Do Violent Offending Rates Differ by Race?

Who gets victimised, who offends against whom, and whether violence was racially motivated are three separate questions, each answered by a different table. This page is not about any of them. It is about offending rates. A companion piece at the foot of this page deals with what goes wrong when one table is swapped for another.

Rates of registered offending are not rates of offending: they record what was reported, investigated, solved and classified. And a difference between groups, before or after statistical adjustment, is not a property of the groups.

The Bureau of Justice Statistics (BJS) has published an offending rate, divided by each group’s resident population. BJS estimated it from the Supplementary Homicide Reports of the Federal Bureau of Investigation (FBI). Averaged over 1980 to 2008, the homicide offending rate was 34.4 per 100,000 for black Americans and 4.5 for white Americans. BJS describes the black rate as almost eight times higher than the white rate. Over the same period the victimisation rates were 27.8 for black Americans and 4.5 for white.

The offending rate stops in 2008. BJS left offending rates out of its homicide report of 2013, doubting the offender data were representative, and has published none since. Any American homicide offending rate by race for a year after 2008 has been built by someone other than BJS.

In the FBI’s 2019 expanded homicide data, of 16,245 murder offenders the race was unknown for 29.3%. In its 2025 data the victim-offender relationship was unknown in 41.1% of murders. The missing data are not random. BJS found that in 2011 the offender’s age, sex and race were all missing for 40% of black homicide victims and 23% of white ones. Its 1980 to 2008 rates estimate each missing offender from similar killings where the offender was known; the raw FBI tables count them as unknown.

A Bureau of Justice Statistics (BJS) report was built to test the obvious objection to any arrest-based figure, that it measures policing.

BJS compared the race of offenders as described by victims in the National Crime Victimization Survey (NCVS) against arrests for the same year, 2018. The arrests were from the Federal Bureau of Investigation’s Uniform Crime Reporting (UCR). If arrest disparities were being generated by police rather than by offending, the two distributions would diverge.

BJS states the result in its own words: there were no statistically significant differences by race between offenders identified in the NCVS and persons arrested per the UCR. White and black people, it adds, were arrested proportionate to their involvement in serious nonfatal violent crime.

Victims’ own accounts of their offenders, gathered by a survey rather than by the police, match the arrests for white and black offenders. The result means the black share of arrests for serious nonfatal violence is not an artefact of differential arrest. It also means the disparity in that arrest figure is a disparity in involvement.

Hispanic arrestees were the exception, 21.4% of arrests against 12.0% of offenders in reported incidents, a statistically significant overrepresentation among those arrested. BJS notes victims could not determine Hispanic origin in about one incident in ten, which may account for some of it.

The BJS test covers nonfatal violence only. It says nothing about homicide, or about any stage of the system after arrest. Offender race or ethnicity also went unreported in 14.2% of all nonfatal violent incidents, and 18.0% where the victim was black.

American crime statistics record race and Hispanic ethnicity in separate fields, and roughly 93% of Hispanics are coded as white in the race field. Steffensmeier and colleagues, in a 2011 study, say this coding inflates white offending figures and deflates black ones.

They re-estimated the national arrest figures by race using California and New York arrest data, which flag Hispanic ethnicity, to separate Hispanics out.

In the uncorrected figures, the black share of violent offending appeared to decline over three decades. Their wider conclusion is that this apparent decline largely disappears once Hispanics are separated out. What looked like a falling share was substantially a growing Hispanic population being counted as white.

Two studies do the work here, and they reach different places.

At the individual level, a 2005 study by Sampson, Morenoff and Raudenbush followed 2,974 people across 180 Chicago neighbourhoods. The participants were aged 8 to 25, and this Chicago study used self-reported violence rather than police records. Allowing only for age and sex, the odds of perpetrating violence were 85% higher for black than white participants.

Controls for family background, verbal and reading ability, impulsivity and neighbourhood context explained over 60% of that gap. Parental marital status, immigrant generation and neighbourhood social context did most of the work. A reduced gap remained after controls.

In the Chicago study the odds were 10% lower for Mexican American participants than for white ones, and 26% higher for Puerto Rican and other Latino participants. The controls explained both differences entirely. Any account that arranges groups into a simple hierarchy has to deal with that.

At the neighbourhood level, Peterson and Krivo, in a 2010 book, analysed roughly 8,900 neighbourhoods (census tracts) across 87 cities. Raw violent crime rates (homicide and robbery, 1999 to 2001) ran 10.0 per 1,000 residents in black neighbourhoods against 2.0 in white ones. That is about five to one.

Holding city characteristics and the share of young men constant, the black-white ratio was 4.27. Adding residential instability, lending and immigration took it to 3.26, and adding concentrated disadvantage took it to 1.65. Adding the surrounding neighbourhoods’ instability, lending, disadvantage and share of white residents took it to 1.19, still significant, and their violent crime to 1.13, which is not. With all these controls, the Latino-white ratio fell to 1.02, also not significant.

The fall to 1.13 looks like the gap disappearing, and the authors’ own data explains why it should not be read that way. In their sample, 88.9% of white neighbourhoods have none of the extreme forms of disadvantage they measure. Among black neighbourhoods, 56.4% have four or more. The distributions barely overlap.

A model that controls for disadvantage therefore rests on an extrapolation: it compares places that mostly do not exist against each other. Its result says that if the neighbourhoods were comparable the crime rates would be too, not that they are comparable.

So the residual, what is left of a gap after the controls, depends on where you measure. At the individual level a reduced but real gap remains after all the controls. At the neighbourhood level it goes to non-significance, on an extrapolation. Neither model contains individual criminal history or individual exposure to violence. "Not explained by this model" is not the same as "unexplainable", and it is not the same as "explained by race".

The Home Office publishes arrest rates per 1,000 population by ethnic group.

Black people were arrested at about 2.2 times the white rate. Among men the rates were 38.2 per 1,000 for black men against 16.0 for white men, about 2.4 times. Asian arrest rates sit below white ones, which again is awkward for a simple hierarchy. Ethnicity was unknown for 12.7% of arrests.

The broad categories hide wide internal ranges. Within the Black group the rate runs from 13.1 for Black African to 52.4 for Black other. Within the Asian group it runs from 2.8 for Chinese to 13.5 for Asian other.

For homicide, the Ministry of Justice and the Office for National Statistics give the ethnic groups of principal suspects convicted.

For the three years to March 2025, the Office for National Statistics puts black suspects at six times their share of the population (4%). Over the same years it puts black homicide victimisation at 36.8 per million against 7.7 for white, over four times higher.

The Ministry of Justice attaches a warning to most of its own statistics on ethnicity and the justice system. The warning governs every figure above. It says no controls have been applied for other characteristics of ethnic groups, such as average income, geography, offence mix or offender history. So, it says, it is not possible to determine what proportion of the differences are directly attributable to ethnicity.

The warning adds that no causal links can be made between ethnicity and outcomes in the justice system. The differences, it says, should not be taken as evidence of bias or as direct effects of ethnicity.

The Home Office’s own Offending, Crime and Justice Survey points the other way from the arrest table. It asked a national household sample in England and Wales about their own offending instead of counting arrests. Its longitudinal analysis followed those aged 10 to 25 from 2003 to 2006. It reports that the 2003 survey had found minority ethnic respondents less likely to offend than white ones. It also reports that once family, peer group, school and area factors were controlled for, ethnic group was not significantly associated with offending in any of its models.

The survey measured mostly minor offending among young people, with assault the most common item and robbery rare. Black respondents were about one in a hundred of the sample. Unlike the Chicago study, the survey does not measure how much of a violent-offending gap controls explain, so it answers the question only in part.

Download this conclusion: England and Wales

This section as one image: the arrest gap, the ranges inside the broad labels, the Ministry of Justice’s own warning, and the Home Office survey that found no significant link between ethnic group and offending once background was controlled for.

Download the card (PNG, 1,080 × 1,350)

No European country in this section measures race. They measure country of birth, parents’ country of birth, or citizenship. Those are different variables, defined differently in each country, and no country’s figures are comparable with another’s.

Sweden’s crime prevention council, Brå, publishes its adjustment step by step. Its measure is the relative risk of being registered as a crime suspect over 2015 to 2018, among residents aged 15 and over. A relative risk is a group’s rate as a multiple of the rate for the baseline group, Swedish-born people with two Swedish-born parents.

About half of the excess for the foreign-born, and two thirds of it for the second generation, is accounted for by the composition of the groups. The rest is not, in that model.

For the second generation the relative risk for attempted and completed homicide was about 11 raw and 3.4 standardised. For robbery it was also about 11 raw, and 3.1 standardised. The category Brå labels attempted and completed homicide is predominantly attempted. In 2005 Brå judged that selection in reporting and police investigation explains only a small part of the relative risk.

Germany publishes a rate of suspects living in Germany per 100,000 residents aged 8 and over. Excluding immigration offences, which almost only foreign nationals can commit, it ran 1,813 for Germans against 4,788 for non-Germans in 2025, about 2.6 times. For violent crime the Federal Criminal Police Office puts the non-German rate at roughly four times the German one. The rates were 628 for non-Germans and 158 for Germans.

The Interior Ministry’s report on these figures notes that they record only crime known to the police. It also notes that the non-German population has a larger share of young men, a group with above-average rates.

The Netherlands recorded 61 suspects per 10,000 residents of Dutch background in 2022, and 150 with a migration background. No adjusted rate sits beside them; Statistics Netherlands says the gaps shrink once age, education and income are taken into account.

Denmark publishes a conviction index, with all men aged 15 to 79 set at 100, and shows the Danish-born children of non-Western immigrants well above that level. Statistics Denmark gives the index unadjusted and standardised four ways, beside raw conviction shares, so a quoted index needs its version named. It standardises for age alone, and for age with socioeconomic status, family education or family income.

Norway reports overrepresentation that shrinks after adjusting for age and sex; it fell in the decade to 2015 and has risen since.

Download this conclusion: Europe, raw and adjusted

This section as one image: what European countries measure instead of race, Sweden’s raw and adjusted figures side by side, and which other countries publish an adjusted figure.

Download the card (PNG, 1,080 × 1,350)

The raw gaps are large. The American black-white homicide offending gap is roughly eightfold on the last series the Bureau of Justice Statistics (BJS) published. The BJS test shows the arrest gap for serious nonfatal violence tracks what victims describe, so policing does not manufacture it. Swedish, Danish and Norwegian overrepresentation survives the adjustments their statistical agencies apply, at a reduced size. Statistics Netherlands says its gaps shrink, and Germany publishes no adjusted rate.

By contrast, the neighbourhood-level black-white gap loses statistical significance under controls for disadvantage and for the surrounding neighbourhoods, the last of them those neighbourhoods’ violent crime. In the Chicago study both Latino-white gaps disappear under controls, although Mexican Americans start below white participants and Puerto Ricans and other Latinos above them. Asian arrest rates in England and Wales sit below white ones. The apparent decline in the black share of American violent offending is largely a Hispanic coding artefact.

The residual at the individual level, the gap left after the controls, neither holds nor collapses. After the Chicago study’s controls it is smaller than the raw gap, still statistically significant, and nothing here says what it consists of. No model on this page holds a person’s own criminal history constant, because the Chicago authors left out factors that might be outgrowths of offending. Reporting the residual as an inherent group difference is unsupported; reporting it as zero is also unsupported.

Yes. Recorded rates of violent offending differ by race in the United States, by ethnic group in England and Wales, and by birthplace or citizenship in Europe. The raw gaps are large. The last homicide offending rate by race from the Bureau of Justice Statistics (BJS), averaged over 1980 to 2008, was 34.4 per 100,000 for black Americans against 4.5 for white. For serious nonfatal violence, victims’ descriptions of white and black offenders match the arrests, so that arrest gap is not made by policing.

Wherever controls are applied the gaps are smaller than the raw figures, and some shrink to nothing measurable. What remains is real in the American individual-level data, and identified by nothing here as an inherent difference between groups.

Download the conclusion

The verdict above as one image, with the neighbourhood-level result as its chart: the black-white ratio of violent crime across 87 American cities as each control is added, 4.27, 3.26, 1.65 and 1.13. It is 1,080 by 1,350 pixels, and anyone may download it and post it.

Download the card (PNG, 1,080 × 1,350)