US Murder (Homicide) Rate: What Drives It?
The honest short answer is that county data cannot prove what produces a murder rate. What it can do is test the usual suspects against each other. It compares places that are alike in the other factors and sees which links are still there. Across the 1,144 counties where the homicide rate is published, home to about 292 million people, homicide is higher in counties with more overdose deaths, more children growing up with one parent, more income inequality, more access to guns, a younger population and less friendship across class lines. It is also higher, by a large amount even after all of those, in counties where more of the residents are Black. What that means is the hardest question on this page. These links hold in every version of the analysis below, but none of them is proof of a cause. Split by race, low income goes with more homicide in every group. Black, white and Hispanic residents are all killed at higher rates in counties where their own group is poorer. Followed over the years, a county's homicide rate also tends to fall when its own residents' incomes rise (−12% for 10% more income). Since that compares each county only with itself, nothing fixed about a place, such as its history, can explain it away.
The measure
The homicide rate here counts deaths recorded as homicide on death certificates, per 100,000 residents, for the years 2016–2022 combined (County Health Rankings, from the National Center for Health Statistics). Most homicides are murders, but the count also includes killings that no court would call murder. A rate is published only where there were enough deaths to estimate one. That is why the analysis covers 1,144 of about 3,100 counties. The smallest and most rural counties are missing, but the places where most Americans live are included. Every other measure is from 2023, the latest year for most.
How to read the numbers
Every figure below is a percentage difference in the homicide rate. Each one compares an average county with a county that is higher on the factor than about 5 in 6 counties. (That is a typical gap between counties, which statisticians call one standard deviation.) So +20% would mean the higher county tends to have a homicide rate 20% above the average one.
Each figure comes with a likely range, where the true value probably lies (statisticians call it a 95% interval). If the range crosses zero, as in −3% to +5%, the link could easily be chance.
Many figures are “after adjusting for” other factors. That means comparing counties that are alike in those factors, so differences in them cannot explain the link. Some comparisons also match each county only against other counties in the same state.
What goes with it, one at a time
First, each factor on its own. Then each factor after adjusting for the three things that most set American counties apart: population density, household income and the share of adults with a college degree. The second column is the one the rest of this site ranks by.
| Factor | On its own | After adjusting for density, income and education |
|---|---|---|
| Children in single-parent households | +80% | +59% |
| Share of residents who are Black | +72% | +52% |
| Poverty rate | +69% | +45% |
| Friendships across income lines | −42% | −34% |
| Empty homes † | +40% | +13% |
| Income inequality | +38% | +23% |
| Gun access (share of suicides by gun) | +36% | +16% |
| Adults with a college degree | −33% | −5% (could be chance) |
| People in jail or prison † | +28% | +8% |
| Drug overdose deaths | +22% | +7% (could be chance) |
| Population density | −22% | +8% (could be chance) |
| Unemployment rate | +21% | +4% (could be chance) |
| Black-white segregation | −20% | −16% |
| Homeownership rate † | −15% | −14% |
| Clubs, churches and civic groups per person | +14% | −1% (could be chance) |
| Median age | −7% | −8% |
| Share of residents who are Hispanic | −6% | −0% (could be chance) |
† Homicide may shape this measure as much as the other way round: people in jail or prison (prisons fill where violence is high); empty homes (people leave violent neighbourhoods); homeownership rate (buyers avoid violent neighbourhoods).
The race question
The share of residents who are Black is one of the two strongest single links to a county’s homicide rate: +74% on its own. For decades, criminologists (researchers who study crime) have argued that this is not about race. They say it is about the conditions Black Americans have far more often been forced to live in: concentrated poverty, joblessness, broken-up families and segregation. We can test that here. The chart adds those conditions one group at a time, using the same 1,106 counties throughout.
Adjusting for density, income and education takes it from +74% to +53%. Adding poverty, unemployment and single-parent households takes it to +37%. Segregation and inequality change almost nothing more (+36%). Comparing counties only with others in their own state leaves +45% (likely range +36% to +54%). That is about 67% of the original link, measured in percentage terms. So the measured conditions account for a real part of the link, but a large part is left.
What is left says nothing about individual people. County averages are rough. The hardship that research links to violence is packed into particular neighbourhoods, and a county-wide poverty rate cannot see it. In Chicago, researchers could measure neighbourhoods directly. There, family structure, neighbourhood conditions and the share of immigrants together accounted for about 60% of the Black–white gap in violence (Sampson, Morenoff & Raudenbush, 2005). The data here lacks other things that might explain the rest: the history of housing and lending policy, how easy guns are to get, and policing and trust in it. Also, roughly half of US homicide victims are Black, so the rate itself is higher where the people most at risk live. The data cannot choose between these explanations.
Does low income go with homicide in every group?
The analysis above compares counties with more and fewer Black residents. A sharper test splits both sides by race. County Health Rankings publishes a separate homicide rate for each group’s residents, by the victim’s race. The Census publishes each group’s own poverty rate and median (middle) income. So we can ask: does a group’s own poverty go with that group’s own homicide rate? As before, each county is compared only with others in its state.
| Residents who are | Counties | Typical homicide rate (median) | Average poverty rate | Change when the group’s poverty is 10 points higher | Extra deaths per 100,000, at the typical rate |
|---|---|---|---|---|---|
| Black | 679 | 25.3 per 100,000 | 25% | +33% (+25% to +41%) | +8.3 |
| White | 980 | 3.7 per 100,000 | 11% | +99% (+77% to +124%) | +3.6 |
| Hispanic | 288 | 6.4 per 100,000 | 18% | +57% (+37% to +81%) | +3.7 |
Yes, in every group. Where a group is poorer, its members are killed at higher rates. In all three groups the likely range is well above zero. The percentages differ partly because the starting rates differ, so the last column shows each one in deaths.
Black and white residents of the same county
501 counties publish both rates. In them, Black residents are killed at about 7.1 times the rate of white residents of the same county. (The typical rates are 23.6 and 3.4per 100,000.) Adjusting for each group’s own income and poverty shrinks the gap:
| Adjusted for (same county) | Black rate as a multiple of the white rate |
|---|---|
| No adjustment | 7.1 times (5.6 to 8.9) |
| Adjusted for each group's own poverty rate | 5.1 times (4.0 to 6.5) |
| Adjusted for each group's own typical income | 3.9 times (3.1 to 4.8) |
| Adjusted for both | 3.9 times (3.2 to 4.9) |
Each group’s median income accounts for about 31% of the gap, measured in percentage terms. Poverty rates account for less. Most of the gap remains. Two known gaps that the data here cannot close are likely part of the rest. First, income is not wealth, and at the same income Black households own far less wealth. Second, a county-wide figure for a group cannot see that poor Black families live in areas of concentrated poverty far more often than poor white families do. Whatever the rest is, county averages of income cannot adjust for it.
Everything at once
Next, every factor goes into one calculation together, so each is judged after adjusting for all the others. Counties are compared only with others in the same state. 879 counties report every measure. Together the factors account for 70% of the differences in homicide rates between counties in the same state. Dark bars are unlikely to be chance; faint ones could be.
Still standing: share of residents who are black (+37%); drug overdose deaths (+15%); children in single-parent households (+8%); income inequality (+7%); gun access (share of suicides by gun) (+5%) go with higher rates; median age (−7%); poverty rate (−8%); adults with a college degree (−10%); friendships across income lines (−21%) go with lower ones.
- Families. The share of children living with one parent is one of the strongest single links (+80% on its own). It is still there after adjusting for everything else (+8%). It is also a sign of poverty, prison and early death among parents. So it sums up a place as much as it describes households.
- Who knows whom.Economic connectedness measures how many of a low-income person’s friends have higher incomes, using Facebook friendships (Chetty et al., 2022). It goes with −21% homicide, after adjusting for poverty, income and the rest. That is one of the largest links to lower homicide here.
- Age. Most killers and most victims are young men. Older counties have less homicide (−7%, comparing an average county with one older than about 5 in 6).
- Drugs and inequality. Overdose deaths (+15%) and income inequality (+7%) both keep a link of their own. Drug markets and violence have long been linked. Of all the broad features of a society, inequality is the one most consistently linked to homicide, across countries as well as counties.
Over time, within the same county
Every comparison so far is between places. Something fixed about a place that was not measured, such as its history, its gangs or its policing traditions, could still be behind any of them. Following the same counties over the years rules all of that out at once. Each county is compared only with itself, and each year only with the same year elsewhere. So a national wave, such as the jump in killings in 2020, counts the same in every county. Here we follow the 187 counties with a homicide rate published every year from 2010 to 2023. These are the large counties, where most homicides happen. We compare them against the three conditions that are measured every year.
Homicides per 100,000 residents in those counties, each year (CDC WONDER).
| When a county’s… | Change in homicide, same year | Change in homicide, next year |
|---|---|---|
| unemployment rate rises, per point | +2% (−1% to +4%) | +3% (+0% to +6%) |
| poverty rate rises, per point | −2% (−3% to +0%) | −2% (−4% to −1%) |
| median household income rises, per 10% higher | −12% (−17% to −6%) | −9% (−14% to −4%) |
Within the same county: when unemployment rises a point, homicide is +3% the following year, when the poverty rate rises a point, homicide is −2% the following year and when median household income is 10% higher, homicide is −12% that year and −9% the next. The poverty figure's direction is not a real finding. Within a county, the Census Bureau's poverty rate (an estimate from a statistical model, not a direct count) and median income move almost in step. Once income is in the calculation, what is left of poverty is just noise around it. Read the two together as one thing: how well off the county's residents are that year. Grey figures could easily be chance. The likely ranges allow for the fact that counties in the same state tend to resemble each other.
Guns
No agency counts guns or gun owners by county. The usual stand-in is the share of suicides that involve a gun. A suicide attempt is far more often fatal with a gun at hand. So where more homes hold guns, more suicides involve guns. Men, older people and white Americans use guns in suicide more often than others. So each county’s share is compared with the share its own mix of sex, age and race would give at national rates (the index). Across states it agrees with RAND’s survey-based estimates of household gun ownership at r = 0.86. (Here r measures how closely two measures rise and fall together, from −1 for opposite to +1 for in step; 0 means no link.)
Counties. This compares counties within the same state and adjusts for every other factor above. Gun access goes with +5% homicide (likely range +1% to +9%). It holds in every version of the analysis. The link is smaller than those for family structure, connectedness or overdoses, but it is real.
States. Across the 50states, RAND’s household gun ownership (averaged over 2012–2016) goes with +14% homicide on its own (−1% to +33%). After adjusting for share of residents who are black, poverty, population density, it is −2% (−19% to +18%). The suicide-based index gives +24% and +9%. So at the state level, legal household gun ownership says little about homicide once you know who lives there and how poor they are.
Both measures capture guns in homes that people will report, or that show up in suicides. Neither captures the illegal market, where many of the guns used in homicides come from. The closest public view of it is the ATF’s count of crime guns recovered by police and traced, by state, for 2025. (The ATF is the federal agency that traces a gun to find where it was first sold.) This count matches legal household ownership only loosely (r = 0.33). But it matches homicide closely. Across the 51 states and DC, r = 0.80 with the 2022–2024 homicide rate. That is +61% on its own and +26% (+14% to +40%) after adjusting for the same three factors.
Read that last figure with care, since it partly counts homicide twice. Guns recovered from homicides are among the guns traced. So a state with more killings recovers and traces more guns. The count also rises with how thoroughly police trace what they find. What it does show is that the guns used in crime are a different thing from how many households own one.
A measure that avoids this double counting is how soon crime guns turn up after they are first sold. The ATF sees a short time from store sale to crime as a sign of trafficking and straw purchasing (one person buying a gun for someone else who may not be allowed to). The share of traced guns recovered within a year of sale is a share, not a count, so more killings do not raise it by themselves. It goes with +26% homicide on its own (r = 0.39), but +5% (−5% to +15%) after adjusting for who lives in a state and how poor it is. That could easily be chance. It also rises with legal ownership (r = 0.48): where more households own guns, crime guns tend to turn up sooner after they are sold. So the one gun measure that does not partly count homicide twice loses its link once these conditions are adjusted for. That suggests much of the trace count's link to homicide is the double counting.
Where crime guns come from tells a related story. In states where few households own guns, most crime guns were first sold in another state. That is about four in five in New York and New Jersey, against one in seven in Texas. This share runs strongly opposite to legal ownership (r = -0.68). States where few households own guns get most of their crime guns from other states. But how much of the supply comes from other states does not go with how much homicide a state has. The link is −9% on its own and −6% after adjusting for the three, and both could easily be chance.
What does not survive
- Population density: +5% (−0% to +11%). Cities are not more violent than the conditions in them would suggest. Density on its own goes with lower rates, and after adjusting for the rest it adds nothing.
- People in jail or prison: +3% (−2% to +7%). More incarceration goes with more homicide on its own. That is what you would expect whichever way the link runs. After adjusting for everything else, the link could easily be chance.
- Clubs, churches and civic groups per person: +2% (−2% to +6%). This counts clubs, churches and civic groups per person, a classic measure of community ties. It does not track homicide once economic connectedness is known.
- Empty homes: +2% (−2% to +6%).
- Unemployment rate: +2% (−1% to +5%). Joblessness goes with homicide on its own, but not after adjusting for the family and friendship measures.
- Homeownership rate: +1% (−3% to +5%).
- Share of residents who are Hispanic: −2% (−7% to +3%). Heavily Hispanic counties have homicide rates like any other county with the same conditions. Researchers call this the immigrant or Hispanic paradox.
- Black-white segregation, measured across each county’s neighbourhoods: +3% (−0% to +6%) after adjusting for everything else, in the 866 counties where it can be measured. On its own it goes with lower homicide, the opposite of what research on cities would suggest. The county measure mixes very different places, from segregated big cities to small towns with a single Black neighbourhood. It is not the same thing as concentrated poverty.
A caution on the hardship measures. These measures overlap heavily: poverty rate, children in single-parent households, adults with a college degree, friendships across income lines, homeownership rate. The calculation struggles to tell them apart. That is why poverty shows a negative sign once the others are adjusted for (−8%). This does not mean poverty protects anyone. It means its link is carried by the measures it moves with. Read that group as one thing: concentrated hardship.
How sure is this?
We ran four versions of the analysis. One uses all counties together. One compares counties only within their own state. One does that but gives bigger counties more weight. One adds segregation. Share of residents who are Black, Drug overdose deaths, Children in single-parent households, Income inequality, Gun access (share of suicides by gun), Median age and Friendships across income lines are unlikely to be chance in every one. Poverty rate and Adults with a college degree are clear in the main version but not in every one.
| Factor | All counties together | Within states | Bigger counties count more | With segregation |
|---|---|---|---|---|
| Share of residents who are Black | +27% | +37% | +43% | +38% |
| Drug overdose deaths | +14% | +15% | +21% | +15% |
| Children in single-parent households | +14% | +8% | +17% | +9% |
| Income inequality | +6% | +7% | +13% | +6% |
| Population density | −4% | +5% | +3% | +5% |
| Gun access (share of suicides by gun) | +11% | +5% | +11% | +5% |
| People in jail or prison | +5% | +3% | +8% | +3% |
| Clubs, churches and civic groups per person | +2% | +2% | −0% | +3% |
| Empty homes | +0% | +2% | +0% | +1% |
| Unemployment rate | +2% | +2% | −0% | +2% |
| Homeownership rate | −1% | +1% | +1% | +2% |
| Share of residents who are Hispanic | +5% | −2% | −0% | −3% |
| Median age | −8% | −7% | −9% | −8% |
| Poverty rate | −5% | −8% | −14% | −7% |
| Adults with a college degree | −6% | −10% | −6% | −10% |
| Friendships across income lines | −13% | −21% | −20% | −20% |
Grey figures could easily be chance. The likely ranges allow for the fact that counties in the same state tend to resemble each other.
What this can and cannot say
- It describes counties, not people. A county with more single-parent households and more homicide does not show that single-parent households commit homicide.
- It cannot say what produces a murder rate. Nothing here was changed on purpose, as in an experiment. Every link could run the other way (violence breaks up families and empties neighbourhoods). Or it could reflect something not measured. The factors that hold up are the ones worth studying, not proven causes.
- Guns are measured indirectly.Gun access here is the share of suicides that involve a gun, adjusted for each county’s mix of sex, age and race. It follows household ownership closely but misses guns held illegally.
- Small and rural counties are missing, since their homicide counts are too small to publish.
Sources: County Health Rankings (homicide, single-parent households, segregation, overdose deaths, membership organisations), Census American Community Survey 2019–2023, BLS local unemployment, Opportunity Insights Social Capital Atlas, Vera Institute incarceration trends, CDC WONDER mortality 2018–2023 (firearm and all suicides) and 1999–2024 (homicides by year), the RAND State-Level Firearm Ownership Database (Open Database License), and ATF Firearms Trace Data 2025. Every measure the homicide rate goes with is on the page for the homicide rate by US county, and any pair can be opened in the US county pair explorer.