Comparative samples comprise 32 Neandertal and 23 European Middle Pleistocene mandibles. Thirty-seven linear distances (56, 57) as well as 2D (58–60) and 3D landmark coordinates were obtained from original fossils or good quality casts. Landmarks were digitized by using a MicroScribe 3DX digitizer (Immersion Corp., San Jose, CA) (58).
2D Geometric Morphometrics. A partial least-squares analysis was performed using the 2D data of 19 mandibles (59, 60). Partial least-squares analysis helped to find correlated pairs of linear combinations (singular vectors) between two blocks of variables (61, 62). Detailed technical information about the method in geometric morphometrics is relatively new (63, 64). The singular vectors are constructed in the form of new, paired "latent" variables (one per block, also called "singular warps") (64) that account for as much as possible of the covariation between the two original sets of variables. In a sense similar to a principal-components analysis (PCA), the singular-value decomposition (SVD) describes the data in terms of scores of each specimen along singular axes, singular values (similar to eigenvalues), and loadings (singular vectors, similar to eigenvectors). However, singular-value decomposition is applied with a different goal, i.e., to maximize low-dimensional representation of between-block covariance (SVD) vs. maximizing low-dimensional representation of total sample covariance (PCA). The singular warps display the maximal covariance between both the shape variables within-block (mandibles) and with the variables of the other block (geographic distribution) (61, 62, 64). The associated morphological patterns are visualized by singular warps (64), which are thin-plate spline transformations of the mean shape into a predicted shape according to singular vector scores of the given axis (north–south).
3D Geometric Morphometrics. Further the hypotheses of geographic differences within northern and southern Neandertals and that of species specific shape differences between those and EMP humans were tested by Goodall's F tests (65) (34 3D landmarks). This test is based on mean shape comparisons. It compares the deviation of squared Procrustes distances in shape space from the group means and compares them with the variance around the grand mean and assesses the ratio of explained and unexplained variances of group factors (65). This test was accompanied by a permutation test (n = 2500), which permutes specimens of groups at random and calculates the F distribution of random group comparisons against which the observed F score can be compared. These analyses give a further estimate of the significance of Goodall's F test, and they were performed by using Simple3D-IMP software (Integrated Morphometrics Package) (15, 66).
We thank A. Estalrrich, S. García, A. Cruz, and M. Pérez for help managing the fossil collection; all of the people working at the El Sidrón excavation for their effort and interest; and E. Trinkaus for improving the quality of the text and providing 3D data from some significant specimens. This work was supported by Principado de Asturias-Universidad de Oviedo Grant CN-04-152. Partial aspects of the investigation are included in Project BOS2003–01531, Spanish Ministerio de Educación y Ciencia.