# Violence and the life-expectancy gap: report

Chapter 8.5 of ZIP Code Destiny, by Vishal Singh (NYU Stern). Generated from the chapter's data files; the reading version is at https://vishalsingh.org/zipcode/violence-and-the-life-expectancy-gap. Table n of this report is figure n of the chapter.

## Identification

- Title: Violence and the life-expectancy gap
- Label: The lost years
- Chapter: 8.5 (Part VIII, Early death; An American life)
- Question: How much of the life-expectancy gap between the most and least deprived neighborhoods comes from violent deaths?
- Dek: A rough accounting that assumes 41.5 years of life lost per homicide attributes 4.8% of the gap of 8.3 years in life expectancy between the most and the least deprived tenths of tracts to homicide.
- Grain: neighborhoods
- Status: published
- Published: 2026-10-09
- Updated: not revised
- Data sources:
  - CDC injury and violence mapping. CDC homicide rates for tract clusters, annual 2022, 2023 and 2024 (modeled estimates for 1 to 9 deaths, zero for none, direct rates for 10 or more); county file 2019 and 2022 to 2024 for one check; years: 2019-2024 for counties, 2022-2024 for tracts (data as of 2026-05-14); weights: ACS 2019-2023 residents of each tract
  - USALEEP life expectancy. Life expectancy at birth by tract, deaths in 2010 to 2015, area-weighted onto 2020 tracts; years: 2010-2015; weights: 2019-2023 tract population
  - Area Deprivation Index. Area Deprivation Index national rank, 2023, for the tenths of tracts; years: 2023 release (version 4.0.1; built on ACS 2019-2023); weights: none (rank)
  - American Community Survey. Tract population (weights) and age shares, 2019-2023; years: 2019-2023 (5-year); weights: none (full count)
  - Published estimates from cited studies. Mean years of potential life lost per homicide victim and per suicide decedent (Zimmerman, Fridel and Trovato 2024) and a decomposition of the Black-white gap in male life expectancy in Washington, D.C. (Roberts, Reither and Lim 2020); years: as reported in each study; weights: none (published estimates)
- Canonical URL: https://vishalsingh.org/zipcode/violence-and-the-life-expectancy-gap
- Report: https://vishalsingh.org/zipcode/violence-and-the-life-expectancy-gap/report
- Markdown: https://vishalsingh.org/zipcode/violence-and-the-life-expectancy-gap/report.md
- BibTeX key: singh2026violenceandthelifeexpectancygap

```bibtex
@misc{singh2026violenceandthelifeexpectancygap,
  author = {Singh, Vishal},
  title = {Violence and the life-expectancy gap},
  year = {2026},
  url = {https://vishalsingh.org/zipcode/violence-and-the-life-expectancy-gap},
  note = {In ZIP Code Destiny. Published 9 October 2026.}
}
```

## Principal estimates

The facts the dek and the opening prose cite, in order of appearance.

| Estimand | Population | Period | Value | Interval | n | Weights | Source | Fact id |
| --- | --- | --- | ---: | --- | ---: | --- | --- | --- |
| mean of life expectancy at birth (specific to the victim's race, ethnicity and sex, 2019) minus age at death, from Zimmerman, Fridel and Trovato (2024), table 1; used here as an assumed setting for every tenth of tracts | homicide victims in the National Violent Death Reporting System, 44 states and DC | deaths in 2003 to 2019 | 41.5 |   | 98,617 homicide victims | none (mean over victims) | Published estimates from cited studies | yll_setting |
| rough accounting: the part of the gap attributed to homicide over the whole gap, bootstrap interval over counties (draws shared) | residents of non-institutional tracts in the 50 states and DC with a life expectancy estimate and an ADI rank | homicide rates for 2022 to 2024 with life expectancy for deaths in 2010 to 2015 | 4.8% | 95% CI 4.1% to 5.4% | 14,089 tracts | ACS 2019-2023 tract population | CDC injury and violence mapping | viol_gap_share |
| difference between the population-weighted mean life expectancy of the least and the most deprived tenth of tracts, bootstrap interval over counties (draws shared by the two means) | residents of non-institutional tracts in the 50 states and DC with a life expectancy estimate and an ADI rank | deaths in 2010 to 2015 (USALEEP, area-weighted onto 2020 tracts); ADI 2023; ACS 2019-2023 population | 8.3 | 95% CI 7.9 to 8.7 | 7,844 tracts | ACS 2019-2023 tract population | USALEEP life expectancy | le_decile_drop |
| the gap in life expectancy less the part attributed to homicide by rough accounting (a remainder, all other causes) | residents of non-institutional tracts in the 50 states and DC with a life expectancy estimate and an ADI rank | homicide rates for 2022 to 2024 with life expectancy for deaths in 2010 to 2015 | 7.9 | 95% CI 7.5 to 8.3 | 14,089 tracts | ACS 2019-2023 tract population | CDC injury and violence mapping | other_gap_years |
| the remainder over the whole gap: the share not attributed to homicide | residents of non-institutional tracts in the 50 states and DC with a life expectancy estimate and an ADI rank | homicide rates for 2022 to 2024 with life expectancy for deaths in 2010 to 2015 | 95.2% | 95% CI 94.6% to 95.9% | 14,089 tracts | ACS 2019-2023 tract population | CDC injury and violence mapping | other_gap_share |
| rough accounting with years of life lost per homicide set at the largest possible setting: the life expectancy of the most deprived tenth, bootstrap interval over counties | residents of non-institutional tracts in the 50 states and DC with a life expectancy estimate and an ADI rank | homicide rates for 2022 to 2024 with life expectancy for deaths in 2010 to 2015 | 8.5% | 95% CI 7.3% to 9.7% | 14,089 tracts | ACS 2019-2023 tract population | CDC injury and violence mapping | share_bound |
| share of residents in the highest group, mean over the three ways of leaving out a year | residents of those tracts that sit in a CDC cluster with a published homicide rate in each year from 2022 to 2024, outside New York City | annual homicide rates for 2022, 2023 and 2024, averaged (2024 provisional); ADI 2023; ACS 2019-2023 population | 1.3% |   |   | none (full count) | CDC injury and violence mapping | band_top_pop_share |
| part of the difference attributed to homicide by rough accounting over the whole difference, bootstrap interval over counties (draws shared) | residents of tracts in the highest group of clusters by homicide rate | homicide rates for 2022 to 2024 with life expectancy for deaths in 2010 to 2015 | 18.5% | 95% CI 16.5% to 20.3% | 1,337 tracts | ACS 2019-2023 tract population | CDC injury and violence mapping | band_top_share |
| difference between the population-weighted mean life expectancy of the lowest and the highest group, bootstrap interval over counties (draws shared) | residents of tracts in the highest group of clusters by homicide rate | homicide rates for 2022 to 2024 with life expectancy for deaths in 2010 to 2015 | 8.1 | 95% CI 7.7 to 8.4 | 1,337 tracts | ACS 2019-2023 tract population | CDC injury and violence mapping | band_top_gap |
| counties with a tract in the highest group; the smallest over the three ways of leaving out a year | residents of tracts in the highest group of clusters by homicide rate | annual homicide rates for 2022, 2023 and 2024, averaged (2024 provisional); ADI 2023; ACS 2019-2023 population | 54 |   |   | none (full count) | CDC injury and violence mapping | band_top_counties_min |
| counties with a tract in the highest group; the largest over the three ways of leaving out a year | residents of tracts in the highest group of clusters by homicide rate | annual homicide rates for 2022, 2023 and 2024, averaged (2024 provisional); ADI 2023; ACS 2019-2023 population | 67 |   |   | none (full count) | CDC injury and violence mapping | band_top_counties_max |

## Findings

Homicide accounts for 4.8% (95% CI 4.1% to 5.4%) of the 8.3 (95% CI 7.9 to 8.7) years by which life expectancy at birth, for deaths in 2010 to 2015, is lower in the most deprived tenth of tracts than in the least deprived. The figure comes from a rough accounting that takes each tenth's homicide rate for 2022 to 2024 and assumes that each death takes 41.5 years of life. The remaining 7.9 (95% CI 7.5 to 8.3) years, 95.2% (95% CI 94.6% to 95.9%) of the gap, are left to every other cause of death.

Neither source gives death rates by age for a tract. This is therefore a rough accounting and not an age-specific decomposition of the kind Arriaga described.[1] It multiplies a crude homicide rate by life expectancy and by an assumed number of years of life per homicide. The two sources also cover different periods, life expectancy for deaths in 2010 to 2015 and homicide rates for 2022 to 2024, and come from different models. The share is proportional to the assumed years and reaches 8.5% (95% CI 7.3% to 9.7%) at the largest value that one setting shared by every tenth could take.

CDC publishes one homicide rate for each cluster of about five neighboring tracts. In the clusters with the highest rates, which hold 1.3% of residents, homicide accounts for 18.5% (95% CI 16.5% to 20.3%) of a difference of 8.1 (95% CI 7.7 to 8.4) years from the clusters with the lowest rates. That group spans only 54 to 67 counties.

The concentration of gun homicide in a few clusters, and the mix of causes of early death, belong to Chapter 8.1, Gun homicide and the causes of early death (https://vishalsingh.org/zipcode/gun-homicide-and-causes-of-early-death). The gradient of life expectancy across deprivation tenths belongs to Chapter 8.4, Place and the length of a life (https://vishalsingh.org/zipcode/place-and-the-length-of-a-life), whose tract sample and gap this chapter reuses. Overdose, alcohol and suicide are in Chapter 8.2, Overdose, alcohol and suicide (https://vishalsingh.org/zipcode/overdose-alcohol-and-suicide), and their change since 2018 is in Chapter 8.3, Mortality trends, 2018 to 2024 (https://vishalsingh.org/zipcode/mortality-trends-2018-2024). The unit here is the tract, with each tract carrying the rate of its cluster, and nothing below describes an individual.

### Homicide rates rise across deprivation tenths, with the steepest step at the top

This chapter counts a violent death as a homicide by any means, as the CDC injury and violence dashboard defines it, and says "homicide" for what it counts. The National Violent Death Reporting System, the source of the years lost per death used below, counts suicide among violent deaths; this chapter does not. Suicides of every means are not counted. The dashboard's data dictionary names deaths in legal intervention only inside its all-firearm measure, which this chapter does not use. One check below adds firearm suicide. Firearms account for 75.1% of the homicide deaths that the rates imply.

The dashboard's tract file gives one rate for each cluster of about five neighboring tracts and repeats it on every tract of the cluster. The rate is a modeled estimate for clusters with a few deaths, zero for none and a direct rate for many, and Chapter 8.1, Gun homicide and the causes of early death (https://vishalsingh.org/zipcode/gun-homicide-and-causes-of-early-death) describes the file. The rates are crude rates per resident. For the clusters where CDC publishes an exact count of deaths for 2022, the rate times the cluster's residents divided by the count has a median of 0.99, as in the check in that chapter. A tract here carries the mean of its cluster's annual rates for 2022, 2023 and 2024, and the 2024 values are provisional.

The tenths are those of Chapter 8.4, Place and the length of a life (https://vishalsingh.org/zipcode/place-and-the-length-of-a-life), 77,936 tracts ranked on the Area Deprivation Index and cut into ten groups with equal numbers of tracts, not of residents. The least deprived tenth holds 32.2 million residents and the most deprived 23.0 million.

The mean rate rises from 1.8 (95% CI 1.3 to 2.2) homicide deaths per hundred thousand residents a year in the least deprived tenth to 14.8 (95% CI 12.8 to 16.9) in the most deprived, a ratio of 8.4 (95% CI 6.5 to 11.3) (Table 1). The rate is higher in each tenth than in the one before, and the smallest step among the first nine is 0.4 (95% CI 0.1 to 0.7). The step from the ninth tenth to the most deprived, 6.4 (95% CI 4.8 to 8.0), is larger than the next-largest step by 5.3 (95% CI 3.6 to 7.0).

Table 1. Homicide deaths per hundred thousand residents a year rise from 1.8 in the least deprived tenth of tracts to 14.8 in the most deprived

Rates are available for 92.8% of the residents in the gradient. New York City holds 2.6% of them and is left out, because its tract rates are near zero. The mean is 0.17 per hundred thousand in the tract file against 4.9 in the county file for the same boroughs, and 75.0% of its tracts sit at exactly zero.

The city holds 12.1% of the least deprived tenth, where the index, which understates deprivation in high-cost cities, ranks many of its tracts. Giving its tracts the county file's rate for their borough raises the least deprived tenth's rate from 1.8 to 2.1 (95% CI 1.5 to 2.6) and moves the share of the gap from 4.8% to 4.6% (95% CI 4.0% to 5.3%).

A further 4.7% of residents live in clusters where CDC withholds the rate in at least one of 2022, 2023 and 2024. The gap uses all of the gradient's tracts, the rates cover 72,225 of them, and each tenth's rate is applied to the whole tenth. Measuring the gap over only the tracts with a rate gives 4.7% (95% CI 4.0% to 5.3%).

### A homicide rate becomes years of life through two assumptions

For a group of tracts $$g$$, the years of life expectancy lost to homicide are approximated by

$$
\text{years lost}_g \;=\; r_g \times e_{0,g} \times L,
$$

where $$r_g$$ is the group's homicide rate per resident per year, $$e_{0,g}$$ is its life expectancy at birth and $$L$$ is the years of life a homicide takes. In a population with constant births and constant age-specific death rates, the crude death rate times life expectancy at birth equals the lifetime probability of dying of that cause. Under this approximation a newborn's lifetime risk of dying of homicide is 0.14% (95% CI 0.11% to 0.18%) in the least deprived tenth and 1.10% (95% CI 0.95% to 1.24%) in the most deprived. Multiplying the risk by the years a death takes gives, to a first approximation, the life expectancy that removing the cause would add. The approximation ignores competing risks, the chance that a person dies of another cause first, and the age mix of each tract. Residents of working age are 61.4% (95% CI 60.1% to 62.3%) of the least deprived tenth and 58.5% (95% CI 58.2% to 58.7%) of the most deprived, and scaling the rates to a common working-age share moves the answer to 5.0% (95% CI 4.3% to 5.7%).

The value of $$L$$ comes from outside both files. Zimmerman, Fridel and Trovato subtract the age at death from a life expectancy specific to the victim's race, ethnicity and sex for homicide victims in the National Violent Death Reporting System between 2003 and 2019. They find a mean of 41.5 years, with the loss floored at zero for a victim who died older than the life expectancy.[2] Most of the records come from 17 states. The accounting applies that national mean to every tenth. In that study Hispanic, Asian and Black victims lose more years on average than white victims, so the years lost would be above the mean where most victims come from those groups and below it elsewhere. Scaling the mean by each tenth's life expectancy over the national one, which lowers it for the most deprived tenth, gives 4.4% (95% CI 3.8% to 5.0%). The sweep below varies the setting.

### Most of the gap belongs to causes of death other than homicide

Across all tracts with a rate, the accounting puts the average loss to homicide at 0.18 (95% CI 0.16 to 0.20) years of life expectancy. Tenth by tenth, the loss is 0.06 (95% CI 0.05 to 0.07) years in the least deprived tenth and 0.45 (95% CI 0.39 to 0.52) in the most deprived.

The difference of 0.39 (95% CI 0.33 to 0.46) years is 4.8% (95% CI 4.1% to 5.4%) of the gap of 8.3 (95% CI 7.9 to 8.7) years between life expectancies of 82.3 (95% CI 82.0 to 82.6) and 74.0 (95% CI 73.7 to 74.3) years that Chapter 8.4, Place and the length of a life (https://vishalsingh.org/zipcode/place-and-the-length-of-a-life) reports. The other 95.2% (95% CI 94.6% to 95.9%) is the remainder, which includes every other cause of death, among them the overdose and suicide deaths in Chapter 8.2, Overdose, alcohol and suicide (https://vishalsingh.org/zipcode/overdose-alcohol-and-suicide) (Table 2). The two lines rest on different tracts: the observed line uses every tract of each tenth, and the homicide line uses those with a rate.

Table 2. Homicide accounts for 0.39 of the 8.3 years by which life expectancy is lower in the most deprived tenth of tracts than in the least deprived

The homicide part is concentrated at the top of the ranking. The homicide rate in the most deprived tenth is 6.4 (95% CI 4.8 to 8.0) higher than in the ninth, so the step in life expectancy between them contains 0.19 (95% CI 0.14 to 0.24) years of homicide, 10.2% (95% CI 8.0% to 12.3%) of that step's 1.9 (95% CI 1.6 to 2.1) years. That is a larger share than across the gradient as a whole. The accounting says how many years of life homicide removes, and says nothing about which of the other causes of death account for the rest.

### The share is proportional to the years of life each homicide takes

Because the accounting multiplies by $$L$$, the share rises in a straight line with that setting (Table 3). At 30 years per death homicide accounts for 3.4% (95% CI 2.9% to 3.9%) of the gap, and at 55 years for 6.3% (95% CI 5.4% to 7.2%).

A death cannot take more years than a person has. The sweep applies one value to every tenth, so the largest value it can take is the life expectancy of the most deprived tenth, 74.0 years, which would mean a death at birth in that tenth. That value gives 8.5% (95% CI 7.3% to 9.7%). No single value shared by every tenth puts the share above that. If every homicide in the most deprived tenth took that tenth's whole life expectancy and none in the least deprived tenth took any, the share would reach 9.8% (95% CI 8.6% to 11.0%).

Table 3. Homicide accounts for 3.4% of the gap if each takes 30 years of life and 6.3% if each takes 55, and for at most 8.5% when one value applies to every tenth

The other settings move the answer less than $$L$$ does (Table 4), and across the other changes tested the share ranges from 3.9% to 6.2%. Rates for 2022 and 2023 alone, the years that are final, give 5.1% (95% CI 4.4% to 5.9%). Neither file covers 2010 to 2015, the years behind the life expectancy estimates, and the county file reaches back only to 2019. Scaling each tenth's accounting to the 2019 county rates gives 3.9% (95% CI 3.3% to 4.5%), and counting only firearm homicide, the measure in Chapter 8.1, Gun homicide and the causes of early death (https://vishalsingh.org/zipcode/gun-homicide-and-causes-of-early-death), gives 4.1% (95% CI 3.4% to 4.7%). These are the two lowest shares in the table, and they cannot be told apart.

Adding firearm suicide, at the shorter 29.5 years that Zimmerman and colleagues find for suicide decedents, gives 6.2% (95% CI 5.6% to 6.9%), the highest. The rows that fill withheld rates with the low or the high end of the range CDC publishes for them bound what those rates could add, and both stay inside the range. So do the rows that use national life expectancy in place of each tenth's and count each tract once.

Table 4. Changing the rate window, the set of deaths counted or the conversion moves the share of the gap between 3.9% and 6.2%

### In the clusters with the highest homicide rates, homicide accounts for a larger share

A national gap averages over a wide range of places. Grouping clusters by their own homicide rate shows the other end. The clusters fall into six groups, from those with rates up to 2 to those over 40 deaths per hundred thousand residents a year, and Table 5 compares the life expectancy of each group with that of the lowest. If the annual errors in a cluster's rate are independent, chance would raise the measured rate of a group chosen for its high rate. Each cluster is therefore placed by two of the three annual rates, and the accounting uses the third year's, scaled to the level of the 2022 to 2024 mean. Forming the highest group and measuring it on the same years would put its mean rate at 59.5 instead of the 52.0 (95% CI 46.9 to 56.1) used here.

Life expectancy is 80.2 (95% CI 80.0 to 80.4) years in the lowest-rate group, whose mean rate is 1.8 (95% CI 1.7 to 1.9), and 72.1 (95% CI 71.8 to 72.4) years in the group over 40. That group holds 3.6 million residents, 1.3% of the residents of tracts with a rate.

The difference of 8.1 (95% CI 7.7 to 8.4) years is close to the gap of 8.3 years between the least and most deprived tenths. By the accounting, homicide costs this group 1.6 (95% CI 1.4 to 1.7) years of life expectancy, and the part beyond the lowest-rate group's loss, 1.5 (95% CI 1.3 to 1.6) years, is 18.5% (95% CI 16.5% to 20.3%) of the difference.

The share is higher in each higher group, from 4.5% (95% CI 3.9% to 5.1%) in the second-lowest to 12.4% (95% CI 11.4% to 13.4%) in the second-highest and 18.5% (95% CI 16.5% to 20.3%) in the highest.

Table 5. Life expectancy is 8.1 years lower in clusters with homicide rates over 40 per hundred thousand than in clusters with rates up to 2, and homicide accounts for 18.5% of the difference

The highest group is a narrow slice and its membership shifts with the year left out. It holds between 1,165 and 1,550 tracts, in between 194 and 264 clusters and between 54 and 67 counties, depending on the split.

The 10 counties with most of its residents hold between 52.7% and 57.1% of them, and Cook County alone between 12.2% and 15.1%. The interval for its share is a bootstrap over those counties.

About 62% of its residents have a 2024 rate that is a modeled estimate and provisional. Using 2022 and 2023 alone, with one year placing a cluster and the other measuring its rate, gives 20.0% (95% CI 17.9% to 21.9%) for the highest group.

An age-specific decomposition from one place gives a figure for scale. In Washington, D.C., Roberts, Reither and Lim attribute 2.43 of the 17.23 years by which Black men's life expectancy trailed white men's for deaths in 2015 to 2017 to homicide, or 14.1%.[3] That is a gap between groups of men in one city, not a gap between kinds of tracts. Light and Vachuska find that homicide contributed more than any other cause of death to the national gap between Black and white men's life expectancy in 2020 and 2021.[4] Chapter 8.1, Gun homicide and the causes of early death (https://vishalsingh.org/zipcode/gun-homicide-and-causes-of-early-death) shows the same pairing of high rates and short lives inside Cook County.

### Within a deprivation tenth, life expectancy is lower where homicide rates are higher

Deprivation and homicide move together, so the comparison of tenths mixes them. Splitting each tenth at its median homicide rate, among its tracts with a rate, holds the deprivation rank nearly fixed. In every tenth, life expectancy is lower in the group with the higher rate (Table 6), by 0.8 (95% CI 0.5 to 1.1) years in the least deprived tenth and 2.1 (95% CI 1.7 to 2.6) in the most deprived. In the most deprived tenth the two groups' mean rates are 5.0 (95% CI 4.7 to 5.4) and 24.6 (95% CI 21.7 to 27.3) per hundred thousand.

Table 6. Within each tenth of tracts on the deprivation index, life expectancy is lower in the group with the higher homicide rate, by 0.8 years in the least deprived tenth and 2.1 in the most deprived

The accounting attributes 0.6 (95% CI 0.5 to 0.7) years of the most deprived tenth's difference to homicide, 27.5% (95% CI 22.4% to 33.8%), and 6.3% (95% CI 4.9% to 8.4%) of the least deprived tenth's. Most of each difference is left to other causes of death. The index is built from census measures of income, education, employment and housing, so tracts of the same rank can differ in conditions it does not measure, and homicide may mark them. The pattern also fits homicide being tied to other deaths, for example among residents who live with it. The data cannot separate the two.

### The homicide rate rises with deprivation under all four rankings

The tenths are again those of the life expectancy chapter, cut over its tracts that have every ranking, and the rate comes from the 66,773 of them with a usable rate outside New York City. Under the ADI baseline the rate runs from 1.8 (95% CI 1.3 to 2.2) homicide deaths per hundred thousand residents a year in the least deprived tenth to 14.3 (95% CI 12.4 to 16.3) in the most deprived, with a rank correlation of +1.00 between tenth and mean (Table 7). The standardized index, which applies the ADI's coefficients after standardizing each of its measures, gives 1.6 (95% CI 1.5 to 1.8) and 15.3 (95% CI 13.0 to 17.9) (+1.00), median household income gives 1.7 (95% CI 1.5 to 1.9) and 14.8 (95% CI 12.9 to 16.9) (+1.00), and the bachelor's degree share gives 2.3 (95% CI 2.0 to 2.7) and 10.8 (95% CI 9.6 to 12.1) (+1.00). Under the rule of Chapter 9.1, How far one ranking goes (https://vishalsingh.org/zipcode/how-far-one-ranking-goes) none of the three counts as a change, since the ends differ under each ranking and every correlation stays at or above 0.9.

Table 7. The homicide death rate is steadily worse toward the more deprived end under the ADI and under each of the three other rankings

### What these data cannot show

The accounting is not an age-specific decomposition. It applies one value of $$L$$ to every tenth, although victims' ages differ by place, and that value is a national mean for victims, most of them from 17 states, specific to each victim's race, ethnicity and sex and floored at zero. It converts a crude rate to a lifetime risk by assuming a stationary population. The share scales with $$L$$, so a reader who believes victims lose more or fewer years should read the answer off Table 3. The intervals around each share treat $$L$$ as known, and the sweep is the answer to that uncertainty. Arriaga's method needs rates by age and cause for each group, and neither file provides them below the county.

The two sources do not describe the same years. Life expectancy is the USALEEP estimate for deaths in 2010 to 2015, area-weighted onto 2020 tracts with standard errors that are good for mapping only, and the homicide rates are for 2022 to 2024. The county file's rates for 2019 are below its rates for 2022 to 2024 in every tenth, and no file here describes 2010 to 2015, so the share for those years could be lower or higher than shown. The rates are CDC's modeled estimates for clusters of neighboring tracts, repeated on each tract, which averages over neighborhoods and lowers measured concentration. That they are crude per-resident rates is inferred from the check against exact counts and not from CDC documentation. The 2024 values are provisional, and the rows that fill withheld rates from CDC's count ranges give bounds and not estimates. The 95% intervals come from a bootstrap that resamples counties, so they describe which counties are in the file and not the error of either model.

The comparisons are of places. A tract's life expectancy and its homicide rate come from different people and different decades, and a difference between a higher-rate and a lower-rate group does not show what homicide does to the people who live there. The accounting counts deaths from homicide and no other consequence of it, so it leaves out any association of violence with other deaths and with health short of death. The Area Deprivation Index understates deprivation in high-cost cities, and New York City, which holds a large share of the least deprived tenth, has published rates near zero that the accounting leaves out. Nothing in this chapter shows what reducing violence would do to life expectancy; Chapter 8.1, Gun homicide and the causes of early death (https://vishalsingh.org/zipcode/gun-homicide-and-causes-of-early-death) reviews evidence on programs aimed at violence.

## Tables

### Table 1. Homicide deaths per hundred thousand residents a year rise from 1.8 in the least deprived tenth of tracts to 14.8 in the most deprived

Mean annual homicide deaths per 100,000 residents, 2022 to 2024, by tenth of tracts on the Area Deprivation Index, tracts outside New York City with a published rate

| conf_high | conf_low | estimate | n_units | tenth |
| ---: | ---: | ---: | ---: | ---: |
| 2.16 | 1.35 | 1.76 | 6,479 | 1 |
| 3.18 | 2.07 | 2.62 | 6,827 | 2 |
| 3.48 | 2.63 | 3.03 | 7,072 | 3 |
| 4.18 | 3.29 | 3.70 | 7,220 | 4 |
| 4.69 | 3.97 | 4.33 | 7,178 | 5 |
| 5.63 | 4.68 | 5.16 | 7,602 | 6 |
| 7.09 | 5.55 | 6.27 | 7,317 | 7 |
| 8.32 | 6.48 | 7.34 | 7,495 | 8 |
| 9.59 | 7.41 | 8.43 | 7,425 | 9 |
| 16.85 | 12.76 | 14.81 | 7,610 | 10 |

Note: The homicide rate is CDC's modeled estimate for a cluster of about five neighboring tracts, repeated on each tract; a tract's rate is the mean of its cluster's annual rates for 2022 to 2024. 2024 is provisional.

Note: Homicide is deaths from homicide by any means, as CDC's dashboard defines it. Suicides of every means and deaths in legal intervention are not counted; the panel publishes legal intervention only inside all firearm deaths.

Note: New York City is left out of every rate: its tracts hold 2.6% of residents in the gradient and carry rates near zero in CDC's tract file, a mean of 0.17 per hundred thousand against 4.9 in the county file, with 75.0% of its tracts at exactly zero. A further 4.7% of residents live in clusters where CDC withholds the rate in at least one year, and 92.8% are in tracts with a rate.

Note: Intervals are 95% intervals from a bootstrap that draws whole counties with replacement (replicates and seed are in the methods note).

Note: Across all the tracts in the comparison the rate is 5.4 per hundred thousand residents a year.

Source: CDC Mapping Injury, Overdose, and Violence, tract file, homicide, 2022-2024; modeled estimates for tract clusters; population-weighted

Data: https://vishalsingh.org/zipcode/report/violence-and-the-life-expectancy-gap/figures/homicide_by_tenth.json

### Table 2. Homicide accounts for 0.39 of the 8.3 years by which life expectancy is lower in the most deprived tenth of tracts than in the least deprived

Years of life expectancy at birth below the least deprived tenth, by tenth of tracts on the Area Deprivation Index: observed for deaths in 2010 to 2015, and the part attributed to homicide by a rough accounting with homicide rates for 2022 to 2024

| conf_high | conf_low | decile | estimate | n_units | series |
| ---: | ---: | ---: | ---: | ---: | --- |
| n/a | n/a | 1 | 0 | 7,820 | All causes (observed) |
| n/a | n/a | 1 | 0 | 6,479 | Homicide (rough accounting) |
| 1.55 | 1.03 | 2 | 1.33 | 7,860 | All causes (observed) |
| 0.040 | 0.015 | 2 | 0.028 | 6,827 | Homicide (rough accounting) |
| 2.25 | 1.65 | 3 | 1.98 | 7,737 | All causes (observed) |
| 0.052 | 0.029 | 3 | 0.041 | 7,072 | Homicide (rough accounting) |
| 2.82 | 2.23 | 4 | 2.54 | 7,757 | All causes (observed) |
| 0.077 | 0.048 | 4 | 0.062 | 7,220 | Homicide (rough accounting) |
| 3.44 | 2.85 | 5 | 3.17 | 7,613 | All causes (observed) |
| 0.097 | 0.066 | 5 | 0.082 | 7,178 | Homicide (rough accounting) |
| 4.16 | 3.59 | 6 | 3.89 | 8,031 | All causes (observed) |
| 0.128 | 0.088 | 6 | 0.108 | 7,602 | Homicide (rough accounting) |
| 4.87 | 4.29 | 7 | 4.60 | 7,722 | All causes (observed) |
| 0.172 | 0.115 | 7 | 0.142 | 7,317 | Homicide (rough accounting) |
| 5.69 | 5.09 | 8 | 5.41 | 7,844 | All causes (observed) |
| 0.209 | 0.143 | 8 | 0.174 | 7,495 | Homicide (rough accounting) |
| 6.70 | 6.09 | 9 | 6.42 | 7,708 | All causes (observed) |
| 0.245 | 0.170 | 9 | 0.205 | 7,425 | Homicide (rough accounting) |
| 8.66 | 7.87 | 10 | 8.28 | 7,844 | All causes (observed) |
| 0.459 | 0.331 | 10 | 0.395 | 7,610 | Homicide (rough accounting) |

Note: Life expectancy is the USALEEP estimate for deaths in 2010 to 2015, area-weighted from 2010 onto 2020 tracts; tracts where group quarters hold a large share of residents are left out. Tenths hold equal numbers of tracts and means are weighted by residents.

Note: The accounting is a rough one, not an age-specific decomposition: it multiplies a crude homicide rate by life expectancy to get the lifetime risk of dying of homicide, and by the years of life a death takes, assumed to be 41.5 for every tenth. The rates are for 2022 to 2024 and life expectancy is for deaths in 2010 to 2015.

Note: The two lines rest on different tracts. The observed line uses every tract of each tenth, and the homicide line uses the tracts with a published rate, applying each tenth's rate to the whole tenth.

Note: New York City is left out of every rate: its tracts hold 2.6% of residents in the gradient and carry rates near zero in CDC's tract file, a mean of 0.17 per hundred thousand against 4.9 in the county file, with 75.0% of its tracts at exactly zero. A further 4.7% of residents live in clusters where CDC withholds the rate in at least one year, and 92.8% are in tracts with a rate.

Note: The homicide rate is CDC's modeled estimate for a cluster of about five neighboring tracts, repeated on each tract; a tract's rate is the mean of its cluster's annual rates for 2022 to 2024. 2024 is provisional.

Note: Intervals are 95% intervals from a bootstrap that draws whole counties with replacement (replicates and seed are in the methods note).

Source: CDC Mapping Injury, Overdose, and Violence, tract file, homicide, 2022-2024 (modeled estimates for tract clusters); USALEEP life expectancy, 2010-2015; Area Deprivation Index 2023; years of life lost per death assumed from published estimates; population-weighted

Data: https://vishalsingh.org/zipcode/report/violence-and-the-life-expectancy-gap/figures/gap_by_tenth.json

### Table 3. Homicide accounts for 3.4% of the gap if each takes 30 years of life and 6.3% if each takes 55, and for at most 8.5% when one value applies to every tenth

Share of the gap in life expectancy between the least and the most deprived tenths of tracts attributed to homicide, by the years of life assumed lost per homicide

| conf_high | conf_low | estimate | years_lost |
| ---: | ---: | ---: | ---: |
| n/a | n/a | 0 | 0 |
| 0.131 | 0.098 | 0.115 | 1 |
| 0.262 | 0.197 | 0.230 | 2 |
| 0.394 | 0.295 | 0.345 | 3 |
| 0.525 | 0.393 | 0.460 | 4 |
| 0.656 | 0.491 | 0.574 | 5 |
| 0.787 | 0.590 | 0.689 | 6 |
| 0.919 | 0.688 | 0.804 | 7 |
| 1.05 | 0.786 | 0.919 | 8 |
| 1.18 | 0.885 | 1.03 | 9 |
| 1.31 | 0.983 | 1.15 | 10 |
| 1.44 | 1.08 | 1.26 | 11 |
| 1.57 | 1.18 | 1.38 | 12 |
| 1.71 | 1.28 | 1.49 | 13 |
| 1.84 | 1.38 | 1.61 | 14 |
| 1.97 | 1.47 | 1.72 | 15 |
| 2.10 | 1.57 | 1.84 | 16 |
| 2.23 | 1.67 | 1.95 | 17 |
| 2.36 | 1.77 | 2.07 | 18 |
| 2.49 | 1.87 | 2.18 | 19 |
| 2.62 | 1.97 | 2.30 | 20 |
| 2.76 | 2.06 | 2.41 | 21 |
| 2.89 | 2.16 | 2.53 | 22 |
| 3.02 | 2.26 | 2.64 | 23 |
| 3.15 | 2.36 | 2.76 | 24 |
| 3.28 | 2.46 | 2.87 | 25 |
| 3.41 | 2.56 | 2.99 | 26 |
| 3.54 | 2.65 | 3.10 | 27 |
| 3.67 | 2.75 | 3.22 | 28 |
| 3.81 | 2.85 | 3.33 | 29 |
| 3.94 | 2.95 | 3.45 | 30 |
| 4.07 | 3.05 | 3.56 | 31 |
| 4.20 | 3.15 | 3.68 | 32 |
| 4.33 | 3.24 | 3.79 | 33 |
| 4.46 | 3.34 | 3.91 | 34 |
| 4.59 | 3.44 | 4.02 | 35 |
| 4.72 | 3.54 | 4.14 | 36 |
| 4.86 | 3.64 | 4.25 | 37 |
| 4.99 | 3.73 | 4.37 | 38 |
| 5.12 | 3.83 | 4.48 | 39 |
| 5.25 | 3.93 | 4.60 | 40 |
| 5.38 | 4.03 | 4.71 | 41 |
| 5.51 | 4.13 | 4.83 | 42 |
| 5.64 | 4.23 | 4.94 | 43 |
| 5.77 | 4.33 | 5.05 | 44 |
| 5.91 | 4.42 | 5.17 | 45 |
| 6.04 | 4.52 | 5.29 | 46 |
| 6.17 | 4.62 | 5.40 | 47 |
| 6.30 | 4.72 | 5.51 | 48 |
| 6.43 | 4.82 | 5.63 | 49 |
| 6.56 | 4.91 | 5.75 | 50 |
| 6.69 | 5.01 | 5.86 | 51 |
| 6.82 | 5.11 | 5.97 | 52 |
| 6.96 | 5.21 | 6.09 | 53 |
| 7.09 | 5.31 | 6.20 | 54 |
| 7.22 | 5.41 | 6.32 | 55 |
| 7.35 | 5.50 | 6.43 | 56 |
| 7.48 | 5.60 | 6.55 | 57 |
| 7.61 | 5.70 | 6.66 | 58 |
| 7.74 | 5.80 | 6.78 | 59 |
| 7.87 | 5.90 | 6.89 | 60 |
| 8.00 | 6.00 | 7.01 | 61 |
| 8.14 | 6.09 | 7.12 | 62 |
| 8.27 | 6.19 | 7.24 | 63 |
| 8.40 | 6.29 | 7.35 | 64 |
| 8.53 | 6.39 | 7.47 | 65 |
| 8.66 | 6.49 | 7.58 | 66 |
| 8.79 | 6.58 | 7.70 | 67 |
| 8.92 | 6.68 | 7.81 | 68 |
| 9.05 | 6.78 | 7.93 | 69 |
| 9.19 | 6.88 | 8.04 | 70 |
| 9.32 | 6.98 | 8.16 | 71 |
| 9.45 | 7.08 | 8.27 | 72 |
| 9.58 | 7.17 | 8.39 | 73 |
| 9.71 | 7.27 | 8.50 | 74 |

Note: The accounting multiplies each tenth's homicide rate for 2022 to 2024 by its life expectancy and by the years lost per death, so the share is proportional to that setting. The largest value that can apply to every tenth is the life expectancy of the most deprived tenth, which would mean that every homicide took a whole life. If every homicide in the most deprived tenth took that tenth's whole life expectancy and none in the least deprived tenth took any, which no single value gives, the share would reach 9.8%.

Note: Intervals are 95% intervals from a bootstrap that draws whole counties with replacement (replicates and seed are in the methods note).

Source: CDC Mapping Injury, Overdose, and Violence, tract file, homicide, 2022-2024 (modeled estimates for tract clusters); USALEEP life expectancy, 2010-2015; Area Deprivation Index 2023; years of life lost per death assumed from published estimates; population-weighted

Data: https://vishalsingh.org/zipcode/report/violence-and-the-life-expectancy-gap/figures/years_lost_sweep.json

### Table 4. Changing the rate window, the set of deaths counted or the conversion moves the share of the gap between 3.9% and 6.2%

Part of the gap in life expectancy between the least and the most deprived tenths of tracts attributed to homicide, in the main setting and in each variation, at the published years of life lost per death

| Setting | Years attributed to homicide | Gap (years) | Share of the gap (%) | Interval, low (%) | Interval, high (%) | Tracts with a rate | Of these, in the end tenths |
| --- | ---: | ---: | ---: | ---: | ---: | ---: | ---: |
| Main setting: homicide rates for 2022 to 2024, all tracts with a rate | 0.40 | 8.28 | 4.8 | 4.1 | 5.5 | 72,225 | 14,089 |
| Rates for 2022 and 2023 only (final data) | 0.42 | 8.28 | 5.1 | 4.4 | 5.9 | 73,581 | 14,275 |
| Withheld rates filled with the low end of CDC's count range | 0.39 | 8.28 | 4.7 | 4.0 | 5.3 | 75,862 | 14,636 |
| Withheld rates filled with the high end of CDC's count range | 0.37 | 8.28 | 4.5 | 3.8 | 5.1 | 75,862 | 14,636 |
| Rates scaled to the 2019 level, from the county file | 0.32 | 8.28 | 3.9 | 3.3 | 4.5 | 70,374 | 13,779 |
| Firearm homicide only | 0.34 | 8.28 | 4.1 | 3.5 | 4.7 | 72,285 | 14,115 |
| Homicide plus firearm suicide | 0.52 | 8.28 | 6.2 | 5.6 | 6.9 | 50,213 | 10,102 |
| National life expectancy in place of each tenth's | 0.43 | 8.28 | 5.2 | 4.4 | 5.9 | 72,225 | 14,089 |
| Rates scaled to a common share of residents aged 18 to 64 | 0.41 | 8.28 | 5.0 | 4.3 | 5.7 | 72,225 | 14,089 |
| Gap taken over the tracts that have a rate | 0.40 | 8.44 | 4.7 | 4.0 | 5.3 | 72,225 | 14,089 |
| Each tract counted once, not weighted by residents | 0.41 | 8.35 | 5.0 | 4.2 | 5.7 | 72,225 | 14,089 |
| New York City tracts at the county file's rate | 0.38 | 8.28 | 4.6 | 4.0 | 5.3 | 74,299 | 15,117 |
| Years lost scaled by each tenth's life expectancy over the national | 0.36 | 8.28 | 4.4 | 3.8 | 5.0 | 72,225 | 14,089 |

Note: Firearm suicide enters at the mean years of life lost per suicide, 29.5, and only for tracts with a published rate for both. Rates scaled to the 2019 level use the county file's ratio of the 2019 rate to the 2022 to 2024 rate in each tenth, so they describe counties and not clusters. The two rows that fill withheld rates use the low and the high end of the count range CDC publishes for them, so they bound what the withheld rates could add and are not estimates. The row for New York City gives the city's tracts the county file's rate for their borough. The row that scales the years lost multiplies them by each tenth's life expectancy over the national one. The rows that measure the gap over the tracts that have a rate and that count each tract once have gaps of their own, which the table gives.

Source: CDC Mapping Injury, Overdose, and Violence, tract file, homicide, 2022-2024 (modeled estimates for tract clusters); USALEEP life expectancy, 2010-2015; Area Deprivation Index 2023; years of life lost per death assumed from published estimates; population-weighted

Data: https://vishalsingh.org/zipcode/report/violence-and-the-life-expectancy-gap/figures/setting_table.json

### Table 5. Life expectancy is 8.1 years lower in clusters with homicide rates over 40 per hundred thousand than in clusters with rates up to 2, and homicide accounts for 18.5% of the difference

Years of life expectancy at birth below the lowest-rate group of tract clusters, by group of clusters ranked on the homicide rate for 2022 to 2024: the observed difference split into the part attributed to homicide and a remainder

| band | component | conf_high | conf_low | years |
| --- | --- | ---: | ---: | ---: |
| Over 2 to 5 | Homicide (rough accounting) | 0.055 | 0.046 | 0.051 |
| Over 2 to 5 | All other causes (remainder) | 1.25 | 0.940 | 1.09 |
| Over 5 to 10 | Homicide (rough accounting) | 0.140 | 0.127 | 0.134 |
| Over 5 to 10 | All other causes (remainder) | 2.42 | 2.03 | 2.22 |
| Over 10 to 20 | Homicide (rough accounting) | 0.334 | 0.305 | 0.319 |
| Over 10 to 20 | All other causes (remainder) | 3.90 | 3.43 | 3.67 |
| Over 20 to 40 | Homicide (rough accounting) | 0.755 | 0.675 | 0.718 |
| Over 20 to 40 | All other causes (remainder) | 5.42 | 4.74 | 5.09 |
| Over 40 | Homicide (rough accounting) | 1.62 | 1.34 | 1.50 |
| Over 40 | All other causes (remainder) | 6.98 | 6.17 | 6.58 |

Note: Clusters are ranked on the mean of two of the three annual rates, and the rate used in the accounting is the third year's, put on the level of the three-year mean by one national factor. The figure averages the three ways of leaving out a year. If the annual errors in a cluster's rate are independent, this keeps chance from inflating the rate of a group chosen for its high rate: the highest group's rate is 52.0 this way and would be 59.5 if the same years chose the group and measured it. About 62% of the highest group's residents have a 2024 rate that is a modeled, provisional estimate.

Note: The groups differ between the three ways of leaving out a year. The highest group holds 1,165 to 1,550 tracts, 194 to 264 clusters and 54 to 67 counties, and the 10 counties with most of its residents hold 52.7% to 57.1% of them, so its interval rests on few counties. Using only 2022 and 2023, one year placing a cluster and the other measuring it, the highest group's share is 20.0%.

Note: Groups hold the residents of tracts outside New York City whose cluster has a rate in every year. The lowest group, rates up to 2, has a mean rate of 1.8 and a life expectancy of 80.2 years.

Note: Life expectancy is the USALEEP estimate for deaths in 2010 to 2015, area-weighted from 2010 onto 2020 tracts; tracts where group quarters hold a large share of residents are left out. Tenths hold equal numbers of tracts and means are weighted by residents.

Note: Intervals are 95% intervals from a bootstrap that draws whole counties with replacement (replicates and seed are in the methods note).

Source: CDC Mapping Injury, Overdose, and Violence, tract file, homicide, 2022-2024 (modeled estimates for tract clusters); USALEEP life expectancy, 2010-2015; Area Deprivation Index 2023; years of life lost per death assumed from published estimates; population-weighted

Data: https://vishalsingh.org/zipcode/report/violence-and-the-life-expectancy-gap/figures/gap_by_violence_band.json

### Table 6. Within each tenth of tracts on the deprivation index, life expectancy is lower in the group with the higher homicide rate, by 0.8 years in the least deprived tenth and 2.1 in the most deprived

Population-weighted mean life expectancy at birth for deaths in 2010 to 2015 in the lower-rate and higher-rate groups of each tenth of tracts on the Area Deprivation Index, split on the homicide rate for 2022 to 2024

| hi1 | hi2 | higher | lo1 | lo2 | lower | n1 | n2 | tenth |
| ---: | ---: | ---: | ---: | ---: | ---: | ---: | ---: | --- |
| 82.98 | 82.28 | 81.97 | 82.63 | 81.73 | 82.81 | 14320000 | 12679000 | 1 (least deprived) |
| 81.64 | 80.94 | 80.57 | 81.31 | 80.29 | 81.47 | 15296000 | 15016000 | 2 |
| 80.99 | 80.19 | 79.90 | 80.73 | 79.64 | 80.86 | 16173000 | 15971000 | 3 |
| 80.42 | 79.42 | 79.22 | 80.09 | 79.01 | 80.26 | 16346000 | 16076000 | 4 |
| 79.79 | 78.81 | 78.61 | 79.50 | 78.40 | 79.64 | 15852000 | 15625000 | 5 |
| 79.18 | 77.95 | 77.74 | 78.90 | 77.55 | 79.04 | 15939000 | 15798000 | 6 |
| 78.45 | 77.25 | 77.07 | 78.16 | 76.91 | 78.30 | 14581000 | 14508000 | 7 |
| 77.77 | 76.32 | 76.15 | 77.47 | 75.98 | 77.62 | 13908000 | 13808000 | 8 |
| 76.75 | 75.34 | 75.14 | 76.45 | 74.94 | 76.59 | 12765000 | 12689000 | 9 |
| 75.45 | 73.16 | 72.90 | 74.71 | 72.68 | 75.04 | 11177000 | 11140000 | 10 (most deprived) |

Note: The groups are formed by cross-fitting, as in the figure of groups: each tenth is split on the mean of two of the three annual rates and the figure averages the three ways of leaving out a year. The n columns give residents. Each tenth is split among the tracts with a usable homicide rate, so a tenth has fewer tracts here than in the other figures.

Note: The split is on the homicide rate and the outcome is life expectancy from another source and another period, so the difference is not built by the sorting. It compares places and does not show what violence does to the people in them.

Note: Life expectancy is the USALEEP estimate for deaths in 2010 to 2015, area-weighted from 2010 onto 2020 tracts; tracts where group quarters hold a large share of residents are left out. Tenths hold equal numbers of tracts and means are weighted by residents.

Note: Intervals are 95% intervals from a bootstrap that draws whole counties with replacement (replicates and seed are in the methods note).

Source: CDC Mapping Injury, Overdose, and Violence, tract file, homicide, 2022-2024 (modeled estimates for tract clusters); USALEEP life expectancy, 2010-2015; Area Deprivation Index 2023; population-weighted

Data: https://vishalsingh.org/zipcode/report/violence-and-the-life-expectancy-gap/figures/same_deprivation.json

### Table 7. The homicide death rate is steadily worse toward the more deprived end under the ADI and under each of the three other rankings

Homicide deaths per 100,000 residents a year by tenth of tracts ranked four ways, over the 66,773 tracts that have all four rankings

| bin | conf_high | conf_low | estimate | n_units | ranking |
| ---: | ---: | ---: | ---: | ---: | --- |
| 1 | 2.19 | 1.33 | 1.77 | 6,021 | Area Deprivation Index |
| 2 | 3.16 | 2.08 | 2.61 | 6,243 | Area Deprivation Index |
| 3 | 3.50 | 2.66 | 3.08 | 6,712 | Area Deprivation Index |
| 4 | 4.16 | 3.33 | 3.73 | 6,635 | Area Deprivation Index |
| 5 | 4.68 | 3.99 | 4.33 | 6,822 | Area Deprivation Index |
| 6 | 5.55 | 4.65 | 5.08 | 6,656 | Area Deprivation Index |
| 7 | 6.96 | 5.50 | 6.15 | 6,942 | Area Deprivation Index |
| 8 | 8.01 | 6.31 | 7.09 | 6,791 | Area Deprivation Index |
| 9 | 9.35 | 7.29 | 8.24 | 6,956 | Area Deprivation Index |
| 10 | 16.28 | 12.40 | 14.34 | 6,995 | Area Deprivation Index |
| 1 | 1.83 | 1.45 | 1.63 | 6,743 | Standardized index |
| 2 | 2.68 | 2.29 | 2.49 | 6,684 | Standardized index |
| 3 | 3.36 | 2.93 | 3.15 | 6,654 | Standardized index |
| 4 | 3.82 | 3.39 | 3.60 | 6,709 | Standardized index |
| 5 | 4.45 | 3.95 | 4.20 | 6,720 | Standardized index |
| 6 | 5.05 | 4.43 | 4.74 | 6,659 | Standardized index |
| 7 | 5.97 | 5.15 | 5.55 | 6,665 | Standardized index |
| 8 | 7.40 | 6.36 | 6.85 | 6,677 | Standardized index |
| 9 | 9.99 | 8.28 | 9.08 | 6,605 | Standardized index |
| 10 | 17.88 | 13.03 | 15.35 | 6,657 | Standardized index |
| 1 | 1.94 | 1.49 | 1.70 | 6,631 | Median household income |
| 2 | 2.84 | 2.41 | 2.63 | 6,584 | Median household income |
| 3 | 3.47 | 2.99 | 3.23 | 6,603 | Median household income |
| 4 | 4.08 | 3.58 | 3.83 | 6,644 | Median household income |
| 5 | 4.72 | 4.11 | 4.41 | 6,627 | Median household income |
| 6 | 5.39 | 4.67 | 5.03 | 6,643 | Median household income |
| 7 | 6.20 | 5.24 | 5.70 | 6,695 | Median household income |
| 8 | 7.46 | 6.23 | 6.83 | 6,750 | Median household income |
| 9 | 9.89 | 8.09 | 8.96 | 6,771 | Median household income |
| 10 | 16.87 | 12.85 | 14.79 | 6,825 | Median household income |
| 1 | 2.71 | 1.99 | 2.33 | 6,499 | Bachelor's degree share |
| 2 | 3.12 | 2.55 | 2.82 | 6,623 | Bachelor's degree share |
| 3 | 3.65 | 3.07 | 3.35 | 6,587 | Bachelor's degree share |
| 4 | 4.53 | 3.79 | 4.14 | 6,595 | Bachelor's degree share |
| 5 | 5.23 | 4.42 | 4.81 | 6,631 | Bachelor's degree share |
| 6 | 5.96 | 4.95 | 5.43 | 6,699 | Bachelor's degree share |
| 7 | 6.71 | 5.49 | 6.06 | 6,727 | Bachelor's degree share |
| 8 | 7.38 | 6.16 | 6.75 | 6,775 | Bachelor's degree share |
| 9 | 9.25 | 7.37 | 8.26 | 6,791 | Bachelor's degree share |
| 10 | 12.14 | 9.55 | 10.79 | 6,846 | Bachelor's degree share |

Note: Least and most deprived tenth: ADI 1.8 and 14.3, rank correlation +1.00; standardized index 1.6 and 15.3, +1.00; median household income 1.7 and 14.8, +1.00; bachelor's degree share 2.3 and 10.8, +1.00.

Note: The standardized index applies the ADI's coefficients to its measures after standardizing each one, so home value and income carry less weight than in the ADI. Income and degree share are single census measures, each ranked from the highest value to the lowest.

Source: CDC Mapping Injury, Overdose, and Violence, tract file, homicide, 2022-2024; modeled estimates for tract clusters; population-weighted; ADI 2023 and ACS 2019-2023 tract tables (B19013, B15003 and the 15 rebuilt ADI measures)

Data: https://vishalsingh.org/zipcode/report/violence-and-the-life-expectancy-gap/figures/alt_rankings.json

## All estimates

Every fact in facts.json (111), sorted by id.

| Fact id | Value | Unit | Estimand | Population | Period | Interval | n | Weights | Source | Note |
| --- | ---: | --- | --- | --- | --- | --- | ---: | --- | --- | --- |
| alt_adi_least | 1.8 | Homicide deaths per 100,000 residents a year | population-weighted mean of the tract homicide rate (each tract carries its cluster's rate); 95% intervals from a bootstrap over counties | the least deprived tenth of tracts ranked by Area Deprivation Index, among the tracts with all four rankings | annual homicide rates for 2022, 2023 and 2024, averaged (2024 provisional); ADI 2023; ACS 2019-2023 population | 95% CI 1.3 to 2.2 | 6,021 tracts | ACS 2019-2023 tract population | CDC injury and violence mapping |   |
| alt_adi_most | 14.3 | Homicide deaths per 100,000 residents a year | population-weighted mean of the tract homicide rate (each tract carries its cluster's rate); 95% intervals from a bootstrap over counties | the most deprived tenth of tracts ranked by Area Deprivation Index, among the tracts with all four rankings | annual homicide rates for 2022, 2023 and 2024, averaged (2024 provisional); ADI 2023; ACS 2019-2023 population | 95% CI 12.4 to 16.3 | 6,995 tracts | ACS 2019-2023 tract population | CDC injury and violence mapping |   |
| alt_adi_rho | +1.00 | rank correlation | Spearman rank correlation between bin order and bin mean, each bin counted once | the bins of tracts ranked by Area Deprivation Index | annual homicide rates for 2022, 2023 and 2024, averaged (2024 provisional); ADI 2023; ACS 2019-2023 population |   |   | each bin counted once | CDC injury and violence mapping |   |
| alt_degree_least | 2.3 | Homicide deaths per 100,000 residents a year | population-weighted mean of the tract homicide rate (each tract carries its cluster's rate); 95% intervals from a bootstrap over counties | the least deprived tenth of tracts ranked by Bachelor's degree share, among the tracts with all four rankings | annual homicide rates for 2022, 2023 and 2024, averaged (2024 provisional); ADI 2023; ACS 2019-2023 population | 95% CI 2.0 to 2.7 | 6,499 tracts | ACS 2019-2023 tract population | CDC injury and violence mapping |   |
| alt_degree_most | 10.8 | Homicide deaths per 100,000 residents a year | population-weighted mean of the tract homicide rate (each tract carries its cluster's rate); 95% intervals from a bootstrap over counties | the most deprived tenth of tracts ranked by Bachelor's degree share, among the tracts with all four rankings | annual homicide rates for 2022, 2023 and 2024, averaged (2024 provisional); ADI 2023; ACS 2019-2023 population | 95% CI 9.6 to 12.1 | 6,846 tracts | ACS 2019-2023 tract population | CDC injury and violence mapping |   |
| alt_degree_rho | +1.00 | rank correlation | Spearman rank correlation between bin order and bin mean, each bin counted once | the bins of tracts ranked by Bachelor's degree share | annual homicide rates for 2022, 2023 and 2024, averaged (2024 provisional); ADI 2023; ACS 2019-2023 population |   |   | each bin counted once | CDC injury and violence mapping |   |
| alt_income_least | 1.7 | Homicide deaths per 100,000 residents a year | population-weighted mean of the tract homicide rate (each tract carries its cluster's rate); 95% intervals from a bootstrap over counties | the least deprived tenth of tracts ranked by Median household income, among the tracts with all four rankings | annual homicide rates for 2022, 2023 and 2024, averaged (2024 provisional); ADI 2023; ACS 2019-2023 population | 95% CI 1.5 to 1.9 | 6,631 tracts | ACS 2019-2023 tract population | CDC injury and violence mapping |   |
| alt_income_most | 14.8 | Homicide deaths per 100,000 residents a year | population-weighted mean of the tract homicide rate (each tract carries its cluster's rate); 95% intervals from a bootstrap over counties | the most deprived tenth of tracts ranked by Median household income, among the tracts with all four rankings | annual homicide rates for 2022, 2023 and 2024, averaged (2024 provisional); ADI 2023; ACS 2019-2023 population | 95% CI 12.9 to 16.9 | 6,825 tracts | ACS 2019-2023 tract population | CDC injury and violence mapping |   |
| alt_income_rho | +1.00 | rank correlation | Spearman rank correlation between bin order and bin mean, each bin counted once | the bins of tracts ranked by Median household income | annual homicide rates for 2022, 2023 and 2024, averaged (2024 provisional); ADI 2023; ACS 2019-2023 population |   |   | each bin counted once | CDC injury and violence mapping |   |
| alt_n_places | 66,773 | tracts | count of tracts in the common sample of the ranking comparison | tracts with the outcome and all four rankings | annual homicide rates for 2022, 2023 and 2024, averaged (2024 provisional); ADI 2023; ACS 2019-2023 population |   |   | none (full count) | CDC injury and violence mapping |   |
| alt_rho_cut | 0.9 | absolute rank correlation | threshold between a steady and an uneven gradient | the rule for changed gradients (docs/RANKING_SWITCH_RULE.md) | annual homicide rates for 2022, 2023 and 2024, averaged (2024 provisional); ADI 2023; ACS 2019-2023 population |   |   | none (a fixed rule) | CDC injury and violence mapping |   |
| alt_rho_margin | 0.1 | absolute rank correlation | smallest move in the rank correlation that counts as a change of shape | the rule for changed gradients (docs/RANKING_SWITCH_RULE.md) | annual homicide rates for 2022, 2023 and 2024, averaged (2024 provisional); ADI 2023; ACS 2019-2023 population |   |   | none (a fixed rule) | CDC injury and violence mapping |   |
| alt_std_least | 1.6 | Homicide deaths per 100,000 residents a year | population-weighted mean of the tract homicide rate (each tract carries its cluster's rate); 95% intervals from a bootstrap over counties | the least deprived tenth of tracts ranked by Standardized index, among the tracts with all four rankings | annual homicide rates for 2022, 2023 and 2024, averaged (2024 provisional); ADI 2023; ACS 2019-2023 population | 95% CI 1.5 to 1.8 | 6,743 tracts | ACS 2019-2023 tract population | CDC injury and violence mapping |   |
| alt_std_most | 15.3 | Homicide deaths per 100,000 residents a year | population-weighted mean of the tract homicide rate (each tract carries its cluster's rate); 95% intervals from a bootstrap over counties | the most deprived tenth of tracts ranked by Standardized index, among the tracts with all four rankings | annual homicide rates for 2022, 2023 and 2024, averaged (2024 provisional); ADI 2023; ACS 2019-2023 population | 95% CI 13.0 to 17.9 | 6,657 tracts | ACS 2019-2023 tract population | CDC injury and violence mapping |   |
| alt_std_rho | +1.00 | rank correlation | Spearman rank correlation between bin order and bin mean, each bin counted once | the bins of tracts ranked by Standardized index | annual homicide rates for 2022, 2023 and 2024, averaged (2024 provisional); ADI 2023; ACS 2019-2023 population |   |   | each bin counted once | CDC injury and violence mapping |   |
| band_cut_ref | 2 | homicide deaths per 100,000 residents a year | upper limit of the lowest group of clusters, the comparison group (definition) | clusters of tracts | annual homicide rates for 2022, 2023 and 2024, averaged (2024 provisional); ADI 2023; ACS 2019-2023 population |   |   | none (definition) | CDC injury and violence mapping |   |
| band_cut_top | 40 | homicide deaths per 100,000 residents a year | cut point between the two highest groups of clusters (definition) | clusters of tracts | annual homicide rates for 2022, 2023 and 2024, averaged (2024 provisional); ADI 2023; ACS 2019-2023 population |   |   | none (definition) | CDC injury and violence mapping |   |
| band_low_share | 4.5% | percent of the difference in life expectancy at birth between the lowest and the second lowest group | part of the difference attributed to homicide by rough accounting over the whole difference, bootstrap interval over counties (draws shared) | residents of tracts in the second lowest group of clusters by homicide rate | homicide rates for 2022 to 2024 with life expectancy for deaths in 2010 to 2015 | 95% CI 3.9% to 5.1% | 19,808 tracts | ACS 2019-2023 tract population | CDC injury and violence mapping |   |
| band_mid_share | 12.4% | percent of the difference in life expectancy at birth between the lowest and the second highest group | part of the difference attributed to homicide by rough accounting over the whole difference, bootstrap interval over counties (draws shared) | residents of tracts in the second highest group of clusters by homicide rate | homicide rates for 2022 to 2024 with life expectancy for deaths in 2010 to 2015 | 95% CI 11.4% to 13.4% | 2,913 tracts | ACS 2019-2023 tract population | CDC injury and violence mapping |   |
| band_ref_le | 80.2 | years of life expectancy at birth | population-weighted mean of tract life expectancy, mean over the three ways of leaving out a year; bootstrap interval over counties | residents of tracts in the lowest group of clusters by homicide rate | deaths in 2010 to 2015 (USALEEP, area-weighted onto 2020 tracts); ADI 2023; ACS 2019-2023 population | 95% CI 80.0 to 80.4 | 26,042 tracts | ACS 2019-2023 tract population | USALEEP life expectancy | The groups differ a little between the three ways of leaving out a year; n is the mean number of tracts in the group over the three. The highest group holds 1,165 to 1,550 tracts, 194 to 264 clusters and 54 to 67 counties, so its interval rests on few counties. |
| band_ref_rate | 1.8 | homicide deaths per 100,000 residents a year | population-weighted mean rate for the year left out of forming the groups, put on the level of the three-year mean, averaged over the three ways of leaving out a year; bootstrap interval over counties | residents of tracts in the lowest group of clusters by homicide rate | annual homicide rates for 2022, 2023 and 2024, averaged (2024 provisional); ADI 2023; ACS 2019-2023 population | 95% CI 1.7 to 1.9 | 26,042 tracts | ACS 2019-2023 tract population | CDC injury and violence mapping | The groups differ a little between the three ways of leaving out a year; n is the mean number of tracts in the group over the three. The highest group holds 1,165 to 1,550 tracts, 194 to 264 clusters and 54 to 67 counties, so its interval rests on few counties. |
| band_top_clusters_max | 264 | tract clusters | clusters in the highest group; the largest over the three ways of leaving out a year | residents of tracts in the highest group of clusters by homicide rate | annual homicide rates for 2022, 2023 and 2024, averaged (2024 provisional); ADI 2023; ACS 2019-2023 population |   |   | none (full count) | CDC injury and violence mapping |   |
| band_top_clusters_min | 194 | tract clusters | clusters in the highest group; the smallest over the three ways of leaving out a year | residents of tracts in the highest group of clusters by homicide rate | annual homicide rates for 2022, 2023 and 2024, averaged (2024 provisional); ADI 2023; ACS 2019-2023 population |   |   | none (full count) | CDC injury and violence mapping |   |
| band_top_cook_max | 15.1% | percent of the residents of the highest group | share of the group's residents who live in Cook County, Illinois; the largest over the three ways of leaving out a year | residents of tracts in the highest group of clusters by homicide rate | annual homicide rates for 2022, 2023 and 2024, averaged (2024 provisional); ADI 2023; ACS 2019-2023 population |   |   | none (full count) | CDC injury and violence mapping |   |
| band_top_cook_min | 12.2% | percent of the residents of the highest group | share of the group's residents who live in Cook County, Illinois; the smallest over the three ways of leaving out a year | residents of tracts in the highest group of clusters by homicide rate | annual homicide rates for 2022, 2023 and 2024, averaged (2024 provisional); ADI 2023; ACS 2019-2023 population |   |   | none (full count) | CDC injury and violence mapping |   |
| band_top_counties_max | 67 | counties and Connecticut planning regions | counties with a tract in the highest group; the largest over the three ways of leaving out a year | residents of tracts in the highest group of clusters by homicide rate | annual homicide rates for 2022, 2023 and 2024, averaged (2024 provisional); ADI 2023; ACS 2019-2023 population |   |   | none (full count) | CDC injury and violence mapping |   |
| band_top_counties_min | 54 | counties and Connecticut planning regions | counties with a tract in the highest group; the smallest over the three ways of leaving out a year | residents of tracts in the highest group of clusters by homicide rate | annual homicide rates for 2022, 2023 and 2024, averaged (2024 provisional); ADI 2023; ACS 2019-2023 population |   |   | none (full count) | CDC injury and violence mapping |   |
| band_top_gap | 8.1 | years of life expectancy at birth below the lowest group of clusters by homicide rate | difference between the population-weighted mean life expectancy of the lowest and the highest group, bootstrap interval over counties (draws shared) | residents of tracts in the highest group of clusters by homicide rate | homicide rates for 2022 to 2024 with life expectancy for deaths in 2010 to 2015 | 95% CI 7.7 to 8.4 | 1,337 tracts | ACS 2019-2023 tract population | CDC injury and violence mapping | The groups differ a little between the three ways of leaving out a year; n is the mean number of tracts in the group over the three. The highest group holds 1,165 to 1,550 tracts, 194 to 264 clusters and 54 to 67 counties, so its interval rests on few counties. |
| band_top_le | 72.1 | years of life expectancy at birth | population-weighted mean of tract life expectancy, mean over the three ways of leaving out a year; bootstrap interval over counties | residents of tracts in the highest group of clusters by homicide rate | deaths in 2010 to 2015 (USALEEP, area-weighted onto 2020 tracts); ADI 2023; ACS 2019-2023 population | 95% CI 71.8 to 72.4 | 1,337 tracts | ACS 2019-2023 tract population | USALEEP life expectancy | The groups differ a little between the three ways of leaving out a year; n is the mean number of tracts in the group over the three. The highest group holds 1,165 to 1,550 tracts, 194 to 264 clusters and 54 to 67 counties, so its interval rests on few counties. |
| band_top_loss | 1.6 | years of life expectancy at birth lost to homicide, by rough accounting | rough accounting: homicide rate x life expectancy at birth x assumed years of life lost per death, tenth of tracts by ADI, bootstrap interval over counties (draws shared with the gap); groups formed by cross-fitting | residents of tracts in the highest group of clusters by homicide rate | homicide rates for 2022 to 2024 with life expectancy for deaths in 2010 to 2015 | 95% CI 1.4 to 1.7 | 1,337 tracts | ACS 2019-2023 tract population | CDC injury and violence mapping | The groups differ a little between the three ways of leaving out a year; n is the mean number of tracts in the group over the three. The highest group holds 1,165 to 1,550 tracts, 194 to 264 clusters and 54 to 67 counties, so its interval rests on few counties. |
| band_top_modeled | 62% | percent of the residents of the highest group | share of the group's residents whose 2024 cluster rate is a CDC modeled estimate (1 to 9 deaths) and not a direct rate | residents of tracts in the highest group of clusters by homicide rate when 2022 and 2023 place the cluster | 2024 (provisional) |   |   | none (full count) | CDC injury and violence mapping |   |
| band_top_pop | 3.6 million | residents | residents in the highest group, mean over the three ways of leaving out a year (cross-fitting) | residents of tracts in the highest group of clusters by homicide rate | annual homicide rates for 2022, 2023 and 2024, averaged (2024 provisional); ADI 2023; ACS 2019-2023 population |   |   | none (full count) | CDC injury and violence mapping |   |
| band_top_pop_share | 1.3% | percent of residents of tracts with a usable homicide rate | share of residents in the highest group, mean over the three ways of leaving out a year | residents of those tracts that sit in a CDC cluster with a published homicide rate in each year from 2022 to 2024, outside New York City | annual homicide rates for 2022, 2023 and 2024, averaged (2024 provisional); ADI 2023; ACS 2019-2023 population |   |   | none (full count) | CDC injury and violence mapping |   |
| band_top_rate | 52.0 | homicide deaths per 100,000 residents a year | population-weighted mean rate for the year left out of forming the groups, put on the level of the three-year mean, averaged over the three ways of leaving out a year; bootstrap interval over counties | residents of tracts in the highest group of clusters by homicide rate | annual homicide rates for 2022, 2023 and 2024, averaged (2024 provisional); ADI 2023; ACS 2019-2023 population | 95% CI 46.9 to 56.1 | 1,337 tracts | ACS 2019-2023 tract population | CDC injury and violence mapping | The groups differ a little between the three ways of leaving out a year; n is the mean number of tracts in the group over the three. The highest group holds 1,165 to 1,550 tracts, 194 to 264 clusters and 54 to 67 counties, so its interval rests on few counties. |
| band_top_rate_naive | 59.5 | homicide deaths per 100,000 residents a year | population-weighted mean of the three-year rate in the group formed on the same three-year rate (not used in the accounting) | residents of tracts in the highest group of clusters by homicide rate | annual homicide rates for 2022, 2023 and 2024, averaged (2024 provisional); ADI 2023; ACS 2019-2023 population |   | 1,358 tracts | ACS 2019-2023 tract population | CDC injury and violence mapping |   |
| band_top_share | 18.5% | percent of the difference in life expectancy at birth between the lowest and the highest group | part of the difference attributed to homicide by rough accounting over the whole difference, bootstrap interval over counties (draws shared) | residents of tracts in the highest group of clusters by homicide rate | homicide rates for 2022 to 2024 with life expectancy for deaths in 2010 to 2015 | 95% CI 16.5% to 20.3% | 1,337 tracts | ACS 2019-2023 tract population | CDC injury and violence mapping | The groups differ a little between the three ways of leaving out a year; n is the mean number of tracts in the group over the three. The highest group holds 1,165 to 1,550 tracts, 194 to 264 clusters and 54 to 67 counties, so its interval rests on few counties. |
| band_top_share_final | 20.0% | percent of the difference in life expectancy at birth between the lowest and the highest group | part of the difference attributed to homicide by rough accounting over the whole difference, one of 2022 and 2023 placing a cluster and the other measuring its rate, the two ways averaged; bootstrap interval over counties (draws shared) | residents of tracts in the highest group of clusters by homicide rate, groups from the 2022 and 2023 rates only | homicide rates for 2022 to 2024 with life expectancy for deaths in 2010 to 2015 | 95% CI 17.9% to 21.9% | 1,576 tracts | ACS 2019-2023 tract population | CDC injury and violence mapping |   |
| band_top_top10_max | 57.1% | percent of the residents of the highest group | share of the group's residents who live in the ten counties with most of them; the largest over the three ways of leaving out a year | residents of tracts in the highest group of clusters by homicide rate | annual homicide rates for 2022, 2023 and 2024, averaged (2024 provisional); ADI 2023; ACS 2019-2023 population |   |   | none (full count) | CDC injury and violence mapping |   |
| band_top_top10_min | 52.7% | percent of the residents of the highest group | share of the group's residents who live in the ten counties with most of them; the smallest over the three ways of leaving out a year | residents of tracts in the highest group of clusters by homicide rate | annual homicide rates for 2022, 2023 and 2024, averaged (2024 provisional); ADI 2023; ACS 2019-2023 population |   |   | none (full count) | CDC injury and violence mapping |   |
| band_top_topn | 10 | counties | number of counties with most residents of the highest group that the concentration shares count (definition) | residents of tracts in the highest group of clusters by homicide rate | annual homicide rates for 2022, 2023 and 2024, averaged (2024 provisional); ADI 2023; ACS 2019-2023 population |   |   | none (full count) | CDC injury and violence mapping |   |
| band_top_tracts_max | 1,550 | tracts | tracts in the highest group; the largest over the three ways of leaving out a year | residents of tracts in the highest group of clusters by homicide rate | annual homicide rates for 2022, 2023 and 2024, averaged (2024 provisional); ADI 2023; ACS 2019-2023 population |   |   | none (full count) | CDC injury and violence mapping |   |
| band_top_tracts_min | 1,165 | tracts | tracts in the highest group; the smallest over the three ways of leaving out a year | residents of tracts in the highest group of clusters by homicide rate | annual homicide rates for 2022, 2023 and 2024, averaged (2024 provisional); ADI 2023; ACS 2019-2023 population |   |   | none (full count) | CDC injury and violence mapping |   |
| band_top_viol | 1.5 | years of life expectancy at birth attributed to homicide by rough accounting, highest group below the lowest | rough accounting: homicide rate x life expectancy at birth x assumed years of life lost per death, tenth of tracts by ADI, bootstrap interval over counties (draws shared with the gap); groups formed by cross-fitting | residents of tracts in the highest group of clusters by homicide rate | homicide rates for 2022 to 2024 with life expectancy for deaths in 2010 to 2015 | 95% CI 1.3 to 1.6 | 1,337 tracts | ACS 2019-2023 tract population | CDC injury and violence mapping | The groups differ a little between the three ways of leaving out a year; n is the mean number of tracts in the group over the three. The highest group holds 1,165 to 1,550 tracts, 194 to 264 clusters and 54 to 67 counties, so its interval rests on few counties. |
| cover_pop | 92.8% | percent of residents of the tracts in the gradient | share of residents who live in a tract of a cluster with a published homicide rate in every year from 2022 to 2024, outside New York City | residents of non-institutional tracts in the 50 states and DC with a life expectancy estimate and an ADI rank | annual homicide rates for 2022, 2023 and 2024, averaged (2024 provisional); ADI 2023; ACS 2019-2023 population |   | 77,936 tracts | ACS 2019-2023 tract population | CDC injury and violence mapping |   |
| dc_gap_years | 17.23 | years of life expectancy at birth, Black men below white men | age-and-cause decomposition of the Black-white gap in male life expectancy at birth, Roberts, Reither and Lim (2020), table 1 | men in Washington, D.C. | deaths in 2015 to 2017 |   |   | none (published estimate) | Published estimates from cited studies |   |
| dc_hom_share | 14.1% | percent of the Black-white gap in male life expectancy at birth | age-and-cause decomposition of the Black-white gap in male life expectancy at birth, Roberts, Reither and Lim (2020), table 1 | men in Washington, D.C. | deaths in 2015 to 2017 |   |   | none (published estimate) | Published estimates from cited studies |   |
| dc_hom_years | 2.43 | years of life expectancy at birth attributed to homicide | age-and-cause decomposition of the Black-white gap in male life expectancy at birth, Roberts, Reither and Lim (2020), table 1 | men in Washington, D.C. | deaths in 2015 to 2017 |   |   | none (published estimate) | Published estimates from cited studies |   |
| fa_share | 75.1% | percent of homicide deaths implied by the cluster rates | firearm homicide deaths implied by the rates (rate x residents) over all homicide deaths implied, tracts with both rates | residents of those tracts that sit in a CDC cluster with a published homicide rate in each year from 2022 to 2024, outside New York City | annual homicide rates for 2022, 2023 and 2024, averaged (2024 provisional); ADI 2023; ACS 2019-2023 population |   | 72,225 tracts | tract residents (implied deaths) | CDC injury and violence mapping |   |
| half_diff_d1 | 0.8 | years of life expectancy at birth lower in the higher-rate group of the tenth | population-weighted mean life expectancy of the lower-rate group minus that of the higher-rate group within the tenth (split at the weighted median rate), mean over the three ways of leaving out a year; bootstrap interval over counties | residents of tracts in the least deprived tenth of tracts with a usable homicide rate | homicide rates for 2022 to 2024 with life expectancy for deaths in 2010 to 2015 | 95% CI 0.5 to 1.1 | 6,479 tracts | ACS 2019-2023 tract population | CDC injury and violence mapping | The tenth has 7,820 tracts; the two groups are formed among the 6,479 with a usable rate, and the life expectancy difference is measured over those tracts. |
| half_diff_d10 | 2.1 | years of life expectancy at birth lower in the higher-rate group of the tenth | population-weighted mean life expectancy of the lower-rate group minus that of the higher-rate group within the tenth (split at the weighted median rate), mean over the three ways of leaving out a year; bootstrap interval over counties | residents of tracts in the most deprived tenth of tracts with a usable homicide rate | homicide rates for 2022 to 2024 with life expectancy for deaths in 2010 to 2015 | 95% CI 1.7 to 2.6 | 7,610 tracts | ACS 2019-2023 tract population | CDC injury and violence mapping | The tenth has 7,844 tracts; the two groups are formed among the 7,610 with a usable rate, and the life expectancy difference is measured over those tracts. |
| half_rate_hi_d10 | 24.6 | homicide deaths per 100,000 residents a year | population-weighted mean rate for the year left out of the split, put on the level of the three-year mean, averaged over the three ways of leaving out a year; bootstrap interval over counties | residents of tracts in the higher-rate group of the most deprived tenth of tracts | annual homicide rates for 2022, 2023 and 2024, averaged (2024 provisional); ADI 2023; ACS 2019-2023 population | 95% CI 21.7 to 27.3 | 7,610 tracts | ACS 2019-2023 tract population | CDC injury and violence mapping | The tenth has 7,844 tracts; the two groups are formed among the 7,610 with a usable rate, and the life expectancy difference is measured over those tracts. |
| half_rate_lo_d10 | 5.0 | homicide deaths per 100,000 residents a year | population-weighted mean rate for the year left out of the split, put on the level of the three-year mean, averaged over the three ways of leaving out a year; bootstrap interval over counties | residents of tracts in the lower-rate group of the most deprived tenth of tracts | annual homicide rates for 2022, 2023 and 2024, averaged (2024 provisional); ADI 2023; ACS 2019-2023 population | 95% CI 4.7 to 5.4 | 7,610 tracts | ACS 2019-2023 tract population | CDC injury and violence mapping | The tenth has 7,844 tracts; the two groups are formed among the 7,610 with a usable rate, and the life expectancy difference is measured over those tracts. |
| half_share_d1 | 6.3% | percent of the difference in life expectancy at birth between the two groups of the tenth | part of the difference attributed to homicide by rough accounting over the whole difference, bootstrap interval over counties (draws shared) | residents of tracts in the least deprived tenth of tracts with a usable homicide rate | homicide rates for 2022 to 2024 with life expectancy for deaths in 2010 to 2015 | 95% CI 4.9% to 8.4% | 6,479 tracts | ACS 2019-2023 tract population | CDC injury and violence mapping | The tenth has 7,820 tracts; the two groups are formed among the 6,479 with a usable rate, and the life expectancy difference is measured over those tracts. |
| half_share_d10 | 27.5% | percent of the difference in life expectancy at birth between the two groups of the tenth | part of the difference attributed to homicide by rough accounting over the whole difference, bootstrap interval over counties (draws shared) | residents of tracts in the most deprived tenth of tracts with a usable homicide rate | homicide rates for 2022 to 2024 with life expectancy for deaths in 2010 to 2015 | 95% CI 22.4% to 33.8% | 7,610 tracts | ACS 2019-2023 tract population | CDC injury and violence mapping | The tenth has 7,844 tracts; the two groups are formed among the 7,610 with a usable rate, and the life expectancy difference is measured over those tracts. |
| half_viol_d10 | 0.6 | years of life expectancy at birth attributed to homicide by rough accounting, higher-rate group below the lower-rate group | rough accounting: homicide rate x life expectancy at birth x assumed years of life lost per death, tenth of tracts by ADI, bootstrap interval over counties (draws shared with the gap); groups formed by cross-fitting | residents of tracts in the most deprived tenth of tracts with a usable homicide rate | homicide rates for 2022 to 2024 with life expectancy for deaths in 2010 to 2015 | 95% CI 0.5 to 0.7 | 7,610 tracts | ACS 2019-2023 tract population | CDC injury and violence mapping | The tenth has 7,844 tracts; the two groups are formed among the 7,610 with a usable rate, and the life expectancy difference is measured over those tracts. |
| hom_d1 | 1.8 | homicide deaths per 100,000 residents a year | population-weighted mean of the tract homicide rate (each tract carries its cluster's rate) in a tenth of tracts by ADI, bootstrap interval over counties | residents of tracts in the least deprived tenth of tracts with a usable homicide rate | annual homicide rates for 2022, 2023 and 2024, averaged (2024 provisional); ADI 2023; ACS 2019-2023 population | 95% CI 1.3 to 2.2 | 6,479 tracts | ACS 2019-2023 tract population | CDC injury and violence mapping |   |
| hom_d10 | 14.8 | homicide deaths per 100,000 residents a year | population-weighted mean of the tract homicide rate (each tract carries its cluster's rate) in a tenth of tracts by ADI, bootstrap interval over counties | residents of tracts in the most deprived tenth of tracts with a usable homicide rate | annual homicide rates for 2022, 2023 and 2024, averaged (2024 provisional); ADI 2023; ACS 2019-2023 population | 95% CI 12.8 to 16.9 | 7,610 tracts | ACS 2019-2023 tract population | CDC injury and violence mapping |   |
| hom_d1_nyc | 2.1 | homicide deaths per 100,000 residents a year | population-weighted mean homicide rate of the least deprived tenth when New York City tracts carry the county file's rate for their borough, bootstrap interval over counties | residents of tracts in the least deprived tenth of tracts, New York City tracts at the county file's rate | annual homicide rates for 2022, 2023 and 2024, averaged (2024 provisional); ADI 2023; ACS 2019-2023 population | 95% CI 1.5 to 2.6 | 7,493 tracts | ACS 2019-2023 tract population | CDC injury and violence mapping |   |
| hom_ratio | 8.4 | ratio of the homicide rate in the most deprived tenth to that in the least deprived | ratio of population-weighted mean homicide rates, most deprived tenth over least deprived, bootstrap interval over counties (draws shared) | residents of the most and the least deprived tenths of tracts with a usable homicide rate | annual homicide rates for 2022, 2023 and 2024, averaged (2024 provisional); ADI 2023; ACS 2019-2023 population | 95% CI 6.5 to 11.3 | 14,089 tracts | ACS 2019-2023 tract population | CDC injury and violence mapping |   |
| hom_step_gap | 5.3 | homicide deaths per 100,000 residents a year, last step minus the next-largest step | the step from the ninth to the most deprived tenth minus the largest other step by point estimate, bootstrap interval over counties (draws shared) | residents of the tenths of tracts with a usable homicide rate | annual homicide rates for 2022, 2023 and 2024, averaged (2024 provisional); ADI 2023; ACS 2019-2023 population | 95% CI 3.6 to 7.0 | 15,035 tracts | ACS 2019-2023 tract population | CDC injury and violence mapping |   |
| hom_step_last | 6.4 | homicide deaths per 100,000 residents a year, most deprived tenth minus ninth tenth | difference between population-weighted mean homicide rates of the most deprived and the ninth tenth of tracts, bootstrap interval over counties (draws shared) | residents of the ninth and the most deprived tenths of tracts with a usable homicide rate | annual homicide rates for 2022, 2023 and 2024, averaged (2024 provisional); ADI 2023; ACS 2019-2023 population | 95% CI 4.8 to 8.0 | 15,035 tracts | ACS 2019-2023 tract population | CDC injury and violence mapping |   |
| hom_step_min | 0.4 | homicide deaths per 100,000 residents a year, smallest step between neighboring tenths among the first nine | difference between population-weighted mean homicide rates of neighboring tenths, bootstrap interval over counties (draws shared); the smallest step by point estimate | residents of two neighboring tenths of tracts with a usable homicide rate | annual homicide rates for 2022, 2023 and 2024, averaged (2024 provisional); ADI 2023; ACS 2019-2023 population | 95% CI 0.1 to 0.7 | 13,899 tracts | ACS 2019-2023 tract population | CDC injury and violence mapping |   |
| le_d1 | 82.3 | years of life expectancy at birth | population-weighted mean of tract life expectancy in a tenth of tracts (rank bins of the ADI; tracts with tied ranks fall in the same tenth), bootstrap interval over counties | residents of tracts in the least deprived tenth of tracts by ADI national percentile | deaths in 2010 to 2015 (USALEEP, area-weighted onto 2020 tracts); ADI 2023; ACS 2019-2023 population | 95% CI 82.0 to 82.6 | 7,820 tracts | ACS 2019-2023 tract population | USALEEP life expectancy |   |
| le_d10 | 74.0 | years of life expectancy at birth | population-weighted mean of tract life expectancy in a tenth of tracts (rank bins of the ADI; tracts with tied ranks fall in the same tenth), bootstrap interval over counties | residents of tracts in the most deprived tenth of tracts by ADI national percentile | deaths in 2010 to 2015 (USALEEP, area-weighted onto 2020 tracts); ADI 2023; ACS 2019-2023 population | 95% CI 73.7 to 74.3 | 7,844 tracts | ACS 2019-2023 tract population | USALEEP life expectancy |   |
| le_decile_drop | 8.3 | years of life expectancy at birth lower in the most deprived tenth of tracts | difference between the population-weighted mean life expectancy of the least and the most deprived tenth of tracts, bootstrap interval over counties (draws shared by the two means) | residents of non-institutional tracts in the 50 states and DC with a life expectancy estimate and an ADI rank | deaths in 2010 to 2015 (USALEEP, area-weighted onto 2020 tracts); ADI 2023; ACS 2019-2023 population | 95% CI 7.9 to 8.7 | 7,844 tracts | ACS 2019-2023 tract population | USALEEP life expectancy |   |
| le_step_last | 1.9 | years of life expectancy at birth, ninth tenth minus tenth tenth of tracts by ADI | difference between the population-weighted mean life expectancy of the ninth tenth and the most deprived tenth, bootstrap interval over counties (draws shared) | residents of non-institutional tracts in the 50 states and DC with a life expectancy estimate and an ADI rank | deaths in 2010 to 2015 (USALEEP, area-weighted onto 2020 tracts); ADI 2023; ACS 2019-2023 population | 95% CI 1.6 to 2.1 | 7,844 tracts | ACS 2019-2023 tract population | USALEEP life expectancy |   |
| n_counties | 3,104 | counties and Connecticut planning regions | count of counties with a tract in the gradient (the units resampled in every bootstrap) | residents of non-institutional tracts in the 50 states and DC with a life expectancy estimate and an ADI rank | deaths in 2010 to 2015 (USALEEP, area-weighted onto 2020 tracts); ADI 2023; ACS 2019-2023 population |   |   | none (full count) | USALEEP life expectancy |   |
| n_gradient_tracts | 77,936 | tracts | count of tracts in the gradient | residents of non-institutional tracts in the 50 states and DC with a life expectancy estimate and an ADI rank | deaths in 2010 to 2015 (USALEEP, area-weighted onto 2020 tracts); ADI 2023; ACS 2019-2023 population |   |   | none (full count) | USALEEP life expectancy |   |
| n_rate_tracts | 72,225 | tracts | count of tracts with a usable homicide rate | residents of those tracts that sit in a CDC cluster with a published homicide rate in each year from 2022 to 2024, outside New York City | annual homicide rates for 2022, 2023 and 2024, averaged (2024 provisional); ADI 2023; ACS 2019-2023 population |   |   | none (full count) | CDC injury and violence mapping |   |
| nyc_county_rate | 4.9 | homicide deaths per 100,000 residents a year | population-weighted mean of the county file's rate (mean of 2022, 2023 and 2024) for the borough of each New York City tract | residents of the New York City tracts in the gradient | annual homicide rates for 2022, 2023 and 2024, averaged (2024 provisional); ADI 2023; ACS 2019-2023 population |   | 2,074 tracts | ACS 2019-2023 tract population | CDC injury and violence mapping |   |
| nyc_pop | 2.6% | percent of residents of the tracts in the gradient | share of residents who live in New York City, whose tract rates are left out | residents of non-institutional tracts in the 50 states and DC with a life expectancy estimate and an ADI rank | annual homicide rates for 2022, 2023 and 2024, averaged (2024 provisional); ADI 2023; ACS 2019-2023 population |   | 77,936 tracts | ACS 2019-2023 tract population | CDC injury and violence mapping |   |
| nyc_tenth1 | 12.1% | percent of residents of the least deprived tenth of tracts | share of the residents of the least deprived tenth who live in New York City, whose tract rates are left out | residents of the least deprived tenth of tracts by ADI national percentile | annual homicide rates for 2022, 2023 and 2024, averaged (2024 provisional); ADI 2023; ACS 2019-2023 population |   | 7,820 tracts | ACS 2019-2023 tract population | CDC injury and violence mapping |   |
| nyc_tract_rate | 0.17 | homicide deaths per 100,000 residents a year | population-weighted mean of the three-year tract-file rate over New York City tracts with a published rate | residents of the New York City tracts in the gradient | annual homicide rates for 2022, 2023 and 2024, averaged (2024 provisional); ADI 2023; ACS 2019-2023 population |   | 2,034 tracts | ACS 2019-2023 tract population | CDC injury and violence mapping |   |
| nyc_zero_share | 75.0% | percent of New York City tracts | share of New York City tracts whose three-year mean rate in the tract file is exactly zero (each tract counted once) | residents of the New York City tracts in the gradient | annual homicide rates for 2022, 2023 and 2024, averaged (2024 provisional); ADI 2023; ACS 2019-2023 population |   | 2,034 tracts | ACS 2019-2023 tract population | CDC injury and violence mapping |   |
| other_gap_share | 95.2% | percent of the gap in life expectancy at birth between the least and the most deprived tenth | the remainder over the whole gap: the share not attributed to homicide | residents of non-institutional tracts in the 50 states and DC with a life expectancy estimate and an ADI rank | homicide rates for 2022 to 2024 with life expectancy for deaths in 2010 to 2015 | 95% CI 94.6% to 95.9% | 14,089 tracts | ACS 2019-2023 tract population | CDC injury and violence mapping | The gap uses all 15,664 tracts of the two end tenths, and the part attributed to homicide uses the 14,089 of them with a published rate; each tenth's rate is applied to the whole tenth. |
| other_gap_years | 7.9 | years of life expectancy at birth, most deprived tenth below the least deprived, from causes other than homicide | the gap in life expectancy less the part attributed to homicide by rough accounting (a remainder, all other causes) | residents of non-institutional tracts in the 50 states and DC with a life expectancy estimate and an ADI rank | homicide rates for 2022 to 2024 with life expectancy for deaths in 2010 to 2015 | 95% CI 7.5 to 8.3 | 14,089 tracts | ACS 2019-2023 tract population | CDC injury and violence mapping | The gap uses all 15,664 tracts of the two end tenths, and the part attributed to homicide uses the 14,089 of them with a published rate; each tenth's rate is applied to the whole tenth. |
| pop_d1 | 32.2 million | residents | sum of tract population in a tenth of tracts (tenths hold equal numbers of tracts, not of residents) | residents of tracts in the least deprived tenth of tracts by ADI national percentile | ACS 2019-2023; ADI 2023 |   | 7,820 tracts | none (full count) | American Community Survey |   |
| pop_d10 | 23.0 million | residents | sum of tract population in a tenth of tracts (tenths hold equal numbers of tracts, not of residents) | residents of tracts in the most deprived tenth of tracts by ADI national percentile | ACS 2019-2023; ADI 2023 |   | 7,844 tracts | none (full count) | American Community Survey |   |
| rate_basis_ratio | 0.99 | ratio of the deaths implied by the rate to CDC's exact count | median over clusters of the annual rate x ACS residents of the cluster / 100,000, divided by CDC's exact count of deaths (a value near one means the rates are crude per-resident rates) | tract clusters outside New York City with an exact 2022 count of deaths at the publication threshold or above | 2022 |   | 229 clusters | none (median over clusters) | CDC injury and violence mapping |   |
| risk_d1 | 0.14% | percent of newborns who would die of homicide, stationary-population approximation | homicide rate per resident per year x life expectancy at birth of the tenth (crude rate times life expectancy equals the lifetime risk in a stationary population), bootstrap interval over counties | residents of tracts in the least deprived tenth of tracts with a usable homicide rate | homicide rates for 2022 to 2024 with life expectancy for deaths in 2010 to 2015 | 95% CI 0.11% to 0.18% | 6,479 tracts | ACS 2019-2023 tract population | CDC injury and violence mapping |   |
| risk_d10 | 1.10% | percent of newborns who would die of homicide, stationary-population approximation | homicide rate per resident per year x life expectancy at birth of the tenth (crude rate times life expectancy equals the lifetime risk in a stationary population), bootstrap interval over counties | residents of tracts in the most deprived tenth of tracts with a usable homicide rate | homicide rates for 2022 to 2024 with life expectancy for deaths in 2010 to 2015 | 95% CI 0.95% to 1.24% | 7,610 tracts | ACS 2019-2023 tract population | CDC injury and violence mapping |   |
| share_2019 | 3.9% | percent of the gap in life expectancy at birth between the least and the most deprived tenth | rough accounting at the published years of life lost, rates scaled to the 2019 level from the county file, bootstrap interval over counties | residents of non-institutional tracts in the 50 states and DC with a life expectancy estimate and an ADI rank | homicide rates for 2022 to 2024 with life expectancy for deaths in 2010 to 2015 | 95% CI 3.3% to 4.5% | 13,779 tracts | ACS 2019-2023 tract population | CDC injury and violence mapping | n counts the tracts of the two end tenths that carry a rate in this setting; the gap in the main setting uses all 15,664 tracts of those tenths. |
| share_agemix | 5.0% | percent of the gap in life expectancy at birth between the least and the most deprived tenth | rough accounting at the published years of life lost, rates scaled to a common share of residents aged 18 to 64, bootstrap interval over counties | residents of non-institutional tracts in the 50 states and DC with a life expectancy estimate and an ADI rank | homicide rates for 2022 to 2024 with life expectancy for deaths in 2010 to 2015 | 95% CI 4.3% to 5.7% | 14,089 tracts | ACS 2019-2023 tract population | CDC injury and violence mapping | n counts the tracts of the two end tenths that carry a rate in this setting; the gap in the main setting uses all 15,664 tracts of those tenths. |
| share_bound | 8.5% | percent of the gap in life expectancy at birth between the least and the most deprived tenth | rough accounting with years of life lost per homicide set at the largest possible setting: the life expectancy of the most deprived tenth, bootstrap interval over counties | residents of non-institutional tracts in the 50 states and DC with a life expectancy estimate and an ADI rank | homicide rates for 2022 to 2024 with life expectancy for deaths in 2010 to 2015 | 95% CI 7.3% to 9.7% | 14,089 tracts | ACS 2019-2023 tract population | CDC injury and violence mapping | The gap uses all 15,664 tracts of the two end tenths, and the part attributed to homicide uses the 14,089 of them with a published rate; each tenth's rate is applied to the whole tenth. |
| share_bound_tenth | 9.8% | percent of the gap in life expectancy at birth between the least and the most deprived tenth | rough accounting with the years lost set at the most deprived tenth's whole life expectancy there and at zero in the least deprived tenth (the largest share any tenth-specific setting could give), bootstrap interval over counties | residents of non-institutional tracts in the 50 states and DC with a life expectancy estimate and an ADI rank | homicide rates for 2022 to 2024 with life expectancy for deaths in 2010 to 2015 | 95% CI 8.6% to 11.0% | 14,089 tracts | ACS 2019-2023 tract population | CDC injury and violence mapping | The gap uses all 15,664 tracts of the two end tenths, and the part attributed to homicide uses the 14,089 of them with a published rate; each tenth's rate is applied to the whole tenth. |
| share_final | 5.1% | percent of the gap in life expectancy at birth between the least and the most deprived tenth | rough accounting at the published years of life lost, rates for 2022 and 2023 only, bootstrap interval over counties | residents of non-institutional tracts in the 50 states and DC with a life expectancy estimate and an ADI rank | homicide rates for 2022 to 2024 with life expectancy for deaths in 2010 to 2015 | 95% CI 4.4% to 5.9% | 14,275 tracts | ACS 2019-2023 tract population | CDC injury and violence mapping | n counts the tracts of the two end tenths that carry a rate in this setting; the gap in the main setting uses all 15,664 tracts of those tenths. |
| share_firearm | 4.1% | percent of the gap in life expectancy at birth between the least and the most deprived tenth | rough accounting at the published years of life lost, firearm homicide only, bootstrap interval over counties | residents of non-institutional tracts in the 50 states and DC with a life expectancy estimate and an ADI rank | homicide rates for 2022 to 2024 with life expectancy for deaths in 2010 to 2015 | 95% CI 3.4% to 4.7% | 14,115 tracts | ACS 2019-2023 tract population | CDC injury and violence mapping | n counts the tracts of the two end tenths that carry a rate in this setting; the gap in the main setting uses all 15,664 tracts of those tenths. |
| share_fsuicide | 6.2% | percent of the gap in life expectancy at birth between the least and the most deprived tenth | rough accounting at the published years of life lost, homicide plus firearm suicide, bootstrap interval over counties | residents of non-institutional tracts in the 50 states and DC with a life expectancy estimate and an ADI rank | homicide rates for 2022 to 2024 with life expectancy for deaths in 2010 to 2015 | 95% CI 5.6% to 6.9% | 10,102 tracts | ACS 2019-2023 tract population | CDC injury and violence mapping | n counts the tracts of the two end tenths that carry a rate in this setting; the gap in the main setting uses all 15,664 tracts of those tenths. |
| share_high | 6.3% | percent of the gap in life expectancy at birth between the least and the most deprived tenth | rough accounting with years of life lost per homicide set at the higher round setting, bootstrap interval over counties | residents of non-institutional tracts in the 50 states and DC with a life expectancy estimate and an ADI rank | homicide rates for 2022 to 2024 with life expectancy for deaths in 2010 to 2015 | 95% CI 5.4% to 7.2% | 14,089 tracts | ACS 2019-2023 tract population | CDC injury and violence mapping | The gap uses all 15,664 tracts of the two end tenths, and the part attributed to homicide uses the 14,089 of them with a published rate; each tenth's rate is applied to the whole tenth. |
| share_low | 3.4% | percent of the gap in life expectancy at birth between the least and the most deprived tenth | rough accounting with years of life lost per homicide set at the lower round setting, bootstrap interval over counties | residents of non-institutional tracts in the 50 states and DC with a life expectancy estimate and an ADI rank | homicide rates for 2022 to 2024 with life expectancy for deaths in 2010 to 2015 | 95% CI 2.9% to 3.9% | 14,089 tracts | ACS 2019-2023 tract population | CDC injury and violence mapping | The gap uses all 15,664 tracts of the two end tenths, and the part attributed to homicide uses the 14,089 of them with a published rate; each tenth's rate is applied to the whole tenth. |
| share_lscale | 4.4% | percent of the gap in life expectancy at birth between the least and the most deprived tenth | rough accounting at the published years of life lost, years lost scaled by each tenth's life expectancy over the national life expectancy, bootstrap interval over counties | residents of non-institutional tracts in the 50 states and DC with a life expectancy estimate and an ADI rank | homicide rates for 2022 to 2024 with life expectancy for deaths in 2010 to 2015 | 95% CI 3.8% to 5.0% | 14,089 tracts | ACS 2019-2023 tract population | CDC injury and violence mapping | n counts the tracts of the two end tenths that carry a rate in this setting; the gap in the main setting uses all 15,664 tracts of those tenths. |
| share_nyc | 4.6% | percent of the gap in life expectancy at birth between the least and the most deprived tenth | rough accounting at the published years of life lost, New York City tracts at the county file's rate for their borough, bootstrap interval over counties | residents of non-institutional tracts in the 50 states and DC with a life expectancy estimate and an ADI rank | homicide rates for 2022 to 2024 with life expectancy for deaths in 2010 to 2015 | 95% CI 4.0% to 5.3% | 15,117 tracts | ACS 2019-2023 tract population | CDC injury and violence mapping | n counts the tracts of the two end tenths that carry a rate in this setting; the gap in the main setting uses all 15,664 tracts of those tenths. |
| share_same_tracts | 4.7% | percent of the gap in life expectancy at birth between the least and the most deprived tenth | rough accounting at the published years of life lost, the gap measured over the tracts that have a rate, bootstrap interval over counties | residents of non-institutional tracts in the 50 states and DC with a life expectancy estimate and an ADI rank | homicide rates for 2022 to 2024 with life expectancy for deaths in 2010 to 2015 | 95% CI 4.0% to 5.3% | 14,089 tracts | ACS 2019-2023 tract population | CDC injury and violence mapping | n counts the tracts of the two end tenths that carry a rate in this setting; the gap in the main setting uses all 15,664 tracts of those tenths. |
| us_homicide_tract | 5.4 | homicide deaths per 100,000 residents a year | population-weighted mean of the tract homicide rate (each tract carries its cluster's rate) over all tracts of the gradient with a rate, bootstrap interval over counties | residents of all tracts with a usable homicide rate, outside New York City | annual homicide rates for 2022, 2023 and 2024, averaged (2024 provisional); ADI 2023; ACS 2019-2023 population | 95% CI 4.9 to 6.0 | 72,225 tracts | ACS 2019-2023 tract population | CDC injury and violence mapping |   |
| variant_max | 6.2% | percent of the gap in life expectancy at birth between the least and the most deprived tenth | largest share over the settings of the table (point estimates) | residents of non-institutional tracts in the 50 states and DC with a life expectancy estimate and an ADI rank | homicide rates for 2022 to 2024 with life expectancy for deaths in 2010 to 2015 |   |   | ACS 2019-2023 tract population | CDC injury and violence mapping |   |
| variant_min | 3.9% | percent of the gap in life expectancy at birth between the least and the most deprived tenth | smallest share over the settings of the table (point estimates) | residents of non-institutional tracts in the 50 states and DC with a life expectancy estimate and an ADI rank | homicide rates for 2022 to 2024 with life expectancy for deaths in 2010 to 2015 |   |   | ACS 2019-2023 tract population | CDC injury and violence mapping |   |
| viol_gap_share | 4.8% | percent of the gap in life expectancy at birth between the least and the most deprived tenth | rough accounting: the part of the gap attributed to homicide over the whole gap, bootstrap interval over counties (draws shared) | residents of non-institutional tracts in the 50 states and DC with a life expectancy estimate and an ADI rank | homicide rates for 2022 to 2024 with life expectancy for deaths in 2010 to 2015 | 95% CI 4.1% to 5.4% | 14,089 tracts | ACS 2019-2023 tract population | CDC injury and violence mapping | The gap uses all 15,664 tracts of the two end tenths, and the part attributed to homicide uses the 14,089 of them with a published rate; each tenth's rate is applied to the whole tenth. |
| viol_gap_years | 0.39 | years of life expectancy at birth, most deprived tenth below the least deprived, attributed to homicide by rough accounting | rough accounting: homicide rate x life expectancy at birth x assumed years of life lost per death, tenth of tracts by ADI, bootstrap interval over counties (draws shared with the gap) | residents of the most and the least deprived tenths of tracts | homicide rates for 2022 to 2024 with life expectancy for deaths in 2010 to 2015 | 95% CI 0.33 to 0.46 | 14,089 tracts | ACS 2019-2023 tract population | CDC injury and violence mapping | The gap uses all 15,664 tracts of the two end tenths, and the part attributed to homicide uses the 14,089 of them with a published rate; each tenth's rate is applied to the whole tenth. |
| viol_loss_all | 0.18 | years of life expectancy at birth lost to homicide, by rough accounting | rough accounting: homicide rate x life expectancy at birth x assumed years of life lost per death, all tracts with a rate, bootstrap interval over counties (draws shared with the gap) | residents of those tracts that sit in a CDC cluster with a published homicide rate in each year from 2022 to 2024, outside New York City | homicide rates for 2022 to 2024 with life expectancy for deaths in 2010 to 2015 | 95% CI 0.16 to 0.20 | 72,225 tracts | ACS 2019-2023 tract population | CDC injury and violence mapping |   |
| viol_loss_d1 | 0.06 | years of life expectancy at birth lost to homicide, by rough accounting | rough accounting: homicide rate x life expectancy at birth x assumed years of life lost per death, tenth of tracts by ADI, bootstrap interval over counties (draws shared with the gap) | residents of tracts in the least deprived tenth of tracts with a usable homicide rate | homicide rates for 2022 to 2024 with life expectancy for deaths in 2010 to 2015 | 95% CI 0.05 to 0.07 | 6,479 tracts | ACS 2019-2023 tract population | CDC injury and violence mapping |   |
| viol_loss_d10 | 0.45 | years of life expectancy at birth lost to homicide, by rough accounting | rough accounting: homicide rate x life expectancy at birth x assumed years of life lost per death, tenth of tracts by ADI, bootstrap interval over counties (draws shared with the gap) | residents of tracts in the most deprived tenth of tracts with a usable homicide rate | homicide rates for 2022 to 2024 with life expectancy for deaths in 2010 to 2015 | 95% CI 0.39 to 0.52 | 7,610 tracts | ACS 2019-2023 tract population | CDC injury and violence mapping |   |
| viol_step_last | 0.19 | years of life expectancy at birth, ninth tenth to most deprived tenth, attributed to homicide by rough accounting | rough accounting: homicide rate x life expectancy at birth x assumed years of life lost per death, tenth of tracts by ADI, bootstrap interval over counties (draws shared with the gap) | residents of the ninth and the most deprived tenths of tracts | homicide rates for 2022 to 2024 with life expectancy for deaths in 2010 to 2015 | 95% CI 0.14 to 0.24 | 15,035 tracts | ACS 2019-2023 tract population | CDC injury and violence mapping | The step in life expectancy uses all tracts of the two tenths; the part attributed to homicide uses the tracts with a published rate. |
| viol_step_share | 10.2% | percent of the step in life expectancy between the ninth and the most deprived tenth | part of the ninth-to-tenth step attributed to homicide by rough accounting over the whole step | residents of the ninth and the most deprived tenths of tracts | homicide rates for 2022 to 2024 with life expectancy for deaths in 2010 to 2015 | 95% CI 8.0% to 12.3% | 15,035 tracts | ACS 2019-2023 tract population | CDC injury and violence mapping |   |
| wa_d1 | 61.4% | percent of residents aged 18 to 64 | population-weighted mean of the tract share of residents aged 18 to 64 (one minus the shares under 18 and 65 and over), bootstrap interval over counties | residents of tracts in the least deprived tenth of tracts | ACS 2019-2023; ADI 2023 | 95% CI 60.1% to 62.3% | 7,820 tracts | ACS 2019-2023 tract population | American Community Survey |   |
| wa_d10 | 58.5% | percent of residents aged 18 to 64 | population-weighted mean of the tract share of residents aged 18 to 64 (one minus the shares under 18 and 65 and over), bootstrap interval over counties | residents of tracts in the most deprived tenth of tracts | ACS 2019-2023; ADI 2023 | 95% CI 58.2% to 58.7% | 7,844 tracts | ACS 2019-2023 tract population | American Community Survey |   |
| withheld_pop | 4.7% | percent of residents of the tracts in the gradient | share of residents outside New York City whose cluster has no published homicide rate in at least one year from 2022 to 2024 | residents of non-institutional tracts in the 50 states and DC with a life expectancy estimate and an ADI rank | annual homicide rates for 2022, 2023 and 2024, averaged (2024 provisional); ADI 2023; ACS 2019-2023 population |   | 77,936 tracts | ACS 2019-2023 tract population | CDC injury and violence mapping |   |
| yll_high | 55 | years of life lost per homicide | higher round setting tried around the published mean (sensitivity) | homicide victims | homicide rates for 2022 to 2024 with life expectancy for deaths in 2010 to 2015 |   |   | none (assumed setting) | CDC injury and violence mapping |   |
| yll_low | 30 | years of life lost per homicide | lower round setting tried around the published mean (sensitivity) | homicide victims | homicide rates for 2022 to 2024 with life expectancy for deaths in 2010 to 2015 |   |   | none (assumed setting) | CDC injury and violence mapping |   |
| yll_setting | 41.5 | years of potential life lost per homicide victim | mean of life expectancy at birth (specific to the victim's race, ethnicity and sex, 2019) minus age at death, from Zimmerman, Fridel and Trovato (2024), table 1; used here as an assumed setting for every tenth of tracts | homicide victims in the National Violent Death Reporting System, 44 states and DC | deaths in 2003 to 2019 |   | 98,617 homicide victims | none (mean over victims) | Published estimates from cited studies |   |
| yll_states | 17 | states | number of states that supply most of the records of the sample (the paper's statement; the other reporting states supply the rest) | states in the National Violent Death Reporting System sample of Zimmerman, Fridel and Trovato (2024) | deaths in 2003 to 2019 |   |   | none (published count) | Published estimates from cited studies |   |
| yll_suicide | 29.5 | years of potential life lost per suicide decedent | mean of life expectancy at birth (specific to the decedent's race, ethnicity and sex, 2019) minus age at death, from Zimmerman, Fridel and Trovato (2024), table 1; used only to add firearm suicide in one check | suicide decedents in the National Violent Death Reporting System, 44 states and DC | deaths in 2003 to 2019 |   | 230,527 suicide decedents | none (mean over decedents) | Published estimates from cited studies |   |

## Definitions

- Census tract: A small area drawn by the Census Bureau to stand in for a neighborhood. Boundaries are redrawn after each decennial census.
- Crude rate: Deaths or births divided by the population, with no adjustment for age. A place with older residents has a higher crude death rate even if health at each age is the same.
- Modeled estimate: A figure produced by a statistical model instead of a direct count or measurement, usually for small areas where direct data are thin, borrowing strength from surveys, neighboring areas or other data. Neighboring areas can share errors, so single extremes call for caution.
- Area Deprivation Index (ADI): A score that ranks neighborhoods by income, education, employment and housing conditions drawn from the American Community Survey. In this book a higher national percentile means a more deprived neighborhood.
- USALEEP: The U.S. Small-area Life Expectancy Estimates Project, a National Center for Health Statistics release of life expectancy at birth for most census tracts. It covers deaths from 2010 to 2015 only and has not been updated.
- Bootstrap interval: A range for an estimate found by recomputing it on many random resamples of the data and reading the range from the results. It shows how much the estimate would vary from sample to sample.

## Methods

This chapter approximates how much of the life-expectancy gap between the most and the least deprived tenths of tracts homicide accounts for. It is a rough accounting and not an age-specific decomposition, because neither source gives age-specific death rates for a tract, and the two cover different periods (life expectancy for deaths in 2010 to 2015, homicide rates for 2022 to 2024) and come from different models. A tract enters only if CDC publishes the homicide rate of its cluster in all three years; New York City is left out, and suicide is not counted. The years of life lost per homicide are an assumption taken from a published study, and the chapter varies it.

### Tracts, periods and homicide rates

The sample, tenths, weights and gap are those of the chapter Place and the length of a life: tracts with a USALEEP life expectancy for deaths in 2010 to 2015 (area-weighted onto 2020 tracts), institutional tracts (group quarters at least 25 percent of residents) dropped, an ADI 2023 national rank, tenths of tracts with equal numbers of tracts, means weighted by ACS 2019-2023 tract population.

The homicide rate is CDC's rate for the cluster of about five tracts that holds the tract, repeated on each member tract; a tract carries the mean of its cluster's annual rates for 2022, 2023 and 2024 and is used only if CDC publishes all three, which 72,225 tracts outside New York City satisfy. New York City is left out because the file records almost no deaths there. Withheld rates are not filled in the main setting; two checks fill them from the published count range.

Explains Table 1 and Table 6.

### How the accounting approximates the share of the gap

The chapter is a rough accounting and not an age-specific decomposition. Neither source gives age-specific death rates for a tract, and the two cover different periods and come from different models, so the part of a life-expectancy gap that homicide accounts for is approximated by the product of three terms: the homicide rate, life expectancy at birth, and the mean years of life a homicide takes.

The first two give the lifetime risk of dying of homicide in a stationary population, where the crude death rate times life expectancy equals the lifetime probability of dying of that cause; the third converts the risk to years. The product is a first-order approximation to the life expectancy that would be gained if the cause were removed, and it leaves out competing risks and the age mix of each tract; it also treats a victim's life expectancy at birth minus age at death, the published measure, as the remaining life expectancy at the age of death. Arriaga's method (1984) needs death rates by age and cause for each group, which neither file provides below the county.

Explains Table 2.

### Years of life lost per homicide are an assumption

The years of life lost per homicide are an assumption: 41.5 years, the mean of life expectancy (specific to race, ethnicity and sex) minus age at death over homicide victims in the National Violent Death Reporting System, 2003 to 2019, from Zimmerman, Fridel and Trovato (2024); the chapter varies it from zero to the life expectancy of the most deprived tenth.

Explains Table 3.

### What counts as a violent death

In this chapter a violent death means a homicide by any means as the CDC injury dashboard defines it (its All_Homicide measure). The National Violent Death Reporting System, the source of the years lost, counts suicide among violent deaths and this chapter does not. Suicides and deaths in legal intervention are not included; one check adds firearm suicide.

### How uncertainty is measured

Intervals are percentile intervals from a bootstrap that draws whole counties (3,104 counties and Connecticut planning regions) with replacement, 2,000 replicates, seed 12345, shared by every statistic so that a share or a difference has one interval; the gap and its interval equal that chapter's.

### Groups formed on the homicide rate and the 2019 check

For groups formed on the homicide rate, each cluster is placed by the mean of two of the three annual rates and the rate in the accounting is the third year's, scaled by one national factor to the level of the three-year mean; the three ways of leaving out a year are averaged. The 2019 check scales each tenth's accounting by the ratio of the county file's 2019 rate to its 2022 to 2024 rate.

Explains Table 5 and Table 4.

### Four rankings of the same places

The comparison follows the book's ranking switch rule, fixed on October 28, 2026 before any of these gradients was computed. The main gradient of the chapter is recomputed under four rankings of the same places, with the outcome, period, weights, bins and intervals unchanged: the published ADI; a standardized index that applies the ADI's coefficients to its measures after standardizing each one across tracts; median household income; and the share of adults 25 and older with a bachelor's degree or more. The three alternatives come from ACS tract estimates for 2019 to 2023, the period of the ADI.

All four rankings use the same places, the 66,773 tracts that have the outcome and all four rankings, so the ADI panel can differ slightly from the chapter's main figure. Tracts enter only if they have every input of the standardized index, which leaves out tracts without a median rent, home value or mortgage cost.

A pattern counts as changed when the worse end switches sides, when the two ends differ under one ranking and not the other, or when the rank correlation of bin order and bin mean falls on the other side of 0.9 and moves by at least 0.1.

Explains Table 7.

## What these data cannot show

The accounting is not an age-specific decomposition. It applies one value of $$L$$ to every tenth, although victims' ages differ by place, and that value is a national mean for victims, most of them from 17 states, specific to each victim's race, ethnicity and sex and floored at zero. It converts a crude rate to a lifetime risk by assuming a stationary population. The share scales with $$L$$, so a reader who believes victims lose more or fewer years should read the answer off Table 3. The intervals around each share treat $$L$$ as known, and the sweep is the answer to that uncertainty. Arriaga's method needs rates by age and cause for each group, and neither file provides them below the county.

The two sources do not describe the same years. Life expectancy is the USALEEP estimate for deaths in 2010 to 2015, area-weighted onto 2020 tracts with standard errors that are good for mapping only, and the homicide rates are for 2022 to 2024. The county file's rates for 2019 are below its rates for 2022 to 2024 in every tenth, and no file here describes 2010 to 2015, so the share for those years could be lower or higher than shown. The rates are CDC's modeled estimates for clusters of neighboring tracts, repeated on each tract, which averages over neighborhoods and lowers measured concentration. That they are crude per-resident rates is inferred from the check against exact counts and not from CDC documentation. The 2024 values are provisional, and the rows that fill withheld rates from CDC's count ranges give bounds and not estimates. The 95% intervals come from a bootstrap that resamples counties, so they describe which counties are in the file and not the error of either model.

The comparisons are of places. A tract's life expectancy and its homicide rate come from different people and different decades, and a difference between a higher-rate and a lower-rate group does not show what homicide does to the people who live there. The accounting counts deaths from homicide and no other consequence of it, so it leaves out any association of violence with other deaths and with health short of death. The Area Deprivation Index understates deprivation in high-cost cities, and New York City, which holds a large share of the least deprived tenth, has published rates near zero that the accounting leaves out. Nothing in this chapter shows what reducing violence would do to life expectancy; Chapter 8.1, Gun homicide and the causes of early death (https://vishalsingh.org/zipcode/gun-homicide-and-causes-of-early-death) reviews evidence on programs aimed at violence.

## References

1. Arriaga, Eduardo E. (1984). Measuring and explaining the change in life expectancies. Demography 21(1): 83-96. https://doi.org/10.2307/2061029
2. Zimmerman, Gregory M.; Fridel, Emma E.; Trovato, Daniel (2024). Disproportionate burden of violence: Explaining racial and ethnic disparities in potential years of life lost among homicide victims, suicide decedents, and homicide-suicide perpetrators. PLOS ONE 19(2): e0297346. https://doi.org/10.1371/journal.pone.0297346
3. Roberts, Max; Reither, Eric N.; Lim, Sojung (2020). Contributors to the black-white life expectancy gap in Washington D.C.. Scientific Reports 10: 13416. https://doi.org/10.1038/s41598-020-70046-6
4. Light, Michael T.; Vachuska, Karl (2024). Increased homicide played a key role in driving Black-White disparities in life expectancy among men during the COVID-19 pandemic. PLOS ONE 19(8): e0308105. https://doi.org/10.1371/journal.pone.0308105

## Provenance

- Pipeline version: 0.1.0
- Build script: book2/chapters/violence-and-the-life-expectancy-gap/build.py (sha256 1a06e4f1501b9a6282fe6d46efe39e8c5fd7a49d408c7e21642113171e360838)
- Built on: 2026-10-09
- Site commit: c2b1ff3309cb4a10833f884da91ac0057a66ad0a

Inputs:

| File | Size (bytes) | SHA-256, first 1 MB | SHA-256, last 1 MB |
| --- | ---: | --- | --- |
| data/analytic/tract_atlas.parquet | 39,659,925 | 29d5e4d943552d54aeb6ba92aadc939d9bd15bdea553c78357dbd516757be627 | 1e10d1ecf722340f23ee2212d4999a90250f85b65fa89b39e32db06d9472e22e |
| data/geo/counties_2020.parquet | 123,034,155 | 2fce837cdc2db1f2c229f2ea453c6793338a462770307e2c64e8d1fd713ec663 | dae928bf1042f3d174d2d64d6e3ec923fa3daa40aee1bb4feb104aac2fe3c04a |
| data/geo/dim_geo.parquet | 1,180,921 | 29d5e4d943552d54aeb6ba92aadc939d9bd15bdea553c78357dbd516757be627 | e58c8c488df54ab3187f5461f179322afdc49c9bd2315404003fe1272eea32c3 |
| data/raw/acs/deprivation_components_2019_2023/tract.parquet | 13,197,793 | 5545cde2cbe96a8b56e468d65388248a241aa6bd300d3c9e325d08d3f86b11da | 5c8ec454d066988b0f1a01aaa6ed34ebfc05d162e01cf9c6bb01f9198cbaff12 |
| data/staged/adi/adi_atlas_tract2020_2023.parquet | 798,078 | 5e409bc212ff871def780fa27b4cf8b616d9da420b2b9f6455517eddcd8c1a95 | 5e409bc212ff871def780fa27b4cf8b616d9da420b2b9f6455517eddcd8c1a95 |
| data/staged/life_expectancy/usaleep_tract2020.parquet | 1,316,169 | 9c9c7f103791b75eddba74e68a89804efdf2486b6b66bb4bc02a46ba5bf8bc23 | 27317107be2465d9fd5e83271dea0cd6e915ee419e2802bedc28747d0a41d38f |
| data/staged/mortality/injury_violence_county.parquet | 722,427 | d0dfd060a3ab9eaa3a87b35d1b24b5a1e9f21af3513c10367edb651c46f40403 | d0dfd060a3ab9eaa3a87b35d1b24b5a1e9f21af3513c10367edb651c46f40403 |
| data/staged/mortality/injury_violence_tract2020.parquet | 8,059,686 | 48f2a99135501cb7c8c7af7c85361808bd998eb4ab06825759340a1befc5f4f1 | e2e5d28442025676279f349563e0eef9235b013d40007e081ba9671697d04b49 |
| data/staged/ses/acs5_tract2020_2019_2023.parquet | 7,232,424 | 5fe6f730a35225f14d5d8abfc6b9bce58b52e545f309a7eac74f1dae74d7a31c | 7b4a2475b3852d3f91804716a04566ef8ceadacec2be9fb85f5eb642078efac6 |
| data/staged/svi/svi_tract2020_2022.parquet | 17,740,415 | cb3860d8edc2aaf1f2a02211a58f7a0b80457eb6ad9a37b5ec8abb2d5da9da10 | e371a4a696d5bed05d54ad7c5b6ca2504508805b847ec415c2322be7f66ec2b2 |

Outputs:

| File | SHA-256 |
| --- | --- |
| facts.json | 4e4801a65b00f2df89da49274a1b39684d11d8a0ddd7a43db764c7e090b9ebde |
| figures/alt_rankings.json | 6bd94f9150a3728846137c9bfb01005366e66051f66f74f12d70780c2b6e1986 |
| figures/gap_by_tenth.json | e6b0c1b66fbc252b1da010c58cde39a8ea20e0741ded55f634938b4fc1f8d4e7 |
| figures/gap_by_violence_band.json | 00d5c34f399a4fcf8c87fb9ceb8a5df94a13d11e557303e11894481c2200f97a |
| figures/homicide_by_tenth.json | 70f57e2d7ce7c5f1c7ec3c733a50b8e094d8a482f41e5bd3958c36d192f3ca47 |
| figures/same_deprivation.json | dc8831e5bf215727f5e2a4155c727e975d606b0b9b90d6a4702eec85b8817c06 |
| figures/setting_table.json | 1f7839a7cb80f6f645fceb7944186d3f4f3c283bdb17c4a4aa471b2c6bde4470 |
| figures/years_lost_sweep.json | 872d92d4e8c5106802bf384cfafed29377fa236e38b6abcc3e3ed93c48951e74 |
| meta.json | 5823f0444142f8328fe76539c2ed863a80902ddc1daa57c5ffa7191d4f1109ba |
| refs.json | e86a987f426691128511d6d87fc4e32e43c80b1dd444ce04c152fc9b1c15119f |

## Machine formats

- Report (HTML): https://vishalsingh.org/zipcode/violence-and-the-life-expectancy-gap/report
- Report (Markdown): https://vishalsingh.org/zipcode/violence-and-the-life-expectancy-gap/report.md
- Every estimate (JSON): https://vishalsingh.org/zipcode/report/violence-and-the-life-expectancy-gap/facts.json
- Table 1, figure "homicide_by_tenth" (JSON): https://vishalsingh.org/zipcode/report/violence-and-the-life-expectancy-gap/figures/homicide_by_tenth.json
- Table 2, figure "gap_by_tenth" (JSON): https://vishalsingh.org/zipcode/report/violence-and-the-life-expectancy-gap/figures/gap_by_tenth.json
- Table 3, figure "years_lost_sweep" (JSON): https://vishalsingh.org/zipcode/report/violence-and-the-life-expectancy-gap/figures/years_lost_sweep.json
- Table 4, figure "setting_table" (JSON): https://vishalsingh.org/zipcode/report/violence-and-the-life-expectancy-gap/figures/setting_table.json
- Table 5, figure "gap_by_violence_band" (JSON): https://vishalsingh.org/zipcode/report/violence-and-the-life-expectancy-gap/figures/gap_by_violence_band.json
- Table 6, figure "same_deprivation" (JSON): https://vishalsingh.org/zipcode/report/violence-and-the-life-expectancy-gap/figures/same_deprivation.json
- Table 7, figure "alt_rankings" (JSON): https://vishalsingh.org/zipcode/report/violence-and-the-life-expectancy-gap/figures/alt_rankings.json
- Index of every report: https://vishalsingh.org/zipcode/llms.txt; all reports in one file: https://vishalsingh.org/zipcode/llms-full.txt