Geographic Variation in Native American Anthropometrics: a Spatial Analysis of the Boas and Gifford Datasets

In 1982, anthropometric data that had been lost for decades was rediscovered, and, with it, another chance was granted to add to our knowledge of the physical anthropology of the American Indian.

Because previous spatial analysis studies either utilized only a portion of the Boas data and either utilized no statistical analyses or were not published, a more comprehensive spatial analysis is still needed. The purpose of this study is to more comprehensively re-analyze the Boas and Gifford datasets using spatial analysis methods to discover the patterns of variation revealed by the data. The following questions using spatial autocorrelation analysis were addressed. First, is there significant heterogeneity in the anthropometric data? Second, what spatial patterns are revealed by the data? Third, do the data show significant spatial structure? Fourth, do the patterns revealed by the analysis show evidence of the migration or migrations that brought Native Americans to the New World? Matrix correlation analysis utilized to examine what influence language may have had on the variation displayed in the Boas and Gifford data sets. This was thought to be important because languages that are mutually unintelligible can affect the amount of gene flow between populations.

The sample sizes consisted of 9024 individuals subdivided into 120 populations for the head dimensions and 8445 individuals spread over 119 populations for the body dimensions. The variables used in the analysis consisted of 12 anthropometric dimensions and 2 additional dimensions, arm length, calculated by subtracting finger height from shoulder height, and leg length, calculated by subtracting sitting height from standing height. The head and body measurements were analyzed separately.

After the data were corrected for inter-observer error, and age and sex variation, a variety of univariate and multivariate methods were used to address the above questions. Canonical discriminant analysis was performed to allow the spatial autocorrelation analysis to be done multivariately. For the spatial autocorrelation analysis, fifteen distance classes were chosen and Moran’s I and Geary’s c coefficients were calculated. Because the one-dimensional correlograms used cannot give the direction of the spatial pattern, two-dimensional correlograms were constructed. Six distance classes were used.

Because spatial autocorrelation does not take the influence of language on the variation into account, matrix correlation analysis was utilized to assess this influence as well as the role of the interplay between language and geographical distance on the variability of the data. Analyses were run checking for correlations between anthropometrics and geographic distance, anthropometrics and language distance, anthropometrics and geographic distance holding language constant, and anthropometrics and language holding geographic distance constant. In order to check for boundaries to gene flow, Wombling was applied to the data. Plots were generated to show the boundaries discovered by this method.

The results of the canonical discriminant analysis, spatial autocorrelation analysis, matrix correlation and Wombling revealed very little evidence for linguistic or geographic patterning in the head data, but evidence of this type was revealed by the body data.

Overall, the results showed that the head data produced little evidence for inter-continental migrations, but the body data revealed evidence for at least one such migration. In addition, a complex network of gene drift, regional gene flow, and natural selection was mainly responsible for the variation in the data.