Chapter 8.5 · The lost years

Violence and the life-expectancy gap

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.

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

Life expectancy at birth, for deaths in 2010 to 2015, is 82.3 years in the least deprived tenth of American census tracts and 74.0 years in the most deprived. Homicide rates for the same tenths, measured in 2022 to 2024, run from 1.8 deaths per hundred thousand residents a year in the least deprived tenth to 14.8 in the most deprived. This chapter asks how much of the 8.3 years between the two ends of the ranking those homicide rates could account for.

Homicide's part of the gap, by a rough accounting

A rough accounting that assumes each homicide takes 41.5 years of life attributes 4.8% (95% CI 4.1% to 5.4%) of that gap to homicide. The other 95.2% is left to every other cause of death, among them the overdose and suicide deaths in Chapter 8.2, Overdose, alcohol and suicide. The accounting gives the years of life homicide takes and says nothing about which of the other causes of death account for the rest.

The unit is the tract. Tracts are ranked on the Area Deprivation Index, which is built from census measures of income, education, employment and housing, and cut into tenths that hold equal numbers of tracts, not of residents. Each tenth's figures are means weighted by residents, so they describe groups of places and not any one person. Both the observed gap and the part attributed to homicide rise with deprivation, and the homicide part is concentrated at the top of the ranking.

Figure 1

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

Years of life expectancy below the least deprived tenth

Show data

Data for 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
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)

Both lines rise with deprivation, and the homicide line stays below the observed line and takes its largest step at the most deprived tenth. The observed line is each tenth's life expectancy below that of the least deprived tenth, using every tract of the tenth. The homicide line is the part the rough accounting attributes to homicide, using the tracts with a published rate, with bands that are 95% intervals from resampling whole counties. Values in Table 2.

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

How we know

Life expectancy is the USALEEP estimate for deaths in 2010 to 2015, for the 77,936 tracts ranked in Chapter 8.4, Place and the length of a life, whose tracts and gap this chapter reuses. The homicide rates come from the CDC injury and violence dashboard, which publishes one rate for each cluster of neighboring tracts and repeats it on every tract of the cluster. The rate is a modeled estimate for clusters with a few deaths, and a tract carries the mean of its cluster's rates for 2022, 2023 and 2024, with 2024 provisional. The chapter counts homicide by any means, as the dashboard defines it, and does not count suicide. New York City is left out, because its tract rates are near zero. Neither source gives death rates by age for a tract, so the chapter cannot make an age-specific decomposition of the kind Arriaga described.[1] It multiplies each tenth's crude homicide rate by its life expectancy. In a population with constant births and constant death rates by age, that product is a newborn's lifetime risk of dying of homicide. The risk is then multiplied by the assumed years per death, the mean for homicide victims in the National Violent Death Reporting System between 2003 and 2019.[2] 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 report gives every estimate with its interval and sample size.

Because the accounting multiplies by the years each homicide takes, the share rises in a straight line with that setting (Figure 2). A death cannot take more years than a person has. The largest value that one setting shared by every tenth could take is the whole life expectancy of the most deprived tenth, which would mean every homicide there was a death at birth. That value gives 8.5% (95% CI 7.3% to 9.7%), and no single value shared by every tenth puts the share above it. A reader who believes victims lose more or fewer years than the published mean can read the answer off the line.

Figure 2

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

Share of the gap attributed to homicide (%)

Show data

Data for 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
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

The share rises in proportion to the years assumed lost per death. The line is the share of the gap between the least and the most deprived tenths that the rough accounting attributes to homicide, with a 95% interval from resampling whole counties. One rule marks the published mean, and the rule on the right marks the largest value a death could take, the life expectancy of the most deprived tenth. Values in Table 3.

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

The other settings the chapter tests move the share less than the years per death do, and across them it ranges from 3.9% to 6.2%. The two lowest shares come from scaling each tenth's accounting to the county file's 2019 rates and from counting only firearm homicide, and the two cannot be told apart. The highest comes from adding firearm suicide. Neither file covers 2010 to 2015, the years behind the life expectancy estimates, and the county file reaches back only to 2019. Its 2019 rates are below its rates for 2022 to 2024 in every tenth, so the share for the years of the life expectancy estimates could be lower or higher than shown.

The tenths average over places whose homicide rates differ. Grouping clusters by their own homicide rate shows the other end. Each cluster is placed in a group by two of its three annual rates and measured by the third. If the annual errors in a cluster's rate are independent, this keeps chance from raising the measured rate of a group chosen for its high rate. In the group with the highest rates, life expectancy is 8.1 (95% CI 7.7 to 8.4) years lower than in the group with the lowest, and homicide accounts for 18.5% (95% CI 16.5% to 20.3%) of that difference. The share is higher in each higher group (Figure 3).

Figure 3

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

Years of life expectancy below the lowest-rate group

Homicide (rough accounting)
All other causes (remainder)

Homicide rate of the tract cluster, deaths per 100,000 residents a year, 2022 to 2024

Show data

Data for 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
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

The difference grows with the homicide rate, and the homicide part grows faster, so it is a larger share of the difference in each higher group. Each bar is the observed difference in life expectancy, for deaths in 2010 to 2015, from the lowest-rate group of clusters. The lower segment is the part the rough accounting attributes to homicide and the upper segment is the remainder. Values in Table 5.

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

The highest group is a narrow slice. It holds 1.3% of the residents of tracts with a rate, and it spans between 54 and 67 counties, depending on which year is left out when the groups are formed.

Within tenths, life expectancy is lower where homicide is 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, and the accounting leaves most of each difference to other causes of death. Tracts of the same rank can differ in conditions the index does not measure, and homicide may mark those conditions. The pattern also fits homicide being tied to other deaths, for example among residents who live with it. The data cannot separate the two. A tract's life expectancy and its homicide rate come from different people and different decades. A difference between a higher-rate and a lower-rate group does not show what homicide does to the people who live there.

What this accounting cannot tell you

The accounting applies one national value of years lost to every tenth, although victims' ages differ by place, and its intervals treat that value as known. The intervals come from resampling counties, so they describe which counties are in the file and not the error of either model. The Area Deprivation Index understates deprivation in high-cost cities, and New York City, which holds 12.1% of the least deprived tenth's residents, is left out of the rates. Giving its tracts the county file's rate for their borough moves the share of the gap to 4.6% (95% CI 4.0% to 5.3%). Nothing here shows what reducing violence would do to life expectancy, and Chapter 8.1, Gun homicide and the causes of early death reviews evidence on programs aimed at violence.

Go deeper

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

Data sources

CDC injury and violence mappingCDC 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 checkYears: 2019-2024 for counties, 2022-2024 for tracts (data as of 2026-05-14) · Weights: ACS 2019-2023 residents of each tract · License: Public domain (CDC, U.S. Department of Health and Human Services)
USALEEP life expectancyLife expectancy at birth by tract, deaths in 2010 to 2015, area-weighted onto 2020 tractsYears: 2010-2015 · Weights: 2019-2023 tract population · License: Public domain (NCHS, CDC)
Area Deprivation IndexArea Deprivation Index national rank, 2023, for the tenths of tractsYears: 2023 release (version 4.0.1; built on ACS 2019-2023) · Weights: none (rank) · License: Free for non-commercial use; cite the Neighborhood Atlas and Kind and Buckingham
American Community SurveyTract population (weights) and age shares, 2019-2023Years: 2019-2023 (5-year) · Weights: none (full count) · License: Public domain (U.S. Census Bureau)
Published estimates from cited studiesMean 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) · License: Quoted from published research for commentary

All data sources

References

Arriaga, Eduardo E. (1984). Measuring and explaining the change in life expectancies. Demography 21(1): 83-96. doi:10.2307/2061029
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. doi:10.1371/journal.pone.0297346

How to cite

@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.}
}

Homicide's part of the gap, by a rough accounting

How we know

The share moves with the assumed years per death

The share is larger where homicide rates are highest

Within tenths, life expectancy is lower where homicide is higher

What this accounting cannot tell you

Go deeper

Data sources

References

How to cite