Craniometry is once again being used as a tool for understand population relationships. OK, it was never abandoned, but it’s undergone something of a resurgence.
But, let’s begin with Franz Boas: a major critic of typological racial thinking, he and his followers were very influential in introducing environmental factors into anthropometry – stressing the importance of health, diet, etc, on patterns in the skeletal form. In a famous study of European immigrants and their children who were raised in the USA, he found the children had different cranial measures than the parents. Though Boas wouldn’t have argued that there was no biological component to human variation, it marked a trend towards focussing on anthropometrics as measuring environmental changes, and a rejection of racial classifications.
But a few years ago, there was some dispute over whether Boaz’ conclusions really held up on re-analysis of his work. Relethford, in “Boas and Beyond: Migration and Craniometric Variation” (2004), responding to both sides, does a good job of getting to what is important: there was indeed a change in cranial measurements – but that did not obscure an underlying pattern, which you can see in this graph:
It turns out that we can get multiple pieces of data from anthropometric data:
– environmental (health/diet/etc)
– adaptative differences
The three all run the risk of obscuring each other, but we have reason to believe that they have not.
Population level differences can be caused by: health, diet, physical activity, etc. but, as shown above, they do not eliminate phylogenetic data. Neither does adaptation. If all populations were adapted for their own environment, then their phenotypic characteristics should be similar only in so far as their habitat is similar. But here we see another table from Relethford, comparing phenotypic difference to geographic distance, which shows that, similarly to the neutral genetic data, phenotypic distance increases with geographics distance, indicating that, like with our genetic data, our phylogenetic information remain intact because much of our phenotypic variation is neutral.
There has, of course, been adaptation as well, it’s just that it hasn’t obscured the neutral cranial variation. In the next table you can see the Fst values for a number of cranial measurements. Fst values measure population differentiation, and thus higher Fst values are likely to indicate selection (those Fst values that are above 0.3 are bolded and may likely indicate selective pressure).
As mentioned before, this is all very convenient: we can look for adaptive changes through high Fst values, but there clearly hasn’t been so much that phylogenetic information has been obscured; and we also can look at anthropometric data for all kinds of information on past health, but neither does this totally obscure phylogeny, which allows us to look at population relationships.
However, the craniometric data is not the same as neutral genetic data – it has less population structure, and is “less able to identify nonclinal variations among populations (which would be in accordance with the existence of biological races in the human species) than molecular data are” (Strauss & Hubbe, 2010).
Strauss & Hubbe use a dissimiliarty fraction (ω): ” the proportion of pairs of individuals from the same population that is genetically more different than pairs sampled from different populations.” In the context of genetic data, using the dissimilarity fraction showed that, once enough loci were sampled, the notion that people from the same population are often more genetically distinct than those from different populations, was falsified – reaching 0 after 800 loci were studied – in other words: “when more than 800 loci are considered, no pair of individuals from the same population is more different than any pair of individuals from any two populations.”
However, when they used this method on craniometric data, the same result was not found. When using genetic data, ω decline to 0; however, with craniometric data ω reached only a mean of 0.3. That is, about a third of the pairs within a population are more different than pairs between populations. This is despite the samples being obtained from widely separate populations, which should have enhanced differentiation.
A helpful note on the lack of resolution of craniometric data is provided by the authors when they note that “the population history signal of human craniometric traits presents the same resolution as a neutral genetic system dependent on no more than 20 loci.”
As mentioned, this means craniometric data supports “the notion of an absence of discrete biological groups… [and] indicates that cranial morphology is less able to identify nonclinal variations among populations.”
But then, why the success in using craniometric data to sort skulls correctly into populations? Strauss and Hubbe point out the importance of ‘centroids’ in these studies: “…classificatory analyses achieve high levels of success because they depend on the a priori definition of group centroids.As a consequence, when a large number of variables is considered, the probability that this kind of analysis will find a dimension in the original data that differentiates among the a priori defined groups is high. Yet the precise biological significance of this kind of difference is hard to establish, especially when the high values of dissimilarity fractions reported here are considered. High rates of correct discrimination of groups can thus be misleading in understanding the structure of human biological diversity.”
However, this is not, I think, the final word on craniometry and population structure. It is possible that newer craniometric techniques are changing the picture. For example, traditionally measurements have been linear, but new 3D methods are being developed that show greater ability to sort into populations.
Consider for example a recent paper: “Identification of Group Affinity from Cross-sectional Contours of the Human Midfacial Skeleton Using Digital Morphometrics and 3D Laser Scanning Technology” (2011).
The authors compared their ability to sort into populations, using 3D measuring of the midfacial region with traditional linear measurements of the same region. They found they were able to sort into their 3 test populations (European, Chinese & Native Californian) based on only the midfacial region with an average accuracy of 86% using 3D measurements compared with 57% using the linear measurements. It may be that the greater sensitivity of these new techniques may change the picture painted above in terms of population structure resolution achievable with craniometric data (Strauss and Hubbe, in calculating the dissimilarity fraction, used the 55 linear measurements of the skull from Howell’s (1995) database). Still, as they note: “the subasymptotic behavior of the ω curves indicates that craniometric measurements become highly redundant when more than 30 variables are included in the analyses.” However, the newer 3D measures have found some contours to be “rich in diagnostic shape information.” I would also be interested in seeing what the results of using all anthropometric (not just cranial) measurements, would be.