Hispaniolan craniometry

Calipers are back

I can't resist blogging this one. First, so it doesn't get lost, and second because it's scientifically interesting. Right now, Haiti is in the news due to social failure and even cannibalism (common among pre-colonization African cultures).

A week ago, Andrew Cutler found this 2019 press release about a study from Boston University:

It sounds almost too honest to be true, but the study is real:

• Herrera, M. D., & Tallman, S. D. (2019). Craniometric variation and ancestry estimation in two contemporary Caribbean populations. Forensic science international, 305, 110013.

Ancestry estimation of skeletonized remains by forensic anthropologists is conducted through comparative means, and a lack of population-specific data results in possible misclassifications. This is especially germane to individuals of Latin American ancestry. Generally, each country in Latin America can trace their ancestral lineage through three main parental groups: Indigenous, European, and African. However, grouping all Latin American individuals under the broad “Hispanic” category ignores the specific genetic contributions from each parental group, which is variable and dependent on the population histories and sociocultural dynamics of each country. This study analyzes the craniometric ancestry of Hispaniola (the Dominican Republic and Haiti) using 190 cranial Computed Tomography (CT) scans (f = 103; m = 87), along with the island’s history, to explore similarities and differences between the two groups. MANOVA results indicate that 53.6% and 71.4% of the 28 cranial measurements differ between the ancestries and sexes, respectively; and intraobserver error analyses demonstrate that 85.7% of measurements from CT scans are good-excellent in reliability. Further, a total of 12 canonical discriminant function analyses produced cross-validated classification accuracies of 73.7–78.6% for females, 71.8–87.5% for males, and 72.0–77.8% for pooled sex. This study demonstrates that, despite sharing a small island, Dominican and Haitian individuals can be differentiated with a fair amount of statistical certainty, which is possible due to complex socio-cultural, -political, and -demographic factors that have produced and maintained genetic heterogeneity. Moreover, the discriminant functions provided here can be used by the international forensic science community to identify individuals living on Hispaniola.

In short, they measured skulls using CT scans from 190 people from the Hispaniola island in the Caribbean to see if they could tell them apart, and they could. Just for reference, it's this island:

Due to colonialism it ended up being split between a French part and a Spanish part. The African slaves in Haiti revolted in 1791 and killed nearly all the French colonizers, so the country has been independent since around that time. Demographically, they are quite distant. I found these 23andme genetic results from 2019, which I assume are based on their customers data (probably people who emigrated or whose parents are from there):

Looking at the results for the island, we see that Haiti is about 80% African, 20% European and with trace amounts of Amerindian/native Caribbean ancestry, which is about the same as African Americans (Blacks in USA). The Dominican republic, on the other hand, is about 60% European (mainly southern), 30% African and 10% Amerindian. I suspect these values are inflated for European ancestry because 23andme customers will be self-selected for being wealthier and more scientifically curious, and thus more European than average.

Using skull measurements (indeed, any detailed bone measurements) one can estimate a person's ancestry/race quite well, especially using modern machine learning approaches. This study used only simple regression methods, 28 cranial metrics, and their sample size was small, but they were still able to classify people into whether they originated from one or the other group with some 80% accuracy versus a baseline of 53% (since the sample has 100 Dominicans, 100/190 = 53%). They didn't compute a good measure of model accuracy (like AUC or R²), so we can't compare their accuracy easily with ours and other MRI-based studies.

Though the sample is small, the study did find that Haitians were easier to classify correctly. This is expected because Haitians are more racially homogeneous as they are mainly of African ancestry. This is an example of the usual finding that social race groups that are more heterogeneous do not function as well as those that are more clear (closer approximations of natural differences).

Looking at the papers citing this study, I found this 2024 study using deep neural networks to classify skulls as European or Chinese. They were able to achieve 99% accuracy despite having a training sample of only 205 people. It looks exactly like how you would think it looks:

These kinds of finding are important because in forensic science, police will sometimes discover a body and be unable to ascertain the identity. By looking at the skull or other bone measurements, it is possible to figure out at least which racial category they belong to (in addition to sex, age, and time of death), thus making the search for the victim's identity easier. Using genomics, one can of course also do this, and even reconstruct the appearance of the person to some extent (forensic genomics).

From a historical perspective, these kinds of findings show that we have returned to where science in this field began. Roughly speaking, in the 1800s, scientists didn't care about egalitarian politics, so they went about their business without such prejudices. In the 1900s, the growing egalitarian mindset among academics served to suppress this kind of work, and especially so in the fallout of Nazism. Now in the 2000s, armed with modern data and machine learning, we are relearning the conclusions of old. The case is also interesting because it parallels the results in archaeology where scholars like Gustaf Kossinna talked about replacement migration and what we would now call gene-culture co-evolution ("Sharply defined archaeological cultural areas correspond unquestionably with the areas of particular people or tribes"). The Nazis were big fans of this (he died in 1931), and because of this, his approach was largely discredited following the war, only to return with a vengeance proven by ancient genomics. I expect other egalitarian models in archaeology and anthropology will be similarly overturned in the coming years.