摘要：Introduction
arising from
: A. H. Ross et al.;
Scientific Reports10.1038/s41598-019-56929-3 (2020).
Ross et al.
1
report results from the analysis of 103 pre-Columbian crania from 10 localities within the insular and circum-Caribbean. Based on geometric morphometric (GMM) and hierarchical cluster analyses, the authors conclude that individuals from Hispaniola (n = 15), Jamaica (n = 7), and the Bahamas (n = 8) form a distinct cluster (HJB) representing the migration of Caribs (peoples belonging to the Cariban language family) into the region, ca. AD 800. Ross et al.
1
link the HJB cluster to a western Venezuelan homeland based on ostensible similarities between Carib pottery found in that region and the Meillacoid ceramics archaeologically associated with HJB populations. A second cluster comprising Cuba (n = 21) and the Yucatán (n = 12) and a third consisting of Puerto Rico (n = 10), Venezuela (n = 4), and Colombia (n = 5) are linked to continental emigration events at ca. 5000 BC and ca. 800–200 BC, respectively. Specimens from Florida (n = 15) and Panama (n = 6) form a fourth cluster. To support the ethnic distinctiveness of the Carib migrants/HJB cluster, the authors rely on Columbus’ account of cannibal marauders, identified as “Caribs”, attacking peaceful Arawakan-speaking communities of the Bahamas. Ross et al.’s
1
study contains important shortcomings that bear on the reliability of their conclusions. We discuss the most significant issues here, and based on these we conclude there is no evidence to substantiate a Carib migration from Venezuela to Hispaniola 700 years before Columbus’ arrival.
The critical weaknesses of the paper lie in the number, chronology, and archaeological contexts of the crania examined. Substantiation of Ross et al.’s
1
proposed migrations requires cranial specimens from a given island cluster to be appropriately associated in time and space with the continental specimens argued to represent source populations. It also requires the latter to be securely assigned to the proposed source culture based on archaeological context. The authors do not meet any of these evidentiary requirements. The four individuals from Venezuela, for instance, are of unknown age, location, and cultural affiliation and are unlikely to represent the full biological diversity and (presumably) associated ethnic and linguistic diversity of this region. The importance of considering archaeological context is critically exemplified by the 12 Chichén Itzá samples. Although not reported by Ross et al.
1
, at least 11 of these derive from the Sacred Cenote, a Maya human sacrificial context (ca. AD 800–1200) recently shown by Sr
87/Sr
86 and δ
18O analyses to contain high proportions of non-local individuals, some possibly originating from as far away as Central America and Mexico’s Central Highlands
2
. Two samples from Ross et al. (SI
1 accession numbers 58219, 58220; recorded as 07–7–20/58219.0 and 07–7–20/58220.0, Harvard Peabody Museum online catalogue) analyzed in that study are consistent with local origins, but the remaining eight, including two reported subadults (07–7–20/58203.0, 07–7–20/58225.0), are unaccounted for and may not represent local individuals and associated craniofacial traits.
Overall, 88 specimens (85%) lack site-specific provenience information and most lack direct radiocarbon dates (Table 2 and SI in Ross et al.
1
). They can be assigned only to 500–700 year ranges post-dating AD 600, or to the pre-contact period more generally. The uncertain spatial and temporal context of some specimens and absence of full details for others makes the sample inadequate for testing the proposed Carib-HJB migration.
Besides sample limitations, the paper employs statistical analysis that cannot reliably represent biological relationships between the series. Ross et al.
1
identify three migrations based on four observed clusters in the morphological series produced by hierarchical cluster analysis. Traditionally used to reconstruct phylogenetic histories
3
, cluster analysis is limited as a method for inferring population histories, particularly for groups or populations that are biologically similar. Different clustering algorithms may result in disparate results
4
and, importantly, clusters may create artificial associations between distant series because series are clustered one pair at a time
5
. To illustrate these issues and their impact on Ross et al.’s
1
conclusions, we present the results of a Kruskal’s Non-Metric Multidimensional Scaling Analysis (MDS)
6
, computed with R
7
using package MASS
6
, based on the Mahalanobis Distances reported in Ross et al.
1
Table 3. This MDS analysis does not presume the existence of clusters among the data and calculates iteratively the closest possible bi-dimensional (or three-dimensional) visual representation of the distance matrix. The MDS reported here represents the data in two dimensions, using the centroid of the points as its origin
6
. The difference between observed distances and point positions in the final MDS output is measured through the stress in the analysis, measured in percentage, where a stress of 0 means the positions of the plotted points represent the distances in the matrix without any distortion. While MDS may result in distorted relationships between series since it attempts to depict multivariate space in two dimensions, in this case it represents more accurately the morphological distances between series (Ross et al.
1
Table 3) and, as we will demonstrate, should be favored over hierarchical clustering.
The MDS in Fig.
1 shows that Puerto Rico is in fact closer to Hispaniola and Jamaica than to the South American series, which is coherent with the D
2 values (Ross et al.
1
Table 3), thus indicating the smallest biological distances to the Puerto Rico series are indeed Jamaica and Hispaniola. Likewise, Florida exhibits similar distances to Yucatán, Cuba, and Panama in the MDS. While the smallest biological distance to Florida in Table 3 of Ross et al.
1
is with Panama (6.10), the similar distances to Yucatán (6.66) and Cuba (6.58) cannot support the existence of a basal cluster between Florida and Panama, as suggested by their cluster analysis. The MDS thus represents the morphological distances among series with significantly less bias, controverting Ross et al.’s
1
conclusions that “individuals from Florida and Panama…show no clear relationship with individuals from the islands” (p. 4), and that “data confirm a biological relationship between individuals from Venezuela and Puerto Rico” (p. 4). Further, these conclusions are undermined by not examining the clustering dendrogram for treeness; not providing confidence measures for the purported nodes; and not acknowledging that canonical variate analysis artificially creates groups, especially with samples as small as these
8,
9
.
Figure 1
Morphological affinities among the series according to Multidimensional Scaling Analysis of the Mahalanobis Distances reported by Ross et al.
1
for 103 pre-Columbian individuals from 10 localities. Series colours follow the four clusters identified in Ross et al.
1
, Fig. 5.
Additionally, in the absence of a demonstrated biological connection to western Venezuela the HJB cluster’s link to this region hinges solely on the claim that “Meillacoid pottery [from Hispaniola, Jamaica, and the Bahamas
