Alex Nowrasteh
Two scholars at the Manhattan Institute (MI) objected to my tweet, noting that Somalis aged 18–54 have an incarceration rate above that of other immigrant groups but below that of native-born Americans in the same age range. They did their own analysis and found that male 18–29 year old Somali immigrants who arrived age 15 or younger have a higher incarceration rate than all native-born American males and white native-born American men in the same age group.
Both analyses are correct, but they claim theirs is an apples-to-apples comparison and that mine is misleading. Their analysis is not apples-to-apples, nor was mine misleading.
How can both be correct? Because they answer different questions by estimating different parameters. Researchers comparing incarceration rates between groups must choose the groups. There must be a comparison because there is no objective high or low incarceration rate without reference to other incarceration rates. My original tweet compared incarceration rates between Somali immigrants, native-born Americans, all legal immigrants, and all illegal immigrants in the 18–54 age range. That age range was chosen to increase the accurate identification of the incarcerated population in the American Community Survey data. That’s an apples-to-apples comparison because it’s a population-level comparison.
MI compares smaller subsets of those populations based on sex, age, and age at arrival. MI’s subgroup analysis does not overturn my population-level finding that Somali immigrants are incarcerated at lower rates than native-born Americans. You wouldn’t know that by reading their headline, which implies a population-level analysis: “Yes, Somali Immigrants Commit More Crime Than Natives.” Should have been “Young Male Somali Immigrants Who Have Been in the US for Many Years Have Higher Incarceration Rates.”
My population level estimate does not invalidate theirs. However, MI goes a step further to claim that theirs is an apples-to-apples comparison that corrects my “misleading comparison.” They made additional choices to compare different slices of the incarceration data to other groups, which I did not. But their choices mean that their analysis is not apples-to-apples. Let me explain.
After the MI analysis spends a few hundred words explaining why age and duration in the United States matter, which makes sense as I use the same variables in different analyses of immigrant incarceration rates, the authors mention that race matters too. And then they quickly drop the race variable, even though they claim they’re making an apples-to-apples comparison. If they’re going to control for age and sex, then they should also control for other known variables that matter, such as race, to achieve a better “apples-to-apples” comparison. Here, that would mean comparing the Somali incarceration rate to native-born black Americans rather than just native-born white Americans.
The same reasons to compare sex-to-sex incarceration rates apply just as well to a race-to-race comparison. Race is a descriptive control here because the comparison is about observed incarceration rates. After standardizing for age and sex, omitting race while claiming an apples-to-apples correction changes the descriptive baseline rather than clarifying it. Picking different comparison groups is fine, but don’t call it apples-to-apples, or else people will get the wrong impression.
Why didn’t I use any controls in my original tweet? Other immigration restrictionists like Jason Richwine argue that we should eschew any controls or restricted comparisons because when comparing immigrant incarceration rates, what matters is how the immigrants who are actually here affect crime, not how a selected subset of them compares to a selected subgroup of natives. Fair enough. We published both, but led with published data comparing all immigrants to all native-born Americans. The result is that all Somalis have slightly lower incarceration rates than all native-born Americans.
MI scholars narrow their comparison to males aged 18–29, a defensible choice given age and sex differences in incarceration rates. They then note that “crime and incarceration rates differ sharply by race in the United States” but do not include race in the analysis. When I asked one of the MI authors about the choice to compare by duration in the US, age, and sex, but not race, he said it was because of the legacy of slavery and racism that black Americans have suffered. That’s not an answer to my question, but I did not expect an MI scholar to credit the relatively high black American incarceration rate to the legacies of slavery and racism. Now I’ve seen everything, and it’s less than two weeks into 2026.
MI scholars are free to compare Somali incarceration rates to any groups they wish, but they cannot claim to have run an apples-to-apples comparison to correct a misleading analysis when neither is true. Yes, male Somali immigrants in the 18–29 age range have a higher incarceration rate than native-born and native-born white American males in the 18–29 age range. The incarceration rate for all Somalis is one-third of that of all black native-born Americans. That latter comparison is more relevant for an apples-to-apples comparison since Somalis are also black.
What does that tell us? Many different questions can be answered by comparing the incarceration rates of various subgroups of populations with each other. MI’s comparison is informative on its own terms. The problem arises only when they present it as a correction to a supposedly misleading population-level claim. If the authors want to claim an apples-to-apples comparison, they must specify which apples matter and why. Otherwise, the comparison answers a different question than the one it claims to correct.