Chapter 1.1 · Weighing the ingredients

What the deprivation index measures

The Area Deprivation Index orders 75,717 census tracts almost in reverse order of median home value, with a rank correlation of −0.988, and the home value term alone accounts for 74.3% of the variance of the score its formula produces.

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

In one San Francisco census tract, 35.9% of residents are below the poverty line, and the Area Deprivation Index (ADI) ranks it at national percentile 1, among the least deprived in the nation. A version of the same formula that puts every measure on a common scale ranks the tract at percentile 99.7, near the most deprived end. The tract's median home value is at the top code of the American Community Survey, the cap the survey puts on the dollar values it reports. The tract illustrates what the index's formula does, and its rank is not an estimate of how deprived the tract is.

The index orders tracts almost as home prices do

Across 75,717 census tracts of the states and the District of Columbia, the ADI has a population-weighted rank correlation of −0.988 with median home value. A rank correlation compares the positions of tracts in two rankings. The sign is negative because the index rises as home values fall, and its size means that the index orders tracts almost as home prices do.

The Neighborhood Atlas at the University of Wisconsin publishes the index. It applies a recipe that Gopal Singh built from 1990 census tracts for a study of mortality in the United States.[1, 2] The recipe multiplies each census measure of education, work, income and housing by a fixed coefficient and adds the products. Median family income, home value, gross rent and the monthly mortgage cost are dollar amounts. The other measures are proportions or a log ratio, and the Atlas enters proportions as fractions of one. The coefficients are close in size, so the unit of a measure, more than its coefficient, sets how much the measure moves the sum.

Variance measures how widely scores spread around their average, in squared units. A dollar amount in the hundreds of thousands varies over a wider range than a proportion, so it contributes more of the variance of a sum. In a replica of the score built from the published formula, the home value term alone accounts for 74.3% of the variance across tracts. The home value and family income terms are correlated, and Figure 1 shares each covariance between two terms equally between them. On that accounting, home value is allocated 85.4% (95% CI 83.1% to 87.0%) of the variance in raw units.

Figure 1

In raw units home value accounts for 74.3% of the variance of the score on its own and 85.4% with its covariance with income shared out, against 5.3% after standardization

Share of the population-weighted variance of the replica score allocated to each measure, same coefficients, measures in raw units and in standard deviation units, 75,717 tracts

In raw units home value and family income carry 99.2% of the variance of the score between them

Share of the variance of the score in raw units allocated to each measure, percent

Measure

Home value's bar is the longest and family income's the second; every other bar is nearly zero.

Show data

Data for In raw units home value accounts for 74.3% of the variance of the score on its own and 85.4% with its covariance with income shared out, against 5.3% after standardization, Raw units
conf_high	conf_low	measure	n	share
86.99	83.08	Home value	75,717	85.40
16.52	12.64	Family income	75,717	14.23
0.245	0.213	Mortgage cost	75,717	0.226
0.164	0.143	Gross rent	75,717	0.152
0.000	0.000	Income disparity	75,717	0.000
0.000	0.000	White-collar jobs	75,717	0.000
0.000	0.000	Below 150% of poverty	75,717	0.000
0.000	0.000	Poverty	75,717	0.000
0.000	0.000	Single parents	75,717	0.000
0.000	0.000	Under 9 years of school	75,717	0.000
0.000	0.000	No plumbing	75,717	0.000
0.000	0.000	Unemployment	75,717	0.000
0.000	0.000	Homeownership	75,717	0.000
0.000	0.000	Crowding	75,717	0.000
0.000	0.000	No vehicle	75,717	0.000

In raw units the home value and family income bars are the only ones long enough to see, and after standardization every measure has a bar and home value's is short. Each bar is a measure's allocated share of the population-weighted variance of the replica score, with whiskers for 95% intervals; the tabs share one order and one axis. Values in Table 3.

Note: Raw units: dollars for the four dollar measures, proportions between zero and one for the rest, and the log ratio for income disparity, as the Atlas applies Singh's coefficients. Standardized: each measure minus its population-weighted mean over its population-weighted standard deviation.Note: The two measures that are not rebuilt (high school diploma and internet access) are proportions, so in raw units they would add little to the variance of the score.Source: ACS 2019-2023 five-year tract estimates (staged ACS file and Census API pull of housing measures); Singh (2003) coefficients; population-weighted (ACS tract population)

A standardized score subtracts each measure's mean and divides by its standard deviation, the typical distance from that mean, so that every measure enters in the same units. With the same coefficients, home value is then allocated 5.3% (95% CI 5.0% to 5.7%) of the variance. The same coefficients thus give weights that depend on the scale of the measures. Petterson showed this across the block groups of the Atlas, and Hannan and colleagues found that in New York State the ADI mainly reflects home value.[3, 4] The shares here describe a population-weighted frame of tracts, a different population from either study.

How the replica was built

The replica uses American Community Survey (ACS) five-year estimates for 2019 to 2023 for 2020 census tracts. The dataset registry records the ADI's 2023 release, used here, as built on the same period. It rebuilds every measure except the share of adults with at least a high school diploma and the share of households without internet. Some rebuilt measures are substitutes, such as the poverty rate of persons where Singh counted families. The Atlas ranks block groups and imputes missing values from the surrounding geography. The replica works with tracts, keeps only those with every rebuilt measure and imputes nothing. Ranked among those tracts, the replica's score has a rank correlation of +0.991 (95% CI 0.990 to 0.992) with the published tract rank. The match shows that the published formula reproduces the Atlas's ranks, and it does not show that the Atlas's code is the same. Each interval comes from resampling whole counties. The report gives every estimate with its interval and sample size.

The ADI reads expensive places as less deprived

The disagreement between the published ranking and the standardized one follows the cost of housing. Figure 2 groups tracts by the median home value of their county, in quartiles with equal numbers of residents. The published percentile minus the standardized percentile averages +11.1 (95% CI 10.5 to 11.6) points in the lowest-cost counties and −16.5 (95% CI −21.3 to −11.5) points in the highest-cost counties. At the same standardized percentile, tracts in the highest-cost counties have a lower published ADI than tracts in the lowest-cost counties in every tenth of the standardized ranking.

Figure 2

The published ADI percentile minus the standardized percentile averages −16.5 in the highest-cost counties and +11.1 in the lowest-cost counties

Tract percentile on the published ADI against the percentile of the standardized replica score, by the cost of the tract's county, and the tracts with the largest gaps

At the same standardized percentile, tracts in the highest-cost counties have a lower published ADI than tracts in the lowest-cost counties

Published ADI against the standardized percentile, one dot per sampled tract, colored by the median home value of the tract's county

Loading data

The line of binned means for the lowest-cost counties lies above the line for the highest-cost counties in every tenth of the standardized ranking, and the gap widens toward the most deprived tenth.

The line of binned means for the lowest-cost counties lies above the line for the highest-cost counties in every tenth of the standardized ranking, and the gap widens toward the most deprived tenth. Each dot is a sampled tract, placed by its standardized percentile and its published ADI, and each marker is a population-weighted mean of the published ADI within a tenth, with a 95% interval. The second tab lists the high-poverty tracts with the largest gaps, with the San Francisco tract in the first row. Values in Table 5.

Note: Counties are grouped by their median home value into quartiles with equal numbers of residents; the middle two quartiles are drawn together in gray. Markers are population-weighted means of the published ADI within tenths of the standardized percentile for the lowest-cost and highest-cost groups, with 95% intervals from a bootstrap over counties; cells with few tracts are left out.Note: The dots are a random sample of the frame's tracts drawn with a fixed seed (seed and share in the methods note); the means, intervals and the table use every tract.Note: A tract's ACS medians rest on the homes and renters in the tract, which are few in some tracts, and the staged file has no margins of error; a college-town tract can show a high poverty rate because of student households. Each row illustrates what the formula does and is not an estimate of its tract's deprivation.Source: Neighborhood Atlas ADI 2023; ACS 2019-2023; population-weighted (ACS tract population) binned means, modeled estimates

The gap matters most where people are poor. In the frame's tracts where 30% or more of people are below the poverty line, the published ADI places 16.1% (95% CI 11.3% to 21.4%) of residents in tracts ranked at or below the 50th national percentile. That end of the ranking is the less deprived one. The standardized replica places 0.34% (95% CI 0.15% to 0.55%) of them there. Across every high-poverty tract with a published rank, including tracts that lack some rebuilt measures, the published share is 20.6% (95% CI 14.1% to 27.9%).

One Medicare payment model, ACO REACH, adjusts payments by a score that combines the ADI with a beneficiary's status as dually eligible for Medicare and Medicaid or as receiving the low-income subsidy. For performance year 2024 the Centers for Medicare and Medicaid Services added the state ADI, which compares tracts within a state. It said the change would better identify underserved beneficiaries in areas with a high cost of living.[5] This chapter does not test how much of the dominance of home value the state ADI removes.

Two outcomes test the recipes

Two outcomes that no ranking used as an input test the recipes. One is life expectancy at birth from USALEEP, for deaths in 2010 to 2015. The other is upward mobility, the mean adult household income rank of children from low-income families raised in the tract, from the Opportunity Atlas, for children born in 1978 to 1983. Every ranking runs from least to most deprived, so every correlation with these outcomes is negative, and the values here are absolute.

For life expectancy the published ADI correlates 0.615 (95% CI 0.594 to 0.633). The standardized replica differs from it by 0.005 (95% CI −0.015 to 0.027), an interval that includes zero, so the two track life expectancy about as closely. For upward mobility the standardized replica correlates more strongly than the published ADI, by 0.100 (95% CI 0.076 to 0.123). Family income alone also correlates more strongly with mobility than the published ADI does, by 0.082 (95% CI 0.063 to 0.101), while for life expectancy the interval of its difference from the ADI includes zero. The rankings that standardize edge out family income alone for both outcomes, for life expectancy by small margins, and the published ADI does not for either.

Figure 3

The published ADI tracks life expectancy about as closely as the standardized score does but upward mobility less closely, by 0.100 in correlation

Absolute population-weighted rank correlation of eight rankings with tract life expectancy at birth and with the adult income rank of children from low-income families, 71,874 and 75,667 tracts

Life expectancy at birth
Upward mobility

Show data

Data for The published ADI tracks life expectancy about as closely as the standardized score does but upward mobility less closely, by 0.100 in correlation
conf_high	conf_low	estimate	n	outcome	ranking
0.653	0.613	0.633	71,874	Life expectancy at birth	First principal component
0.607	0.566	0.587	75,667	Upward mobility	First principal component
0.638	0.601	0.620	71,874	Life expectancy at birth	Replica, raw units
0.520	0.466	0.492	75,667	Upward mobility	Replica, raw units
0.642	0.596	0.619	71,874	Life expectancy at birth	Replica, standardized
0.603	0.560	0.582	75,667	Upward mobility	Replica, standardized
0.633	0.594	0.615	71,874	Life expectancy at birth	Published ADI
0.511	0.455	0.483	75,667	Upward mobility	Published ADI
0.630	0.587	0.609	71,874	Life expectancy at birth	Family income
0.586	0.543	0.564	75,667	Upward mobility	Family income
0.605	0.563	0.584	71,874	Life expectancy at birth	Home value
0.473	0.415	0.443	75,667	Upward mobility	Home value
0.553	0.485	0.518	71,874	Life expectancy at birth	SVI socioeconomic theme
0.571	0.519	0.545	75,667	Upward mobility	SVI socioeconomic theme
0.529	0.484	0.508	71,874	Life expectancy at birth	Poverty rate
0.522	0.486	0.504	75,667	Upward mobility	Poverty rate

The dots for life expectancy bunch together across the composite rankings and family income, and for upward mobility the published ADI, its raw replica and home value alone have the lowest point estimates. Each dot is the absolute population-weighted rank correlation between a ranking and an outcome across the frame's tracts that have that outcome, with a 95% interval. Values in Table 7.

Note: Both outcomes were modeled and neither was an input to any ranking. Each outcome uses the tracts that have it and every ranking, and institutional tracts are dropped from the life expectancy comparison. USALEEP covers deaths in 2010 to 2015 and its 2020-tract file is interpolated from 2010 tracts; the Opportunity Atlas covers children born in 1978 to 1983 and its 2020-tract file is interpolated from 2010 tracts.Note: Family income and home value enter as negative values so that a higher rank means more deprived. The SVI theme and the poverty rate are not Atlas products.Source: Neighborhood Atlas ADI 2023; ACS 2019-2023; CDC/ATSDR Social Vulnerability Index 2022; NCHS USALEEP 2010-2015; Opportunity Atlas 1978-1983 birth cohorts; population-weighted (ACS tract population) rank correlations, modeled estimates

A factor model of the measures sorts them into housing money, economic hardship, and a group of crowding and adults who left school before ninth grade. The number of groups rests on a common cutoff. In raw units the single number follows the housing-money factor. A ranking dominated by home value measures the cost of housing, a property of a market, and some uses of a deprivation index may want that. The two outcomes disagree about the recipes, and the data cannot say which recipe is right.

What this cannot tell you

A tract's rank says little about how deprived any one resident is, because the data are averages for places (ecological inference). The replica is a tract-level rebuild of a block-group index, and its ACS estimates carry sampling error that the staged file does not report. Both outcomes are older than the ACS measures and are modeled estimates placed on 2020 tracts from 2010 tracts. The frame leaves out Puerto Rico, which the Atlas ranks with the states.

Go deeper

The report: every estimate with its interval and sample size.
Tables: every figure of the report as a table.

Same question at another grain:

Deprivation index
- Chapter 1.2, What a county average leaves out Neighborhoods

Reading paths that include this chapter:

Methods first

Data sources

Area Deprivation IndexNeighborhood Atlas ADI, 2023 release (version 4.0.1), national percentile ranks of block groups averaged to 2020 tractsYears: 2023 release (version 4.0.1; built on ACS 2019-2023) · Weights: none (index); tract values are unweighted means of block-group ranks · License: Free for non-commercial use; cite the Neighborhood Atlas and Kind and Buckingham
American Community SurveyACS 2019-2023 five-year tract estimates: the staged tract file and a Census API pull of housing, occupation, income-distribution and schooling tablesYears: 2019-2023 (5-year) · Weights: ACS 2019-2023 tract population · License: Public domain (U.S. Census Bureau)
CDC/ATSDR Social Vulnerability IndexCDC/ATSDR Social Vulnerability Index 2022, tracts: the socioeconomic theme percentile and, for a sensitivity check, the shares of adults without a high school diploma and of households without internet (ACS 2018 to 2022)Years: 2022 release (ACS 2018-2022) · Weights: none (index); percentile ranks of tracts · License: Public domain (CDC/ATSDR)
USALEEP life expectancyNCHS small-area life expectancy at birth, deaths 2010 to 2015, 2020 tracts (interpolated), institutional tracts droppedYears: 2010-2015 · Weights: ACS 2019-2023 tract population · License: Public domain (NCHS, CDC)
Opportunity AtlasMean adult household income rank of children from low-income families, 1978 to 1983 birth cohorts, 2020 tracts (interpolated)Years: 1978-1983 birth cohorts (2018 release) · Weights: ACS 2019-2023 tract population · License: Public use (Opportunity Insights and the U.S. Census Bureau)

All data sources

References

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
Kind, Amy J. H.; Buckingham, William R. (2018). Making Neighborhood-Disadvantage Metrics Accessible: The Neighborhood Atlas. New England Journal of Medicine 378(26): 2456-2458. doi:10.1056/NEJMp1802313
Petterson, Stephen (2023). Deciphering the Neighborhood Atlas Area Deprivation Index: the consequences of not standardizing. Health Affairs Scholar 1(5): qxad063. doi:10.1093/haschl/qxad063
Hannan, Edward L.; Wu, Yifeng; Cozzens, Kimberly; Anderson, Brett (2023). The Neighborhood Atlas Area Deprivation Index For Measuring Socioeconomic Status: An Overemphasis On Home Value. Health Affairs 42(5): 702-709. doi:10.1377/hlthaff.2022.01406
Centers for Medicare & Medicaid Services (2024). ACO REACH Model Performance Year 2024 Model Update: Health Equity Benchmark Adjustment. CMS Innovation Center web page. www.cms.gov/priorities/innovation/innovation-models/reach-py24-model-perf

How to cite

@misc{singh2026anatomyofdeprivation,
  author = {Singh, Vishal},
  title = {What the deprivation index measures},
  year = {2026},
  url = {https://vishalsingh.org/zipcode/anatomy-of-deprivation},
  note = {In ZIP Code Destiny. Published 9 October 2026.}
}