a separate sample of 20,000 tracts.Note: The dots are a random sample of 6,000 of the typed tracts drawn with a fixed seed (replicates and seed in the methods note); the components, the types and the centroids use every typed tract. Components are in standardized units, and their signs are set so that life expectancy and density rise with the first and second component.Note: The first component weights income, college share, life expectancy and adult income rank positively and poverty, smoking, diabetes, single-parent share and Black share negatively. The second component weights density and Hispanic share positively and homeownership and the share of older adults negatively.Source: k-means types and principal components of the standardized tract inputs (ACS 2019-2023, USALEEP 2010-2015, Opportunity Atlas, PLACES 2024 release); unweighted sample of tracts, centroids population-weighted

Causal evidenceMovers design

Whether growing up in a different place changes a child's adult income

The comparisons above describe tracts and cannot show that a tract's type changes anyone's life. Chetty and Hendren used federal tax records for children born in 1980 to 1988 whose families moved once between commuting zones, and compared children who moved at different ages, including siblings in the same family. The identifying assumption is that the selection that sends a family to a better or a worse place does not vary with the child's age at the move. They test it with sibling comparisons, with controls for changes in parental income and marital status, and with moves caused by displacement shocks. Children's adult income ranks move toward the ranks of children of permanent residents of the destination who had parents at the same income, by 0.04 percentile for each year of childhood spent there and each percentile by which the destination's ranks are higher.[2]

The design applies to commuting zones and counties, and its benchmark is permanent residents of the destination at the same parent income. It does not test the eight types, and it does not say that moving a child from one type to another changes an outcome by the gap between the types' means.

What these data cannot show

The types describe tracts, and every comparison is ecological: the life expectancy of the high-poverty type is an average over tracts and says little about any one resident of one (ecological inference). The names are the book's labels for the profiles. A name such as high-poverty summarizes one input of a profile that also has a high Black share and a short life expectancy, and it does not describe the people who live there.

The typology summarizes inputs that include outcomes. Four of its inputs are outcomes, so the order of the types on those measures is built in, and the held-out measures share data sources or model inputs with the inputs. The types show how tightly conditions and outcomes move together across tracts and do not show that a tract's conditions produced any outcome. The measures come from different periods, with life expectancy for deaths in 2010 to 2015, mobility for children born in 1978 to 1983, the ACS for 2019 to 2023 and PLACES built on 2022 survey data, so a type does not describe one moment. The USALEEP estimates and the Opportunity Atlas ranks for 2020 tracts are interpolated from 2010 tracts by land area, not by population. The USALEEP estimates suit maps and gradients, the ACS file carries no margins of error, and PLACES values are modeled and carry no intervals at the tract level.

The intervals in this chapter condition on the types. Tracts that lack an input are not typed. The 6,670 untyped tracts hold 20.6 million residents, and they differ from the typed tracts: their poverty rate averages 18.6% against 12.2%, their Black share 16.0% against 11.8%, and 72% of them lack a life expectancy estimate. The residents of typed tracts, 93.8% of all residents, are therefore not a random sample of the country. The county map shows one type per county and hides the mix, and Puerto Rico is outside the typology and the map.

Data sources

American Community SurveyACS 5-year estimates 2019-2023 for 2020 tracts: income, poverty, education, race and ethnicity, age, households, tenure, populationYears: 2019-2023 (5-year) · Weights: ACS 2019-2023 tract population · License: Public domain (U.S. Census Bureau)
USALEEP life expectancyNCHS small-area life expectancy at birth, deaths in 2010 to 2015, 2020 tracts (area-weighted), institutional tracts left outYears: 2010-2015 · Weights: ACS 2019-2023 tract population · License: Public domain (NCHS, CDC)
Opportunity AtlasAdult income rank, teen births and incarceration of children born 1978 to 1983 raised in families at the 25th income percentile, 2020 tracts (interpolated by land area)Years: 1978-1983 birth cohorts (2018 release) · Weights: ACS 2019-2023 tract population · License: Public use (Opportunity Insights and the U.S. Census Bureau)
CDC PLACESCDC PLACES 2024 release (BRFSS 2022): diabetes, smoking, frequent mental distress and obesity among adults, modeled crude prevalenceYears: 2024 release (tracts); 2025 release (ZCTAs, 2023 data year) · Weights: ACS 2019-2023 tract population · License: Public domain (CDC, U.S. Department of Health and Human Services)
Area Deprivation IndexArea Deprivation Index 2023 national rank, a profile column and not an inputYears: 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
Census TIGER/Line tracts2020 tract land area for population densityYears: 2020 · Weights: none (land area) · License: Public domain (U.S. Census Bureau)
Published estimates from cited studiesThe exposure effect quoted in the causal box (Chetty and Hendren 2018)Years: as reported in each study · Weights: as reported in the study · License: Quoted from published research for commentary

All data sources

Methods

The typology groups tracts of the 50 states and DC by 13 inputs that come from different periods: ACS 2019-2023 measures, life expectancy for deaths in 2010 to 2015, the Opportunity Atlas adult income rank of children born 1978 to 1983, and modeled PLACES estimates of diabetes and smoking. Life expectancy, the adult income rank, diabetes and smoking are among the inputs and are also reported by type, so a fit without them is described below. The clustering counts every tract once while type means are weighted by tract population, and the intervals come from a bootstrap over counties that conditions on the types and does not carry uncertainty about the typology.

Which tracts are in the typology

The typology covers the tracts of the 50 states and DC that have every one of the 13 inputs and a population, are not institutional (25 percent or more of residents in group quarters) and have more than 200 residents. Puerto Rico has no life expectancy or PLACES estimates and is out.

Figure explained here: Figure 1

What the inputs measure and when

The inputs are log median household income, the poverty rate, the college share of adults, the Black share (Black alone, Hispanic residents included), the Hispanic share, the share aged 65 and over, the single-parent share of households and the homeownership rate (ACS 2019-2023); life expectancy at birth (USALEEP, deaths 2010 to 2015); the adult income rank of children from low-income families (Opportunity Atlas, children born 1978 to 1983, interpolated from 2010 to 2020 tracts by land area); diabetes and smoking among adults (PLACES, modeled); and the log of residents per square mile of land (ACS population, 2020 TIGER land area).

Figure explained here: Figure 2

How the types are numbered and named

Types are numbered by the population-weighted mean of the first principal component of the inputs, largest first, so the numbers do not depend on the initialization.

The names are the book's labels and are written in types_data.py, with an assertion that each profile fits its name. They describe density, income, poverty, education, tenure and age and name no race or ethnicity.

Figures explained here: Figure 2 and Figure 5

How a county gets a type

A county's type is the type that holds the most typed residents; a county has one when its typed tracts hold at least 50% of its residents. The count of types per county uses counties with at least 100 typed tracts and counts a type when it holds at least 10% of the typed residents.

Figure explained here: Figure 3

Weights and intervals

Type means are weighted by ACS 2019-2023 tract population.

Intervals are 95% percentile intervals from 2,000 draws of whole counties with replacement (seed 12345), with Connecticut tracts assigned to the eight legacy counties; differences between two types and the shares of variance between types use the same draws. The intervals condition on the types and do not carry uncertainty about the typology.

The spread of life expectancy is the highest minus the lowest population-weighted group mean, and the R squared is the variance between groups over the total, both with the county bootstrap above.

Figure explained here: Figure 4

How the clustering was fitted and checked

The inputs are standardized with each tract counted once and clustered by k-means (scikit-learn, 8 groups, 20 starts, seed 12345, one thread). The fit weights every tract equally; a fit weighted by tract population gives a similar partition (agreement 0.81).

Refits on 8 bootstrap resamples of tracts, assigned back to every tract by nearest centroid, agree with the typology at a median of 0.95.

The same k-means (8 groups, 20 starts, seed 12345) on the inputs other than life expectancy, the adult income rank, diabetes and smoking agrees with the typology at 0.55.

The mean silhouette is computed on a random sample of 20,000 tracts (seed 12345) for the fit with 8 groups (20 starts) and for fits with 5 to 10 groups (10 starts each; the mean silhouette is highest at 5 groups). A benchmark fit to simulated data with no clusters, a multivariate normal with the correlations of the inputs and one draw per tract, gives 0.10; the simulated margins are symmetric and tract margins are not, so the benchmark is a guide.

The principal-components figure shows a random sample of 6,000 typed tracts (seed 12345).

Figure explained here: Figure 5

References

Murray, Christopher J. L.; Kulkarni, Sandeep C.; Michaud, Catherine; Tomijima, Niels; Bulzacchelli, Maria T.; Iandiorio, Terrell J.; Ezzati, Majid (2006). Eight Americas: Investigating Mortality Disparities across Races, Counties, and Race-Counties in the United States. PLoS Medicine 3(9): e260. doi:10.1371/journal.pmed.0030260
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

How to cite

@misc{singh2026eighttypesofneighborhood,
  author = {Singh, Vishal},
  title = {Eight types of neighborhood},
  year = {2026},
  url = {https://vishalsingh.org/zipcode/eight-types-of-neighborhood},
  note = {In ZIP Code Destiny. Published 9 October 2026.}
}