The Not-So-Nefarious Reason for White Crime Stat Confusion

Are statistics on white crime being manipulated?

Over the past few years, anyone using Twitter/X or even Facebook has probably seen racialist but often technically skilled bloggers post pictures of exotic-looking criminal defendants -- bearded Middle Easterners, Latin American gang sicarios -- who have been classified as “white” by regional sheriff’s departments. The implication, of course, is that this may well be a national statistical problem. However, until this year, no one has conducted any rigorous, at-scale quantitative test of the potential issue.

That recently changed. Danish researcher Emil Kirkegaard and several other analysts “scraped 5.5 million criminal records and 1.5 million mugshots from 39 U.S. states,” which make legal data publicly available via portals such as Judici -- and then trained a program to identify the race of arrested criminal suspects based on 18 individual and case features. This was then compared to the officially recorded race given for each suspect.

What did these rogues find? At the highest level of analytical confidence possible -- an impressive 95-100 percent -- their model estimates that 22.4 percent of Hispanic suspects were almost certainly misclassified as white. In contrast, relatively few white men or women were misclassified as Hispanic, and almost no whites or blacks were labeled as members of the other race. If all of these identified misclassifications in fact occurred, and pure national crime rates for each race could be calculated, Kirkegaard et al. argue that the overall “criminal record rate” for U.S. Hispanics would increase 20-31 percent, the rate for the far larger white population would drop by more than 4 percent, and the rate for the black population would decrease by 1 percent.

The next question, at least online, seems to be: is the misclassification part of some deliberate plot? Some certainly think so, referring to the process as “white labeling” and other such terms.

However, the short answer is: almost certainly not. Many or most U.S. Latinos -- about 58 percent -- identify as being white along with Hispanic, and officers tend to respect the self-identifications that defendants give. For that matter, the substantial majority of these people are more than 50 percent Caucasian in DNA terms: Spain is a country in Europe. It is far more likely that what we are seeing is conflict between the offhand police classification of Cubans and Argentines as “white” and ruthless AI analysis than an intentional plot by right-leaning police departments against the nation’s majority race and/or its justice system.

There is another, more important point here, concerning the actual impact of the (fairly impressive) Kirkegaard data. The primary source of U.S. national criminal justice data, the FBI’s Uniform Crime Reporting (UCR) and National Incident-Based Reporting System (NIBRS) database, already combines whites and Hispanics into a “Caucasians” or “whites: all” category when summarizing crime totals. Thus, in a standard FBI data table from the typical year of 2019 -- Expanded Homicide Data Table #6 -- when we see that there were 3,299 single-offender white murder victims, “white” is already defined so as to include Jews, non-black Hispanics, Middle Easterners, and other groups technically classified as Caucasian.

In an entirely separate portion of the FBI-UCR table, Hispanic/Latino is broken out as an ethnicity, rather than a race -- this is already done, secondarily, on most police reports -- so we can see that 738 of the offenders against the victim pool just mentioned were identified as Hispanic while the rest were not. Another 566 killers of whites were blacks (and about 250 killers of blacks were white). Identification of Hispanics as Hispanic, rather than as white -- certainly on the “race” line -- on a police report later sent to the FBI would change either very little or none of this data.

To provide an even deeper level of data comprehension here: the UCR and NIBRS aren’t the only crime databases available, and an even better-designed option already makes it possible to see the white crime rate with Hispanics removed. Unlike the UCR, the National Crime Victimization Survey from the Department of Justice’s Bureau of Justice Statistics (the BJS-NCVS) is based around Census-style interviews of several hundred thousand Americans annually about their experiences with crime. Police departments are not involved, making extensive data manipulation impossible at that level: respondents who were crime victims are asked precise questions about their attacker(s) by trained interviewers, and their answers are extrapolated out to the national population.

In 2023, the most recent year fully on record, 53.7 percent of the violent criminals responsible for “Index” offenses such as assault, aggravated assault, robbery, and rape were white, about 24 percent were black, and a bit under 15 percent were Hispanic -- with whites and Hispanics broken out here as totally separate categories. Given the methodology of the NCVS, it is possible that some Hispanics could have been misclassified as white, but it seems equally plausible that Italian, Greek, and Albanian thugs were sometimes mislabeled as Latinos. Certainly, there is no reason to believe that victims of real-world batteries or sex crimes are describing black or Mexican assailants who attacked them as blonds to make some bizarre political point.

Furthermore, while there isn’t much reason to doubt the NCVS numbers in the first place, it’s worth noting that adjusting for the Kirkegaard team’s max-confidence estimate of 22.4 percent Hispanic misclassifications would only move the Hispanic percentage of U.S. criminals up to 19.2 percent and the white percentage down to almost exactly 50 percent. Blacks would fall to 23 percent, if my back-of-matchbook scratchings are correct -- and “FBI Caucasians” combined would be responsible for an almost identical 68-70 percent of crime across datasets.

We really do, pretty much, know the crime rate in the United States. While it is not the focus of this criminological essay, a frankly more interesting question is: do we really know what it means to be white or black (or, indeed, American)? To a very striking extent, the contemporary right-wing debate about “population replacement” centers on this question.

So, for example, we often hear that the United States’ population has “fallen” to being just 61.6 percent or even 58 percent white. However, this is in fact a population figure for “(Anglo) white alone”; with white-identifying Hispanics added to the equation, the USA is at minimum 71 percent white. Nor, as we have seen above, is “white Hispanic” a spurious category.

It is somewhat taboo to remember this today, but “Hispanics” as a recognized Census population group date back only to debates in 1970 about the reach of affirmative action law and whether this covered Spanish speakers. Prior to that year, all or almost all Caucasian or mestizo Latinos were legally white, in much the same way Italians and (entertainingly) most Moors are still thought to be.

Even beyond these internal “white” struggles, the American classification debate is a fascinating one. Two other giant U.S. populations today are “Asian alone or in combination (7.2%)” -- the highest-performing group in the country, by several metrics -- and “multi-racial (and generally ½ white)(10.2%).” The black population has remained stable at roughly 12 percent for multiple decades.

To some extent, questions about whether a Jewish or Arab embezzler, a Cuban cocaine cowboy, a Turkish tax cheat, or a plain Italian or Mexican domestic abuser is a “white criminal” don’t reflect some tricky plot by anyone. They reflect, in fact, a deeper debate about what race even means in the era of benefits and subsidies.