Are the Racial Violence Statistics Being Misread?

Three questions get argued with the same statistics, and each needs different data.

Whether the violence was racially motivated is a question about hate crime; it needs a record that someone judged the motive, a police bias flag, a prosecution decision or a victim saying so in a survey. Whether it crossed racial lines is a question about interracial crime; it needs only the recorded race of victim and offender, whatever the motive. Whether rates differ between groups is a question about disparity; it needs a numerator, a population denominator and a reason to think the two describe the same people.

The first thing to ask of any number in this area is which of the three it answers.

Four systems produce these figures. Arrest data records who police apprehend; police-recorded crime, incidents reported and logged under a counting rule; victimisation surveys, what people say happened to them; conviction data, court outcomes. A claim that holds in one and reverses in another is a finding about the instrument.

A recorded count can rise through reporting, recording, investigation, clearance or classification, none of which is offending, so a rise has to be attributed before it means anything. And missing data is not neutral: where many records have no offender attached, the missing cases are not a random sample, and a claim built on the known share covers only recorded, classified crime with a known offender.

The table in the screenshots is Expanded Homicide Data Table 6 of the FBI's Crime in the United States, "Race, Sex, and Ethnicity of Victim by Race, Sex, and Ethnicity of Offender"; below is its 2019 edition.

It holds under half of the FBI's own file, which had supplementary data for 13,927 murder victims in 2019. Of those, 7,047, 50.6%, died in incidents with one victim and one offender; Table 6 keeps only those, and only where police knew something about the offender. Its four rows sum to 6,578.

The relationship between victim and killer was unknown for 6,808 victims, 48.9%, but the offender was unknown for fewer, 3,870. Police cleared 61.4% of 2019's murders, by charging someone or identifying an offender they could not charge. The offender's race was unknown for 4,752 of the file's 16,245 offenders. In the FBI's figures for 2025, the relationship was unknown for 41.1% and 68.8% of murders were cleared.

A table of victim race by offender race can only hold cases where something is known about the offender, and those are not a random sample of all murders: clearance ran from 51.5% to 69.4% across the FBI's kinds of place, and in 2023, by the Bureau of Justice Statistics' estimates, 54% of homicides of black victims went uncleared against 42% of white victims.

Every figure in the table is a count. Comparing 566 with 246 means dividing each by its group's population, which the table does not give.

Race and ethnicity are separate fields, and nearly a third of the offender ethnicity entries are unknown. For the 3,299 white victims, 738 offenders were recorded Hispanic or Latino, 1,542 not Hispanic or Latino, and 1,019 unknown. A Hispanic offender is also counted under a race, so "White offender" here is not a count of non-Hispanic white offenders, and nothing in the table lets you produce one.

The table has no motive field, so it cannot show hate crime.

Police records also give a victim rate, which the screenshots leave out.

The Bureau of Justice Statistics puts the homicide victimisation rate in 2024 at 18.3 per 100,000 for black Americans and 2.2 for white Americans, about 8 to 1, with Hispanic victims of any race counted separately. It is a rate with a stated population base, estimated from police records the FBI collects through its newer incident-based system rather than the supplementary reports behind Table 6, and a victim's race is a required entry in those records, so the rate does not depend on the case being solved.

The Violence Policy Center, working from CDC mortality data for the same year because it finds the FBI's data less reliable for some states since the 2021 reporting change, puts the rates at 22.4 and 3.5, about 6 to 1, counting Hispanic victims inside each race. The figures differ in source, in method, since the Center adjusts its rates for age, and in where Hispanic victims sit, and they point the same way.

Download this conclusion: the rate the screenshots leave out

This section as one image: the victimisation rate from police records, and the CDC rate beside it.

Download the card (PNG, 1,200 × 630)

On 1 January 2021 the FBI stopped taking data through its old Summary Reporting System and accepted only the National Incident-Based Reporting System, so agencies that had not converted were missing. In the FBI's and BJS's figures, 65% of the population was covered in 2021, against about 95% in earlier years, and some large agencies did not submit. The FBI began collecting summary data again in 2022.

The FBI's 2021 figures are estimates, with the missing agencies filled in statistically and confidence intervals attached, and the FBI now notes that they are not nationally representative or comparable with other years.

Most police agencies report hate crime to the FBI voluntarily, and an agency can report incidents, report that it had none, or send nothing. In 2025, 16,791 agencies took part and about 2,800 enrolled ones sent nothing; more than four in five of those taking part reported no hate crime at all.

To get what it calls "a meaningful national trend", the FBI counts only agencies that reported the same six or more months in both years. On that like-for-like basis incidents fell 0.6% from 2022 to 2023, the year the all-agency total rose to 11,862 and groups such as the Anti-Defamation League called it a record. In 2025 both counts fell, the like-for-like one by 7.0%.

The National Crime Victimization Survey estimated about 244,000 hate crime victimisations a year in 2010 to 2019; police recorded 6,628 hate crime incidents in 2010 and 7,314 in 2019. The Bureau of Justice Statistics puts most of the gap down to victims not reporting and to how police classify what is reported. The FBI series counts what participating police record as hate crime.

Section 95 of the Criminal Justice Act 1991 requires the government to publish information that helps those working in criminal justice avoid discrimination. The Ministry of Justice's compendium on ethnicity, published under it, follows each group from stop and search to prison, though at least a third of court defendants have no recorded ethnicity.

The ministry warns that for most of its figures "no controls have been applied" for other characteristics, such as average income, geography, offence mix or offender history, and that differences "should not be taken as evidence of bias or as direct effects of ethnicity".

The Home Office compares each group's stop and search rate per 1,000 residents with the white rate; in the year to March 2025 the rate was 22.4 for black people and 5.9 for white people, 3.8 times as high. It also warns that a national rate "presumes that all stop and searches and ethnic groups are spread evenly across forces". London's Metropolitan Police made 23% of searches that year, in an area where 46% of residents are not white, against 13% elsewhere.

Two studies from 2000 and 2004 counted who was out where searches happened. A 2007 review by Ben Bowling and Coretta Phillips reports that, measured against those counts, white people tended to be searched at a higher rate and Asian people to be under-represented, while black people were sometimes under- and sometimes over-represented. The Home Office's figures also show about three searches in ten find something, for every ethnic group. None of these figures settles whether the searching is biased.

The claim is that 84% of grooming gang offenders are Asian. It comes from a 2017 Quilliam Foundation report that, according to a Home Office literature review in 2020, counted 264 people convicted between 2005 and 2017 but did not say how it found them or classified their ethnicity. The review said the findings could not support conclusions about offenders' ethnicity.

The Home Office's 2020 paper on group-based child sexual exploitation (CSE) says in its foreword that some studies "indicated an over-representation of Asian and Black offenders" but that it is "difficult to draw conclusions" because research is limited and data collection poor. Its summary says research has found offenders "most commonly White", that some studies suggest black and Asian offenders are over-represented, and that "it is not possible to conclude that this is representative of all group-based CSE offending".

Paragraph 81 finds it is not possible to say whether Asian or black offenders are over-represented, yet that "it seems most likely" offenders' ethnicity is in line with the general population, mostly white; the paper's own analysis of police data "would not answer the question". The line "no credible evidence that any one ethnic group is over-represented" is not in the paper. It is two academics' summary in the Guardian four days later.

In June 2025 Baroness Casey's national audit found that ethnicity is still not recorded for two thirds of perpetrators, so the national data cannot support conclusions either way. Local police data from three force areas showed "disproportionate numbers of men from Asian ethnic backgrounds amongst suspects"; with local reviews and prosecutions, she judged this enough "to at least warrant further examination". She wrote that flawed data is used "repeatedly to dismiss claims about 'Asian grooming gangs' as sensationalised, biased or untrue", and that claims of an overwhelming problem with white perpetrators "can't be proved".

So neither side's national claim has a source that can carry it. The government promised then to make police record perpetrators' ethnicity and nationality; a year later the Home Secretary said she lacked the power to require it and would seek it in a Police Reform Bill.

Download this conclusion: the grooming gangs figure

This section as one image: where the 84% came from, what the 2020 Home Office paper judged, and what the 2025 Casey audit found nationally and locally.

Download the card (PNG, 1,200 × 630)

Police in England and Wales recorded 31,844 hate crimes in the year to March 2013 and four times as many, 128,485, in the year to March 2022; in the year to March 2025 they recorded 115,990. The Home Office puts the rise down to better crime recording and better identification of hate crime, and says its figures do not provide reliable trends since 2014 and are not a good measure of prevalence.

Those figures leave out London's Metropolitan Police. Its old system marked every offence in a crime report as a hate crime if one of them was, so the higher totals published before October 2025, which included its figures, overstated the count.

The Crime Survey for England and Wales, which asks victims directly and is unaffected by changes in police recording, shows a 43% fall instead, from 307,000 incidents a year in the surveys of April 2007 to March 2009 to about 176,000 in those of April 2022 to March 2025. The Home Office puts the difference mainly down to better police recording, and notes that public order offences, over half of the police count, are not well covered by the survey.

French law forbids processing data that reveals racial or ethnic origin, with exceptions, and French crime statistics sort suspects by nationality. German police statistics, as Bundestag papers and the fact-checker CORRECTIV describe them, count suspects as German or non-German by citizenship and record no migration background. Sweden's crime prevention council's 2021 study counted suspects by background, meaning where the person and their parents were born. The European Commission reports a lack of comparable equality data across member states, with some collecting such data and others consciously avoiding it.

So "in Europe, group X commits Y times more crime" rests on a proxy, and the proxies differ. Nationality changes on naturalisation; country of birth misses the second generation; migration background is defined by citizenship at birth in Germany, where one parent suffices, and in Sweden by where the person and both parents were born.

The Nordic countries that publish crime by background publish adjusted versions too. Norway's statistical office found 11.3% of the Norwegian-born children of immigrants registered as offenders in 2010 to 2013, against 4.5% of the rest of the population; adjusting for age and sex cut that to 6.8%.

The EU's Fundamental Rights Agency surveys, including one of almost 6,800 people of African descent, measure discrimination, harassment, violence and police stops as respondents experienced them, not offending rates for any group. ODIHR, which publishes the OSCE's hate crime figures, says its data cannot be used to compare countries, because what each state records and classifies as hate crime differs.

Yes, both ways. The FBI table at the centre of the argument is genuine: a count of one-victim, one-offender murders in which police knew something about the killer, with no motive field and no population figure, holding under half of the 13,927 victims in the 2019 file. In that file the relationship between victim and killer was unknown for 48.9% of victims and the killer for 27.8%. Read as a rate or as motive, it answers a question it was never built for.

Police records also give a figure that is a rate. For 2024 the Bureau of Justice Statistics estimates 18.3 homicide victims per 100,000 black Americans against 2.2 for white, about eight to one with Hispanic victims counted apart; CDC mortality data give 22.4 against 3.5, about six to one with them counted inside each race.

Britain’s 84% grooming gangs figure has no source that can carry it, and the counter-claim, that the Home Office found no evidence of over-representation, is not in the paper credited with it. That paper said the data could not answer the question, then judged offenders’ ethnicity most likely in line with the general population, mostly white; the 2025 Casey audit found ethnicity unrecorded for two thirds of perpetrators nationally and disproportionate numbers of Asian men among suspects in three force areas. Hate crime counts move with who reports and how, and France, Germany and Sweden record nationality or background, not ethnicity. Counts get read as rates and motive, and missing data as a finding of no difference.

Download the conclusion

The verdict above as one image, with the FBI’s 2019 file drawn to scale: all its murder victims, those with no known relationship to their killer, those in the table people screenshot, and the two interracial counts. Anyone may download and post it.

Download the card (PNG, 1,200 × 630)