Hominid mandibular corpus shape variation and its utility for recognizing species diversity within fossil Homo


Michael R. Lague, Natural Sciences & Mathematics, The Richard Stockton College of New Jersey, Pomona, NJ 08240-0195, USA. E: michael.lague@stockton.edu


Mandibular corpora are well represented in the hominin fossil record, yet few studies have rigorously assessed the utility of mandibular corpus morphology for species recognition, particularly with respect to the linear dimensions that are most commonly available. In this study, we explored the extent to which commonly preserved mandibular corpus morphology can be used to: (i) discriminate among extant hominid taxa and (ii) support species designations among fossil specimens assigned to the genus Homo. In the first part of the study, discriminant analysis was used to test for significant differences in mandibular corpus shape at different taxonomic levels (genus, species and subspecies) among extant hominid taxa (i.e. Homo, Pan, Gorilla, Pongo). In the second part of the study, we examined shape variation among fossil mandibles assigned to Homo (including H. habilis sensu stricto, H. rudolfensis, early African H. erectus/H. ergaster, late African H. erectus, Asian H. erectus, H. heidelbergensis, H. neanderthalensis and H. sapiens). A novel randomization procedure designed for small samples (and using group ‘distinctness values’) was used to determine whether shape variation among the fossils is consistent with conventional taxonomy (or alternatively, whether a priori taxonomic groupings are completely random with respect to mandibular morphology). The randomization of ‘distinctness values’ was also used on the extant samples to assess the ability of the test to recognize known taxa. The discriminant analysis results demonstrated that, even for a relatively modest set of traditional mandibular corpus measurements, we can detect significant differences among extant hominids at the genus and species levels, and, in some cases, also at the subspecies level. Although the randomization of ‘distinctness values’ test is more conservative than discriminant analysis (based on comparisons with extant specimens), we were able to detect at least four distinct groups among the fossil specimens (i.e. H. sapiens, H. heidelbergensis, Asian H. erectus and a combined ‘African Homo’ group consisting of H. habilis sensu stricto, H. rudolfensis, early African H. erectus/H. ergaster and late African H. erectus). These four groups appear to be distinct at a level similar to, or greater than, that of modern hominid species. In addition, the mandibular corpora of H. neanderthalensis could be distinguished from those of ‘African Homo’, although not from those of H. sapiens, H. heidelbergensis, or the Asian H. erectus group. The results suggest that the features most commonly preserved on the hominin mandibular corpus have some taxonomic utility, although they are unlikely to be useful in generating a reliable alpha taxonomy for early African members of the genus Homo.


The hominin fossil record contains a relatively large number of mandibular specimens. For example, out of 126 hominin cranial fossils from Koobi Fora examined by Wood (1991), the largest single subset (making up 40% of the total) comprises well-preserved mandibles or recognizable fragments of the mandibular corpus. As a result of this biased preservation, mandibular data have played an important role in the definition of a number of hominin fossil species. Almost 50% of the hominin species identified to date have a mandibular specimen as their holotype [e.g. Australopithecus anamensis, Leakey et al. 1995; A. afarensis, Johanson et al. 1978; A. bahrelghazali, Brunet et al. 1996; Paranthropus aethiopicus, Arambourg & Coppens, 1968, Chamberlain & Wood, 1985; P. crassidens (now P. robustus, Broom, 1938), Broom, 1949; Homo ergaster, Groves & Mazák, 1975; H. heidelbergensis, Schoetensack, 1908; H. antecessor, Bermúdez de Castro et al. 1997]. Accordingly, it is important to verify the ability of mandibular morphology to identify taxonomic affiliations, particularly those aspects of morphology most frequently preserved on fossil specimens.

Discussion regarding the utility of hominid mandibular morphology for delineating species and reconstructing their relationships has continued from the early 1900s to the present. Early studies focused on modern human mandibular morphology with a view to gaining an understanding of inter- and intrapopulation variation, including sexual dimorphism (e.g. Harrower, 1928; Martin, 1936; Morant et al. 1936; Cleaver, 1937; Hrdlička 1940a,b). As computational power increased, researchers applied multivariate statistics and other new methodologies, such as geometric morphometrics, to describe and quantify mandibular variation in modern human populations (Humphrey et al. 1999; Oettlé et al. 2005; Nicholson & Havarti, 2006; Schmittbuhl et al. 2007) and fossil hominin species (Chamberlain & Wood, 1985; Wood & Lieberman, 2001; Rak et al. 2002; Kaifu et al. 2005; Nicholson & Havarti, 2006). Other recent studies have used multivariate mandibular data to evaluate a fossil specimen's taxonomic identity, test hypotheses of species integrity, or reconstruct phylogenetic relationships (Bromage et al. 1995; Rosas, 1995; Lam et al. 1996; Rosas & Bermúdez de Castro, 1998; Stefan & Trinkaus, 1998a,b; Rosas & Bermúdez de Castro, 1999; Schwartz & Tattersall, 2000; Silverman et al. 2000, 2001; Quam et al. 2001; Rosas, 2001; Rak et al. 2002; Rosas & Bastir, 2004; Rightmire et al. 2006; Skinner et al. 2006). A number of studies examining great ape mandibular morphology have also been published (Aitchison, 1963, 1965; Kinzey, 1970; Wood, 1985; Daegling & Jungers, 2000; Taylor, 2002, 2003; Taylor & Groves 2003; Taylor, 2005, 2006a,b,c; Schmittbuhl et al. 2007).

Several of the above studies have analysed the ability of mandibular morphology to accurately predict group membership of known specimens from extant groups (e.g. Humphrey et al. 1999; Silverman et al. 2000; Taylor & Groves 2003; Schmittbuhl et al. 2007) but only rarely do those studies subsequently use those validated measures to assess species diversity in the fossil record. One reason for this is that the measurements used on the extant specimens (e.g. measures of the mandibular ramus) are often not readily available from the fossil specimens due to poor preservation.

In this study, we assessed the taxonomic utility of mandibular morphology by focusing on the traditional linear measurements that are most widely available in the hominin fossil record, namely those of the mandibular corpus. More specifically, we explored the extent to which mandibular corpus morphology can be used to: (i) discriminate among extant hominid taxa and (ii) support species designations (largely based on non-mandibular evidence) among fossil specimens assigned to the genus Homo.

In the first part of this study, we used discriminant analysis (DA) to test the hypothesis that size-adjusted linear measurements of the mandibular corpus can be used to sort extant hominid taxa. We cannot test the taxonomic value of mandibular variables using the fossil record, as we have no independent means of determining the taxonomy of the specimens concerned. We can, however, investigate the taxonomic utility of mandibular corpus variables in extant hominoid taxa closely related to the hominin clade (e.g. Taylor & Groves, 2003; Wildman, et al. 2003). We are aware of the arguments suggesting that, with respect to fossil hominins, genetic propinquity is not the only criterion to use for selecting appropriate extant analogues (e.g. Aiello et al. 2000; Jolly, 2001; Plavcan, 2002). Nonetheless, we believe that it is unlikely that reliable taxonomic decisions about the mandibular corpora of fossil hominin taxa can be made if mandibular corpus morphology is not taxonomically informative in closely related extant taxa.

In the second part of this study, we employed a novel probabilistic approach to assess within- and between-group variation of mandibles assigned to several species of Homo, including H. habilis sensu stricto, H. rudolfensis, early African H. erectus/H. ergaster, late African H. erectus, Asian H. erectus, H. heidelbergensis, H. neanderthalensis and H. sapiens. More specifically, we used randomization of ‘distinctness values’ (RDV) to examine whether shape variation in mandibular morphology is consistent with conventional taxonomy. Following the argument of Tattersall (1986, p. 166) that ‘what is important in distinguishing among species is between-species variation’, we examined the cohesion of mandibular morphology within a proposed taxonomic group relative to between-group variation. It is worth noting that our purpose was not to overturn conventional species designations but rather to assess the extent to which shape variation of the mandibular corpus can be used as a reliable taxonomic indicator within the hominin clade.

Materials and methods

The extant hominid sample comprises 457 adult individuals (both male and female specimens) representing four genera: Gorilla, Homo, Pan and Pongo (Table 1). All of the comparative specimens are adult, based on the presence of wear facets on M3. Measurements were taken on mandibles with the teeth present or with pristine alveoli; mandibles with substantial alveolar bone resorption [e.g. cases where more than half of the tooth root(s) is exposed] were excluded from the study (cf. Vinter et al. 1996).

Table 1.  List and composition of extant taxa
Extant taxonNMalesFemales
  1. All measurements made by N.J.C. NMNH = National Museum of Natural History.

Total Gorilla:1467967
 Gorilla gorilla beringei 231310
 Gorilla gorilla gorilla 543222
 Gorilla gorilla graueri 693435
Total Homo: 91
 Homo sapiens (Terry Collection, NMNH) 522626
 Homo sapiens (Tel Aviv University) 39
Total Pan:1647787
 Pan paniscus 421725
 Pan troglodytes schweinfurthi 673037
 Pan troglodytes troglodytes 553025
Total Pongo: 562927
 Pongo pygmaeus abelii 12 6 6
 Pongo pygmaeus pygmaeus 442321

The 34 fossil hominin mandibles included in the study are all conventionally included in the genus Homo (Weidenreich, 1936; Day & Leakey, 1973; Wood, 1991; Wood & Richmond, 2000; Rosas, 2001). Based on previous studies, we have divided the fossils into nine taxonomic groups (see Table 2). We recognize that not all researchers will agree on the extent of diversity represented by these specimens (e.g. Wood, 1985, Stringer, 1986; Bilsborough & Wood, 1988; Lieberman et al. 1988; Rightmire, 1990; Miller, 1991; Tobias, 1991; Wood, 1991; Bräuer & Mbua, 1992; Wood, 1992; Bräuer, 1994; Wolpoff et al. 1994; Wood, 1994; Kramer et al. 1995; Grine et al. 1996; Wolpoff, 1996; Rightmire, 1998; Wood & Collard, 1999; Miller, 2000). Nonetheless, as our goal was to test group integrity (based on mandibular corpus morphology), we began our analyses by ‘splitting’ our fossil groups and subsequently ‘lumping’ those for which no mandibular corpus morphological justification for separation could be found (i.e. one cannot assess whether groups are distinct if they are not initially considered as separate groups). For example, although specimens from both Africa and Asia have been assigned to H. erectus, we thought it prudent to maintain the geographic integrity of the samples and initially consider them as separate groups. We also began with a separate Middle Pleistocene ‘late African H. erectus’ group consisting of two specimens whose taxonomic allocation is considered unresolved by some authors (e.g. Rosas, 2001). Finally, we began by considering the Asian H. erectus specimens as two separate groups (Sangiran and Zhoukoudian), as these samples differ both geographically and temporally.

Table 2.  List and composition of fossil taxa
Fossil taxonCodeNMeasured by
  • *

    Measured from casts (Smithsonian Institution, National Museum of Natural History).

  • Amud and Kebara measured from original fossils (Tel Aviv University), remainder measured from casts (Smithsonian Institution, National Museum of Natural History).

  • Data taken from Wood (1991).

Homo sapienssapi11N.J.C.*
Combe Capelle, Choukoutien/PA 101, Choukoutien/PA 104, Choukoutien/PA 109, Eyasi I, FishHoek I, Minatogawa I, Predmosti, Qafzeh 9, Skhul IV, Tabun II
Homo neanderthalensisnean8N.J.C.
Amud, de la Naulette, Kebara H 2, Krapina E, Krapina H, Shanidar I, Spy I, Tabun I
Homo heidelbergensisheid2N.J.C.*
Arago II, Mauer
Asian Homo erectus (Zhoukoudian)zhou2N.J.C.*
Ckn. G1.6, Ckn. H1.12
Asian Homo erectus (Sangiran)sang2B.A.W.
Sangiran 1, Sangiran 9
Late African Homo erectuslahe2B.A.W.
KNM-BK 8518, OH 22
Early African Homo erectus/H. ergastererga2B.A.W.
KNM-ER 730, KNM-ER 992
Homo habilishabi2B.A.W.
OH 13, OH 37
Homo rudolfensisrudo3B.A.W.
KNM-ER 1482, KNM-ER 1483, KNM-ER 1802

Linear dimensions (Table 3) were chosen to capture morphological information from that part of the mandible (i.e. the corpus) that is most often preserved in the early hominin fossil record. Measurements were taken with digital calipers using the measurement definitions given in Wood (1991). In order to maximize the fossil hominin sample size, only eight variables were selected from a larger data set of 19 corpus variables. Interobserver (B.A.W. and N.J.C.) and intraobserver (N.J.C.) measurement errors were < 3% in all cases. Measurements were taken on the original fossils where possible (see Table 2). The values of Weidenreich (1936) for the originals of Ckn. G1.6 and Ckn. H1.12 were not used because Weidenreich employed different landmarks than those employed in this study.

Table 3.  Measurements of the mandibular corpus
  • *

    The number at the beginning of the definitions refers to the measurement number in Wood (1991).

Symphyseal depth142: Maximum depth, at right angles to symphyseal height (Wood, 1991, pp. 295)
Corpus height at P4147: Minimum distance between the most inferior point on the base and the lingual alveolar margin at the midpoint of P4 (Wood, 1991, pp. 295)
Corpus width at P4148: Maximum width at right angles to 147, taken at the midpoint of P4 (Wood, 1991, pp. 296)
Corpus height at M1150: Same as 147 (Wood, 1991)
Corpus width at M1151: Same as 148 (Wood, 1991)
Canine socket (labiolingual length)164: Maximum internal breadth of the canine alveolus in the labiolingual axis (Wood, 1991, pp. 296)
Canine socket (mesiodistal length)165: Maximum internal breadth of the canine alveolus in the mesiodistal axis (Wood, 1991, pp. 296)
P3–P4 alveolar length167: Minimum chord distance between the midpoints of the interalveolar septa between C/P3 and P4/M1 (Wood, 1991, pp. 97)

Assessment of extant taxa: DA

Discriminant function analysis (Klecka, 1980; Rencher, 1995) was applied to the extant specimens to assess whether metrical data from the mandibular corpus can be used to discriminate among extant hominid genera, species, and subspecies. Size-adjusted, or ‘shape’, values were generated by dividing each variable (of a given specimen) by that specimen's geometric mean (cf. Darroch & Mosimann, 1985; Jungers et al. 1995). Taxa included in the genus level analysis were Gorilla, Homo, Pan, and Pongo. The species level analysis was confined to Pan, and involved only P. paniscus and P. troglodytes. The subspecies level analyses examined three sets of subspecies, including three subspecies of Gorilla (G. gorilla beringei, G. gorilla graueri and G. gorilla gorilla), two subspecies of Pan (P. troglodytes schweinfurthi and P. troglodytes troglodytes) and two subspecies of Pongo (P. pygmaeus pygmaeus and P. pygmaeus abelii). We have opted for a conventional taxonomy in the absence of a firm consensus about an alternative (for Gorilla see Groves 1967, 1970, 1989, Albrecht et al. 2003; Thalmann et al. 2007; for Pongo see Groves, 1971; Courtenay et al. 1988; Muir et al. 1998, 2000; Zhang et al. 2001). The DAs were performed using Statistica (Statsoft, Inc.,Tulsa, OK, USA) as well as an algorithm written by M.R.L. for MATLAB software (R2006a, version; The Mathworks, Inc., Natick, MA, USA).

Discriminant analysis is computationally equivalent to manova and, for each analysis, we tested the null hypothesis that there is no difference among groups. We also used ‘structure coefficients’ (i.e. the product-moment correlation between a given variable and a given canonical function) to identify those variables that are most closely associated with group discrimination along a given canonical function. Variables with larger structure coefficients (i.e. ≥ | 0.40 | by convention; also see Schneider, 2006) are considered the most meaningful for group separation. We also used posterior probabilities to assess whether individuals could be correctly assigned to a given taxon.

For purposes of significance testing, additional tests were run in which data were ranked following the rank transformation approach of Conover & Iman (1981). This non-parametric approach relaxes the assumption of normality without significant loss of power. As the results based on ranked vs unranked data do not differ, we present only the latter results here.

Assessment of fossil taxa: randomization analysis (RDV test)

The small sizes of the fossil hominin samples preclude the use of DA for exploring group differences, as significance testing requires more cases within each group than the total number of variables. As an alternative, we employed an RDV test adapted from Sokal & Rohlf (1995). This is a non-parametric probabilistic approach that assesses whether the a priori fossil groups are random with respect to mandibular morphology. The RDV test assesses the cohesiveness of a group of individuals by calculating a ‘distinctness value’ (DV) defined by Sokal & Rohlf (1995, p. 806) as ‘... a measure of homogeneity or cohesion of the members of a group relative to their similarity with other groups’. The DV for a given group is calculated as the average correlation within the group (i.e. average of all pairwise within-group correlation coefficients) minus the average correlation between groups (i.e. average of all possible correlation coefficients between members and non-members). Hence, high positive DVs indicate that the chosen specimens form a distinct group (relative to the other specimens) in which members are more similar in shape to each other than they are to non-members. (Note that the use of correlation coefficients implicitly adjusts for scale, although it does not control for size-related shape variation.) Negative values indicate that members of the chosen group are generally more similar in shape to outside members than they are to one another.

In the procedure described by Sokal & Rohlf (1995), randomization is used to determine whether the DV for a given group can be considered significantly high. We required something different, which was to establish whether there is any morphological justification for a particular group configuration (e.g. 34 fossil Homo specimens divided into nine taxa), i.e. based on the morphometric information that we have captured, do the species designations represent the ‘best’ way to sort n specimens into N groups or are these a priori groups random with respect to the captured morphology? Hence, rather than test the ‘distinctness’ of one group at a time, the algorithm was modified to consider the average distinctness of multiple groups. The average DV is simply the average of the N DVs of the N groups (e.g. average of nine distinctness values of nine groups of fossil Homo). The more ‘distinct’ each particular group is, the greater the value of the average DV. Therefore, a high positive DV indicates that the particular group configuration under consideration is largely supported by the mandibular corpus morphology captured by our eight linear measurements. In contrast, group configurations with negative values indicate that, on average, between-group correlations are higher than within-group correlations and therefore the particular group configuration is not supported by mandibular corpus morphology.

The null hypothesis of the RDV test is that the average DV (of the given group configuration) is not significantly higher than expected by chance alone. The observed average DV is compared with a distribution of average DVs obtained via a randomization procedure. For each iteration, the sample is split into a random number of groups (N ≥ 2), whereby each group has at least two members (the minimum necessary to calculate a within-group correlation coefficient). The average DV is calculated from this random group configuration and the procedure is repeated for up to 10 000 iterations. In those cases where the number of possible novel combinations was less than 10 000 (e.g. there are only 2079 ways to combine eight specimens into multiple groups of two or more), exact randomization was used.

We began by assessing a group configuration in which 34 fossil hominin specimens were divided into the nine groups indicated in Table 2, i.e. we tested whether the average DV based on this a priori taxonomic grouping is significantly higher than one would expect for a random taxonomic allocation in which the 34 specimens are divided into a random number of groups (N ≥ 2) of random sample size (n ≥ 2). We subsequently used the RDV procedure on smaller subsets of the fossil data (e.g. all pairwise comparisons as well as tests suggested either by previous taxonomic hypotheses or by the RDV results themselves). We used principal components analysis (PCA) of group means to visually depict the morphological affinities among the different fossil taxa being tested. The RDV tests and PCA were performed using algorithms written by M.R.L. for MATLAB software (R2006a, version; The Mathworks, Inc.).

It is known that correlation coefficients (r) are distributed in an asymmetrical fashion. To assess the potential effect of such asymmetry on our results, we also ran all of the fossil RDV tests described below using Fisher's z-transformation of r (see Sokal & Rohlf, 1995). Although the use of z instead of r changed the resulting P-values somewhat, without exception, all of the results were the same in terms of whether or not statistical significance (P < 0.05) was observed. Only the results based on untransformed correlation coefficients are presented below.

Validation of the RDV test

In order to validate the method, and compare the results with those obtained using more traditional DA, we applied the RDV test to a number of extant group configurations. Two sets of RDV tests were run using the same extant data: (i) one set of tests using complete samples and (ii) another set of tests using small samples (three to four specimens for each group). The latter set of tests was designed to assess the efficacy of the RDV test on sample sizes similar to those of the fossil taxa. A small number of individuals was randomly chosen from each extant group. Only these randomly chosen individuals were used to calculate the average DV and the randomized distribution. This procedure was repeated 1000 times (using different randomly chosen individuals each time), resulting in a total of 1000 RDV tests for a given group configuration. We then computed the percentage of significant results (P ≤ 0.05) out of these 1000 RDV tests to assess the probability of rejecting the null hypothesis when using small samples. To conserve computing time, we used a maximum of 1000 (rather than 10 000) iterations per RDV test.

Sexual dimorphism

Previous studies have demonstrated that sexual dimorphism is an important component of intraspecific mandibular corpus size and shape variation among hominoids (Wood, 1976; Smith, 1983; Chamberlain & Wood, 1985; Kimbel & White, 1988; Wood et al. 1991; Humphrey et al. 1999; Plavcan, 2002; Taylor, 2006c). Nonetheless, we did not conduct separate analyses of male and female extant specimens, particularly given that the goal of this study was to examine the extent to which mandibular corpus shape is distinct among taxa, including those with substantial sexual dimorphism. As our focus was on mandibular shape, it is of interest to note those groups known to be characterized by significant mandibular shape dimorphism (especially as related to the mandibular corpus), as such taxa may have an impact on our results. In a recent study of hominoid mandibles by Taylor (2006c), Pongo and Gorilla were found to exhibit significant shape dimorphism, although only one of the significant shape dimensions (i.e. corpus width relative to corpus depth at M1) used by Taylor is considered in the present study. As our ability to recognize taxa (species and subspecies) is partly predicated on the extent of shape dimorphism present in any given taxon, it is conceivable that substantial shape dimorphism could compromise our ability to define taxonomic boundaries. Nevertheless, even given significant mandibular corpus shape dimorphism, our analyses will recognize distinct taxa as long as any sex-related shape variation does not exceed taxonomic variation (such that males and females of a given taxon are more similar in shape to one another than to members of other taxa).


Extant hominids: DAs

Genus level

The DA based on the four-genus configuration (Homo, Pan, Gorilla and Pongo) produced a significant result (Λ = 0.04, χ2 = 1497, P < 0.001) and sorted the extant genera with 80% success. On the corresponding plot (Fig. 1a), Homo is distinguished from Pan, Pongo and Gorilla on the first canonical axis, which accounts for the large majority (93%) of the variation among groups. There is con- siderable overlap along the second axis (only 6% of the variation), although overall generalized distances between group centroids are all significantly large (P < 0.001). Based on the structure coefficients of the first discriminant axis (Table 4), mandibular corpus height is relatively larger in extant modern humans (than in non-human hominids) at both P4 and M1, whereas the symphyseal depth of the modern human mandibles is relatively smaller.

Figure 1.

Canonical variates plots based on size-adjusted data for (a) four extant hominid genera and (b) the same taxa with the exclusion of Homo. Genera are indicated as follows: H, Homo; P, Pan; G, Gorilla; O, Pongo.

Table 4.  Structure coefficients for genus-level analysis*
VariablesAll four generaHomo excluded
Axis 1Axis 2Axis 1Axis 2
  • *

    Values > | 0.400 | are shown in bold to highlight the most important discriminating variables.

Symphyseal depth–0.519 0.395 0.362 0.498
Height at P4 0.556 0.121 0.170–0.477
Width at P4 0.186–0.047–0.060 0.578
Height at M1 0.534 0.183 0.238–0.299
Width at M1 0.346 0.078 0.087 0.316
Canine socket (labiolingual)–0.100–0.198–0.187–0.354
Canine socket (mesiodistal)–0.268–0.748–0.729–0.235
P3–P4 alveolar length–0.105 0.499 0.505 0.076
Percentage of explained variation93.1 6.385.414.6

As most of the shape variation among extant genera is accounted for by differences between modern humans and non-human hominids, we ran an additional analysis in which Homo was excluded. Even without the influence of Homo (and despite considerable overlap), the overall result is significant (Λ = 0.43, χ2 = 305, P < 0.001) and distances between group centroids are found to be significantly large (P < 0.001). Specimens were allocated to the correct genus with 74% success. Most of the variation between groups (ca 85%) is accounted for by the first axis, which is similar to the second axis of the previous analysis (which included Homo) in that Pan is somewhat separated from the other two genera (Fig. 1b); as expected, the most influential variables associated with these axes are the same (see Table 4).

Species level

The two species of Pan were found to differ significantly (T2 = 1.389, F = 26.9, P < 0.001) and 91% of the individuals were allocated correctly. Based on structure coefficients (Table 5), P. troglodytes has a relatively smaller P3–P4 alveolar length and a relatively taller symphysis than P. paniscus.

Table 5.  Structure coefficients for species-level and two subspecies-level analyses*
Variables SpeciesSubspecies
  • *

    Values > | 0.400 | are shown in bold to highlight the most important discriminating variables.

  • ‘Group 1’ designated as follows: P. troglodytes for Pan (species), P. t. troglodytes for Pan (subspecies), and P. p. pygmaeus for Pongo.

Symphyseal depth  0.408 0.163 0.135
Height at P4  0.066–0.112 0.586
Width at P4 –0.184 0.083 0.058
Height at M1 –0.158–0.194 0.250
Width at M1  0.053–0.045–0.380
Canine socket (labiolingual)  0.149 0.425–0.296
Canine socket (mesiodistal)  0.034–0.335–0.171
P3–P4 alveolar length –0.567–0.047 0.056
Group centroidsGroup 1 0.69–0.952 0.295
 Group 2–2.00 0.782–1.081

Subspecies level

Tests of the three Gorilla subspecies produced a significant result (Λ = 0.48, χ2 = 102, P < 0.001), with all pairwise intercentroid distances being significantly large (P < 0.001). Specimens were allocated to the correct subspecies with 76% success. The corresponding canonical variates plot is presented in Fig. 2; structure coefficients are provided in Table 6.

Figure 2.

Canonical variates plot based on size-adjusted data for the three gorilla subspecies. Group centroids are indicated with black dots. Subspecies are indicated as follows: g, G. g. gorilla; b, G. g. beringei; i, G. g. graueri.

Table 6.  Structure coefficients for subspecies-level analysis of Gorilla*
VariablesAxis 1Axis 2
  • *

    Values > | 0.400 | are shown in bold to highlight the most important discriminating variables.

Symphyseal depth–0.034–0.456
Height at P4 0.409–0.064
Width at P4–0.353 0.204
Height at M1 0.352 0.331
Width at M1–0.408–0.023
Canine socket (labiolingual)–0.245–0.056
Canine socket (mesiodistal) 0.177 0.092
P3–P4 alveolar length 0.323 0.043
Percentage of explained variation66.633.4

The two subspecies of P. troglodytes were also found to differ significantly (T2 = 0.75, F = 10.68, P < 0.001). The extent of successful allocation to the correct taxon was 80%. Based on structure coefficients (Table 5), discrimination between Pan subspecies appears to be based largely on the shape of the canine socket, which has a relatively larger labiolingual dimension in P. t. troglodytes.

In contrast to the above analyses, no significant difference was found between the two subspecies of Pongo (T2 = 0.33, F = 0.95, P > 0.05), although specimens were allocated with a success rate of 80%. Structure coefficients are listed in Table 5.



The DAs indicate that significant differences in mandibular corpus morphology exist among extant taxa even at the subspecific level (with the exception of Pongo). Hence, it is reasonable to consider whether mandibular corpus morphology is of value for distinguishing fossil hominin taxa. As noted above, before proceeding with RDV tests on the fossil specimens, we applied the RDV test to a number of extant group configurations based on complete data sets, as well as on small random samples.

The RDV results based on complete samples indicate that, in comparison to DA, the RDV test is more conservative (Table 7). None of the three RDV tests based on subspecies indicate a significant difference among groups. In contrast, none of the 1000 randomized DVs for the two species of Pan is higher than the observed value, indicating that these two species are significantly distinct with respect to mandibular corpus morphology. The same result is obtained for a four-group configuration of extant non-human hominids in which all of the subspecies are combined into their respective species (i.e. Gorilla, P. troglodytes, P. paniscus and Pongo).

Table 7.  Results of randomization of ‘distinctness values’ (RDV) tests using extant taxa
Groups testedN of groupsComplete sample sizesSmall samples
Total nAverage DVP value*n per groupTotal nIterations per RDV test% sig. result
  • *

    Based on 1000 iterations.

  • Not applicable. No RDV test necessary for negative average distinctness value.

  • ‡DV = distinctness values.

Pongo subspecies2 56–0.0005N/A3 6  70 0.022
Pan subspecies2122 0.00170.2363 6  70 0.062
Gorilla subspeciesPan species3146 0.00100.2593 91000 0.070
2164 0.010403 6  7020.5
4 8100028.9
Gorilla, P. pan., P. trog., Pongo4366 0.00640312100021.4

For the RDV tests using small random samples (n = 3), one is extremely unlikely to reject the null hypothesis for any of the hominid subspecies; based on 1000 tests for each set of subspecies, the probability of doing so equals < 0.001 (0.1%) in all three cases. The small sample test for the two species of Pan yields a higher percentage of rejected null hypotheses (21%), as does the test for the four-group configuration of non-human hominids noted above (21%). Nonetheless, it is clear that the chances of rejecting a ‘false’ (based on full data) null hypothesis is not particularly large when utilizing small samples. Increasing group sample sizes to four individuals each increases the percentage of significant results (to 29% in both cases), albeit not substantially.

The results of the above tests based on extant data demonstrate that the RDV test is more conservative than DA, even with large samples. At small sample sizes (such as those characterizing the fossils), there is a good chance that the null hypothesis will be falsely accepted unless the groups under consideration are particularly ‘distinct’. Hence, with respect to the small fossil samples examined below, rejection of the null hypothesis for a given RDV test can be accepted as strong evidence that the groups involved are morphologically distinct at a level that is equal to or greater than that of modern hominid species.

Fossil Homo

The average DV for the complete nine-group configuration of fossils (DV = 0.0182) is found to be significantly high based on sampled randomization; only 11 of the 10 000 random group configurations produced an average DV higher than that observed for the nine a priori groups (i.e. P = 0.0011). Hence, there is some morphological support from the mandibular corpus for the way in which specimens have been assigned to fossil Homo taxa.

The above result does not necessarily indicate that all nine groups are significantly distinct from one another but it does suggest that further analysis is worthwhile. We proceeded by considering three logical pairwise comparisons involving groups that have been (or are) considered to be conspecific: (i) the two sets of H. erectus specimens from Asia (Sangiran vs Zhoukoudian), (ii) early African H. erectus/ergaster vs late African H. erectus and (iii) H. habilis vs H. rudolfensis. The sample sizes of all six of the aforementioned groups are very small (n = 2 or 3); hence, these comparisons do not lend themselves to the randomization test (as there are only three possible ways to combine four specimens). Nonetheless, an assessment of the ‘distinctness’ of each group can be made by simply considering the average DV (i.e. is it positive or negative?) and the pattern of correlations. With respect to the latter, we can justify maintaining a distinction between two groups if both (or all three) members of each group are more highly correlated with another group member than they are to members of the alternative group.

The average DV for the comparison between H. habilis and H. rudolfensis is negative (DV = –0.0028). Hence, no justification can be made for maintaining a distinction between these two groups based on our data; in the remainder of this work, these two taxa are referred to collectively as ‘early Homo’.

Although the average DVs for the other two comparisons are positive, their respective correlation patterns are not supportive of a morphological distinction between the two groups in question (Table 8). In the case of the two Asian H. erectus groups (DV = 0.0024), both Sangiran specimens are more highly correlated with one of the Zhoukoudian specimens than they are to one another. Hence, in the analyses below, they are considered as a single ‘Asian H. erectus’ group. With respect to the two African H. erectus groups (DV = 0.0042), only one of the four specimens (KNM-ER 730) is most highly correlated with the specimen belonging to the same group (KNM-ER 992). As such, the two African H. erectus groups were considered as a single group (‘African H. erectus’) in subsequent analyses.

Table 8.  Correlation matrix for select fossil hominin groups
  Sang. 1Sang. 9Ckn. G1.6Ckn. H1.12ER 1482ER 1483ER 1802OH 37OH 13ER 730ER 992BK 8518OH 22
H. erectus (Asian)Sangiran 11            
Sangiran 90.9651           
H. rudolfensisER 14820.9510.9820.9710.9831        
ER 14830.9580.9890.9780.9810.9841       
ER 18020.9780.9780.9870.9930.9860.9801      
H. habilisOH 370.9770.9890.9820.9840.9800.9940.9791     
OH 130.9450.9720.9560.9690.9830.9840.9780.9771    
H. ergasterER 7300.9780.9820.9660.9740.9820.9750.9880.9800.9811   
ER 9920.9620.9760.9650.9770.9890.9850.9860.9840.9960.9901  
Late African H. erectusBK 85180.9350.9320.9300.9430.9550.9590.9670.9520.9830.9670.9841 
OH 220.9350.9750.9500.9630.9840.9900.9730.9790.9970.9770.9930.9781

To visually depict the morphological affinities among the resulting six groups of fossil hominins, we used a PCA (Fig. 3) based on the group means for size-adjusted data. Along the first axis (91% of the variation), the African H. erectus and ‘early Homo’ groups fall to the right side of the graph. Three taxa collectively referred to as ‘later Homo’ (H. sapiens, H. heidelbergensis and H. neanderthalensis) fall to the opposite side of the graph, whereas the Asian H. erectus group has an intermediate position. Interestingly, the African H. erectus group does not show particular affinities with the Asian H. erectus group; the former is on the extreme right side of the graph, separated from the Asian H. erectus group by the two taxa of the ‘early Homo’ group.

Figure 3.

Principal components analysis (PCA) plot of group means for select fossil hominin groups based on the variance-covariance matrix of size-adjusted data.

We performed RDV tests for all possible pairwise comparisons of the six fossil groups noted above (see Table 9). Of particular interest are those pairs of taxa on opposite sides of the PCA plot (i.e. those that should be most ‘distinct’ from one another and are most likely to yield significant RDV results). Based on pairwise testing, all three of the ‘later Homo’ taxa (nean, sapi, heid) are significantly distinct from both of the groups on the opposite side of the plot (i.e. ‘early Homo’ and African H. erectus). The intermediate position of Asian H. erectus (as suggested by the PCA) is also reflected in the randomization results; with respect to pairwise comparisons, Asian H. erectus is not distinct from two of the three ‘later Homo’ groups (i.e. H. neanderthalensis and H. heidelbergensis) or from ‘early Homo’. If we impose a more inclusive two-group configuration of ‘later Homo’ (hsap, nean, heid) vs ‘African Homo’ (rudo, habi, erga, lahe) while excluding the intermediate ‘Asian H. erectus’ individuals, the resulting average DV (0.035) is significantly high (P = 0). Hence, the RDV results complement the visual PCA results and suggest that the fossil specimens can be split into at least two distinct morphological groups (i.e. ‘later Homo’ and ‘African Homo’), within which further testing can be done.

Table 9.  Randomization of ‘distinctness values’ (RDV) results for pairwise comparisons*
  NNumber of iterationsAverage DVP value
  • *

    Taxa included within brackets were analysed as a single group: [zhou, sang]=‘Asian H. erectus’, [rudo, habi]=‘early Homo’, [erga, lahe]=‘African H. erectus’.

  • Not applicable. No test necessary for negative average distinctness value.

  • ‡DV = distinctness value.

neansapi1910 000 0.00110.3329 ns
heid1010 000 0.00250.2854 ns
[zhou, sang]12N/A–0.0011N/A ns
[rudo, habi]1310 000 0.01740.0054 **
[erga, lahe]1210 000 0.03920.0005 ***
sapiheid1310 000 0.00470.0334 *
[zhou, sang]1510 000 0.00530.0449 *
[rudo, habi]1610 000 0.03100.0001 ***
[erga, lahe]1510 000 0.06000.0003 ***
heid[zhou, sang] 6    70 0.00450.0857 ns
[rudo, habi] 7   266 0.01890 ***
[erga, lahe] 6    70 0.04610 ***
[zhou, sang][rudo, habi] 910 000 0.00610.0750 ns
[erga, lahe] 8 2 079 0.02160.0029 **
[rudo, habi][erga, lahe] 910 000 0.002830.2019 ns

Among the ‘African Homo’ group (right side of the PCA plot), the ‘early Homo’ group (habi and rudo) is not found to be distinct (DV = 0.0028, P = 0.202) from the African H. erectus group (erga and lahe). Hence, on the basis of traditional linear measurements of the mandibular corpus, the two groups should be treated as a single group (i.e. ‘African Homo’).

Similarly, the three ‘later Homo’ taxa (left side of the PCA plot) also do not produce a significantly high average DV (DV = 0.0031, P = 0.117) when compared in a three-group configuration (Table 10). The low average DV for this configuration, however, appears to be due mainly to the influence of H. neanderthalensis, which, unlike the other two species of ‘later Homo’, has a negative DV. Indeed, pairwise comparisons (Table 9) indicate that although H. neanderthalensis is not significantly distinct from either H. sapiens or H. heidelbergensis, the latter two species are significantly different from one another (P = 0.033). Based on our eight dimensions, shape variation of the H. neanderthalensis mandibular corpus appears to overlap that of H. sapiens, H. heidelbergensis and Asian H. erectus.

Table 10.  Results of additional randomization of ‘distinctness values’ (RDV) tests*
Groups includedN (groups)n (specimens)Average DVP value
  • *

    10 000 iterations used in all cases.

  • Taxa included within brackets were analysed as a single group: [sapi, heid, nean]=‘later Homo’; [zhou, sang]=‘Asian H. erectus’, [habi, rudo, erga, lahe]=‘African Homo’.

  • ‡DV = distinctness value.

All nine groups9340.01820.0011 **
‘Later Homo’ group
sapi, heid, nean3210.00310.1173 ns
Tests with ‘Asian H. erectus’[zhou, sang]
[zhou, sang], [habi, rudo, erga, lahe]2130.01220.0049 **
[zhou, sang], [sapi, heid, nean]2250.00220.1836 ns
[zhou, sang], [sapi, heid]2170.00450.0489 *
[zhou, sang], sapi, heid3170.00530.0269 *
Tests with ‘African Homo
[habi, rudo, erga, lahe], sapi2200.04310 ***
[habi, rudo, erga, lahe], heid2110.03030 ***
Final four ‘distinct’ groups
sapi, heid, [zhou, sang], [habi, rudo, erga, lahe]4260.02150.0012 **

As noted above, the PCA (and pairwise RDV tests) suggests that Asian H. erectus is morphologically intermediate between the ‘later Homo’and ‘African Homo’ groups. Comparison of the ‘African Homo’ and Asian H. erectus groups (in a two-group configuration) yields a significantly high average DV (P = 0.005; Table 10). In contrast, when Asian H. erectus is compared with the combined ‘later Homo’ group (in a different two-group configuration), the average DV is non-significant (P = 0.184). However, when the Neanderthals are removed from the ‘later Homo’ group (for reasons described above), the average DV is higher and significant (P = 0.049) in a similar two-group test. In addition, the average DV of a three-group configuration consisting of H. sapiens, H. heidelbergensis and Asian H. erectus (but not H. neanderthalensis) is also significantly high (P = 0.027). The pairwise RDV test between Asian H. erectus and H. sapiens indicates that these two groups are also significantly distinct. Although the same cannot be said for the pairwise test between Asian H. erectus and H. heidelbergensis, the associated average DV does have a very low probability of being randomly sampled (P = 0.086; Table 9). Hence, among the ‘later Homo’ species, although H. neanderthalensis is not distinct from either H. heidelbergensis or H. sapiens, the latter two species are distinct from one another, as well as from Asian H. erectus.

In summary, evidence from the randomization analyses suggests that shape variation of the mandibular corpus, based on traditional linear measurements, is not random among fossil Homo taxa. Among the 34 fossil mandibles examined here, there is sufficient evidence to recognize at least four distinct groups [i.e. modern H. sapiens, H. heidelbergensis, Asian H. erectus and a group consisting of exclusively African species (H. habilis, H. rudolfensis, early African H. erectus/H. ergaster and late African H. erectus)]. With the exception of the comparison between H. heidelbergensis and Asian H. erectus (P = 0.086), these four groups are all significantly distinct from one another (Tables 9 and 10). A four-group RDV test based on the above four groups also yields a significantly high average DV (P = 0.001; Table 10).

In contrast, we cannot reject the null hypothesis that the four exclusively African taxa are random with respect to mandibular corpus morphology. In addition, although the mandibular corpora of H. neanderthalensis are found to be distinct from those of the ‘African Homo’ group, they cannot be distinguished from those of H. sapiens, H. heidelbergensis or ‘Asian H. erectus’.

To provide a visual summary of the RDV results, we performed a canonical variates analysis (CVA) based on shape data for the four ‘distinct’ groups noted above (Fig. 4a), each of which was entered into the analysis as an a priori group. The majority of variation among groups is accounted for by the first axis (88.5%), with extensive overlap along the second axis (with the exception of H. heidelbergensis). Structure coefficients (Table 11) indicate that H. sapiens and H. heidelbergensis differ from the other three fossil hominin groups in that the mandibular corpus is relatively higher at both P4 and M1, and the P3–P4 alveolar length is relatively smaller. As expected, when Neanderthals are added as a sixth group by projecting them onto the same canonical vectors (Fig. 4b), they show substantial overlap with H. sapiens, H. heidelbergensis and Asian H. erectus.

Figure 4.

Canonical variates plot based on size-adjusted mandibular data for four ‘distinct’ groups of fossil hominins, presented alone (a) and with Neanderthals projected onto the same canonical vectors (b). The fossil specimens were configured into four groups based on the results of randomization of ‘distinctness values’ (RDV) tests, which also indicate that Neanderthal mandibular morphology is not significantly distinct from that of H. sapiens, H. erectus or H. heidelbergensis. The ‘African Homo’ group consists of specimens assigned to H. habilis, H. rudolfensis and early African H. erectus/H. ergaster.

Table 11.  Structure coefficients for canonical variates analysis (CVA) of the four ‘distinct’ fossil groups depicted in Fig. 4a (based on ‘shape’ data) *
VariablesAxis 1Axis 2
  • *

    Values > | 0.400 | are shown in bold to highlight the most important discriminating variables.

Symphyseal depth 0.177 0.523
Height at P4–0.501–0.210
Width at P4 0.351 0.338
Height at M1–0.487–0.112
Width at M1 0.291–0.086
Canine socket (labiolingual)–0.053 0.004
Canine socket (mesiodistal)–0.205–0.117
P3–P4 alveolar length 0.442–0.292
Percentage of explained variation88.5 8.8


The results of our DAs indicate that the hominid mandibular corpus is taxonomically informative, even when its morphology is captured by eight ‘low-tech’ traditional linear measurements commonly preserved in the hominin fossil record. To some extent, we can successfully discriminate extant hominids at the genus, species and subspecies levels (with the exception of the Pongo subspecies). These results are consistent with those of previous studies that used a broader set of mandibular variables (including measurements of the ramus and different corpus measurements; Humphrey et al. 1999; Taylor & Groves, 2003). As noted in the Introduction, it is theoretically possible that mandibular corpus shape dimorphism may be substantial enough to obfuscate the detection of distinct taxa, especially at the subspecific level. Nonetheless, despite reports of significant mandibular shape dimorphism in Gorilla (Taylor, 2006c), we are able to distinguish among three gorilla subspecies with moderate success (using DA but not RDV). Our inability to differentiate between the orangutan subspecies may be partly related to the relatively small sample sizes for P. p. abelli or perhaps to significant mandibular shape dimorphism (Taylor, 2006c), although an analysis of sexual dimorphism in Pongo is beyond the scope of this study.

Analyses based on the extant data demonstrate that our RDV test is more conservative than DA, even at similar sample sizes. For example, although the complete-sample RDV results mirror the DA results in recognizing morphological differences among hominids at the generic and species levels, RDV does not return statistically significant results for analyses at the subspecific level. In addition, at small sample sizes (three to four individuals per group), it is likely that the null hypothesis of RDV will be falsely accepted unless the groups under consideration are particularly ‘distinct’ (e.g. only 29% of 1000 RDV tests indicated a significant distinction among the four hominin genera when sample size was set to four). As such, with respect to our fossil samples, we believe that any significant results obtained from the RDV analyses provide strong evidence for morphological distinction among groups, probably at a level similar to (or greater than) that of extant congeneric hominid species (e.g. P. troglodytes and P. paniscus).

Evidence from RDV of hominin fossil samples suggests that the mandibular corpus does have limited taxonomic utility. Among the 34 fossil Homo mandibles examined here (potentially divisible into nine groups), we can recognize at least four unambiguously distinct groups: (i) modern H. sapiens; (ii) H. heidelbergensis; (iii) Asian H. erectus and (iv) ‘African Homo’. The latter combined group (iv) consists of four predominately African species (H. habilis, H. rudolfensis, early African H. erectus/H. ergaster and late African H. erectus) and we cannot reject the null hypothesis that these ‘African Homo’ taxa are random with respect to mandibular corpus morphology (hence, their collection into a single group). In addition, although the mandibular corpora of H. neanderthalensis are found to be distinct from those of the ‘African Homo’ group, they cannot be distinguished from those of H. sapiens, H. heidelbergensis or ‘Asian H. erectus’.

We note that our results are contingent upon the small number of corpus linear dimensions that we chose to measure; alternative variables may have produced significant results in those cases where the null hypothesis was accepted. For example, it is well established that modern humans differ from other hominins in the presence of a mental eminence and, although we did not measure this feature, it should serve to distinguish H. sapiens from the other specimens. In addition, although it has been argued that H. neanderthalensis can be defined by a suite of mandibular characters (see Rosas, 2001), our limited set of variables did not include most of the characters commonly cited as distinctive.

Of all of the fossil species in our sample, H. erectus sensu lato samples the longest time period and has the widest geographic distribution. Specimens range in time from as early as 1.7 mya (KNM-ER 730; Feibel et al. 1989) to as late as ~200 kya (Zhoukoudian specimens G1.6 and H1.12; Antón, 2003). Not only does this fossil group span a time depth of over 1.5 million years, it also has an extensive geographic distribution ranging from Africa to China and Indonesia. Various researchers have examined the integrity of H. erectus sensu lato with the understanding that the extreme time depth and geographic distribution sampled by the taxon may introduce variation that is too great for a single hominin species (Groves & Mazák, 1975; Tyler, 1991; Bräuer & Mbua, 1992; Bräuer, 1994; Rightmire, 1998; Antón, 2002; Kidder & Durband 2004). Our own results suggest that mandibular corpus morphology does not allow us to discriminate between Chinese and Indonesian H. erectus specimens, nor are we able to recognize any diagnostic differences between early and late African H. erectus. In contrast, the division between African and Asian H. erectus specimens is supported by our analyses.

Another issue in fossil Homo taxonomy that has long been of interest concerns the scope of the hypodigm of H. habilis. Arguments have been made both for (e.g. Alexeev, 1986; Wood, 1991) and against (e.g. Howell, 1978; Tobias, 1985) the division of the H. habilis sensu lato hypodigm. Based on our limited eight-variable data set, the morphology of the mandibular corpus does not provide any additional evidence for splitting the H. habilis hypodigm. However, some of the mandibular morphology that is said to differ between H. habilis and H. rudolfensis (e.g. the size of the alveolar planum; Lieberman et al. 1996) is not captured by our analysis and much of the argument for splitting the H. habilis hypodigm is based on dental and other non-mandibular cranial data.

We deliberately limited the scope of this study to simple traditional linear measurements that can be taken by any researcher on the part of the mandible (i.e. the corpus) that is best represented in the hominin fossil record. We also deliberately focused on fossil taxa within a single genus and did not compare the mandibles of more distantly related taxa such as H. ergaster and P. boisei (which differ markedly in size and shape), nor did we investigate mandibular corpus variation more widely within and among fossil hominoid taxa. Despite constraining the study in these ways, it is clear that these commonly available data do carry a taxonomic signal and we anticipate that such a signal will have at least some taxonomic utility within and among fossil hominoid taxa. Although linear measurements of the mandibular corpus do not distinguish among all fossil Homo taxa (particularly not among early African specimens), they do distinguish among several of them. These results imply that, to a limited extent, it is possible to test hypotheses concerning hominin alpha taxonomy using the mandibular corpus, particularly when new discoveries increase the fossil sample sizes.


We would like to thank Linda Gordon, David Hunt and Rick Potts (National Museum of Natural History, Smithsonian Institution) for providing access to fossil casts and primate skeletal remains. We are grateful to Bruce Latimer and Lymen Jellema (Cleveland Museum of Natural History) and Wim Van Neer (Royal Museum of Central Africa, Tervuren, Belgium) for access to the primate collections in their care. Thanks to Yoel Rak for allowing access to the fossil specimens and archaeological human skeletal remains stored at Tel Aviv University, Israel, and for his comments on a previous version of this manuscript. Thanks also to Mark Collard, Robbin Chatan, Susan Antón, Dan Lieberman, and several anonymous reviewers for their comments. This research was funded in part by the Louis B. Cotlow Research Fund, The Henry Luce Foundation, the University of Illinois at Urbana-Champaign and Lee Silverman.