Multivariate classification of human populations. I. Allocation of Yanomama indians to villages

A set of 12 anthropometric measures and six genetic traits, available for 520 Yanomama Indians from 19 villages in nine clusters, were used to allocate individuals to villages. On the basis of anthropometrics alone, 36% of the individuals were allocated to the right village and 60% to the right cluster. On the basis of genetic traits alone, 16% were allocated to the right village and 26% to the right cluster. A combination of all 18 characters yielded 41% allocation to the right village and 63% to the right cluster. Of the 924 possible combinations of six anthropometric measures, only one provided poorer resolution than did the six genetic traits. We explain the better resolution of the anthropometric traits by noting that the anthropometric traits are not totally heritable and that genetic traits are not continuously distributed. Randomization studies indicated that all of the observed correct-allocation fractions are far in excess of random expectation. We infer that the village phenotype distributions overlap only partially, and that they represent real and substantial population differentiation.