AMERICAN JOURNAL, OF PHYSICAL ANTHROPOLOGY 95:53-62 (1994) Craniometric Variation Among Modern Human Populations JOHN H. RELETHFORD Department of Anthropology, State Uniuersity of New York College at Oneontu, Oneontu, New York 13820 KEY WORDS Craniometrics, Genetic differentiation, Human variation, Modern human origins ABSTRACT Previous studies of genetic markers and mitochondrial DNA have found that the amount of variation among major geographic groupings of Homo sapiens is relatively low, accounting for roughly 10% of total variation. This conclusion has had implications for the study of human variation and consideration of alternative models for the origin of modern humans. By contrast, it has often been assumed that the level of among-group variation for morphological traits is much higher. This study examines the level of among-group variation based on craniometric data from a large sample of modern humans originally collected by W.W. Howells. A multivariate method based on quantitative genetics theory was used to provide an estimate of FST - a measure of among-group variation that can be compared with results from studies of genetic markers. Data for 57 craniometric variables on 1,734 crania were analyzed. These data represent six core areas: Europe, SubSaharan Africa, Australasia, Polynesia, the Americas, and the Far East. An additional set of analyses was performed using a three-region subset (Europe, Sub-Saharan Africa, and the Far East) to provide comparability with several genetic studies. The minimum FST (assuming complete heritability) for the three-region analysis is 0.065, and the minimum F S T for the six-region analysis is 0.085. Both of these are less than the average FST from genetic studies (average estimates of 0.10-0.11). The smaller value of the minimum F S T estimates is expected since it provides an estimate of FsTexpected under complete heritability. Using an estimate of average craniometric heritability from the literature provides an estimate of FST of 0.112 for the three-region analysis and 0.144 for the six-region analysis. These results show that genetic and craniometric data are in agreement, qualitatively and quantitatively, and that there is limited variation in modern humans among major geographic regions. 0 1994 Wiley-Liss, Inc. In recent years increasing attention has been paid to the implications of patterns of modern human variation for understanding the origin and evolution of modern humans (Stringer and Andrews, 1988). One important finding has been the relatively low level of variation among geographic regions compared to high levels of variation within regions. A variety of data (genetic markers, nuclear DNA, and mitochondrial DNA) all suggest low levels of population differentiation in modern humans, often less than 0 1994 WILEY-LISS, INC. found among subspecies of other organisms (e.g., Lewontin, 1972; Latter, 1980; Nei and Roychoudhury, 1982; Ryman et al., 1983; Livshits and Nei, 1990; Vigilant et al., 1991; Stoneking, 1993; Takahata, 1993). These findings have had several implications for the study of human variation and human evolution. First, such low levels of population differentiation imply that the vast maReceived October 24,1993; accepted March 7,1994 54 J.H. RELETHFORD jority of genetic variation is among individuals within groups, and not due to variation among groups. As such, these findings show little support for the concept of race or racial classifications (Lewontin, 1972, but see Nei and Roychoudhury, 1982). Second, some authors have suggested the low level of population differentiation supports a relatively recent divergence of modern humans into distinct geographic regions-the “Garden of Eden” or replacement model (e.g., Cann et al., 1987; Stringer and Andrews, 1988; Rouhani, 1989).An alternative hypothesis, multiregional evolution, posits an earlier divergence of Homo erectus approximately 1 million years ago (e.g., Wolpoff, 1989). Here, the low level of differentiation is seen as due to relatively high rates of gene flow among geographic regions. To date, all studies of population differentiation on a worldwide scale have used data from genetic markers (blood groups, serum proteins and enzymes, and so forth) and mitochondrial DNA (mtDNA). Differentiation of complex traits, such as cranial and facial measurements, has seldom been addressed in this context. Many discussions assume greater morphological differentiation relative to genetic differentiation. Nei and Roychoudhury, for example, state that morphological variation among major races is “conspicuous”(1982: 40). Stringer and Andrews state that our species “shows great morphological variation. However, in contrast to this, genetic variation between human populations is low overall.” (1988: 1264). These and other statements presume extensive worldwide morphological variation. One notable exception is Howell’s (1989) study of worldwide craniometric variation. On the basis of comparisons of modern crania with several archaic forms, he concluded that craniometric variation among modern Homo sapiens is fairly limited. However, he performed no formal comparison of differentiation based on craniometrics and genetic markers. The study of quantitative traits has for many decades followed what Howells (1973b)termed a model-free approach. Here, traditional statistical methods are applied and interpreted in light of general predictions from population genetics theory. This approach contrasts with the model-bound approach often used in analysis of genetic markers, where specific parameters of models are estimated. The past decade or so has seen a resurgence in the study of quantitative traits in anthropology, focusing more and more on model-bound approaches and parameter estimation (e.g., Relethford and Lees, 1982; Rogers and Harpending, 1983; Williams-Blangero et al., 1990). Recent developments in methodology allow estimation of parameters of microevolutionary change from quantitative traits, which can then be compared with similar estimates from genetic markers and mtDNA. The purpose of this paper is to present estimates of the degree of population differentiation among world regions based on craniometric data. These estimates provide an index of the relative degree of variation among and within regions. These estimates are then compared to typical values found from studies of genetic markers and mtDNA. The results indicate that the degree of differentiation is essentially the same in both genetic markers and craniometric traits. These results lend further support to a relatively recent divergence of modern humans, and also suggest that multivariate measures of average worldwide craniometric variation may primarily reflect a balance between gene flow and genetic drift. MATERIALS AND METHODS The data used in this study were graciously provided by Dr. W.W. Howells and are from his long-term study of craniometric variation in modern humans (Howells, 1973a, 1989).Portions o r all of Howells’data have been used in a variety of studies focusing on within-group and among-group craniometric variation of modern humans (e.g., Guglielmino et al., 1977; Lynch, 1989; Franciscus and Long, 1991; Konigsberg and Blangero, 1993; Kramer, 1993; Relethford and Harpending, 1994). The present study uses data on 57 cranial and facial measurements taken on 1,734 modern human crania (907 males and 827 females) from six world regions: Europe, Sub-Saharan Africa, Australasia, Polynesia, Americas, and the Far East. Howells chose three local populations CRANIOMETRIC VARIATION AMONG MODERN HUMANS TABLE 1 . Samule sizes Region Europe Sub-Saharan Africa Australasia Polynesia Americas Far East Total 55 TABLE 2. List of craniometric variables Males Females Total 164 135 153 157 148 150 907 153 148 145 137 133 111 827 317 283 298 294 281 261 1734 from within each of these six regions for maximum representation of average patterns of variability within each region. This sampling strategy avoids potential problems that might come from reliance on a single local population to represent a geographic region. Even given different locations and time periods for local populations, Howells’ (1989) analyses show that the local samples cluster together relative to differences among regions. Dates for samples range from the early 20th century to 2,000 years ago. Even “recent” samples were chosen to provide the best representation of geographic regions as might have existed prior to 500 years ago; thus, the samples are not influenced to any great extent by the dramatic changes in migration and population size in Homo sapiens (Howells, 1989). Sample sizes are reported in Table 1.A complete list of variables used is provided in Table 2. The degree of population differentiation is most often expressed as the ratio of amonggroup variation to total variation expected under panmictic conditions. For genetic markers, the proportion of total diversity attributed to variation among regions is often computed as Glabello-occipital length Nasio-occipital length Basion-nasion length Basion-bregma height Maximum cranial breadth Maximum frontal breadth Bistephanic breadth Bizygomatic breadth Biauricular breadth Minimum cranial breadth Biasterionic breadth Basion-prosthion length Nasion-prosthion height Nasal height Orbit height, left Orbit breadth. left Bigugal breadth Nasal breadth Palate breadth, external Mastoid height Mastoid breadth Bimaxillary breadth Zygomaxillary subtense Bifrontal breadth Nasio-frontal subtense Biorbital breadth Dacryon subtense Interorbital breadth Naso-dacrval subtense Simotic chord Simotic subtense Malar length, inferior Malar length, maximum Malar subtense Cheek height Supraorbital projection Glabella projection Foramen magnum length Nasion-bregma chord Nasion-bregma subtense Nasion-subtense fraction Bregma-lambda chord Bregma-lambda subtense Bregma-subtense fraction Lambda-opisthion chord Lambda-opisthion subtense Lambda-subtense fraction Vertex radius Nasion radius Subspinale radius Prosthion radius Dacryon radius Zygoorbitale radius Frontomalare radius Ectoconchion radius Zygomaxillare radius M1 alveolus radius See Howells (1973a, 1989) for additional description number of studies have since applied this (or a similar) method to genetic data from human populations. These studies, based on a larger number of loci, suggest a value of roughly 10% for among-group variation (Latter, 1980; Nei and Roychoudhury, 1982; Ryman et al., 1983; Livshits and Nei, 1990). Similar results have been found for mitochondrial DNA after adjustment for a maternal mode of inheritance-two studies reported by Takahata (1993) give an average value of 0.097. The partitioning of variances using equation (1)cannot be applied directly t o quantiHs - Hr (1) tative traits because total genotypic or pheHs ’ notypic variation (all groups pooled) is not where Hs is the heterozygosity expected if the same as that expected under panmixia. the entire species were a single panmictic The total genotypic variance of a total sampopulation and Hr is the average heterozy- ple (such as a species) reflects a combination gosity within regions. Lewontin (1972) was of within-group and among-group variation one of the first to use this method on genetic (Relethford and Blangero, 1990). An altermarkers from major human regions native method based on another method of (“races”)and found an average value across genetic analysis is required. Equation (1) loci of 0.063; that is, roughly 6% of species- and similar derivations provide an estimate wide variation was due to among-group of Wright’s (1951) familiar FST measurevariation, while almost 94% of total varia- the ratio of among-group variation to total tion was due to within-group variation. A variation expected under panmixia (note 56 J.H.RELETHFORD that the quantity computed in equation (1) also goes by a variety of other names, such as f GsnR , , r,, and others). An estimate of FST can be computed from the v a r i a n c s o variance matrix of population relationships, better known as an R matrix. The diagonal elements of this matrix, rLi,represent the genetic distance of population i to a regional centroid, defined by the mean allele frequencies over all populations being studied. The diagonal elements are computed as where p i is the allele frequency for population i and p is the mean allele frequency averaged over all populations. The rii value is then averaged over all alleles, and FST is estimated as the average rii value over all populations, weighted by population size (Harpending and Jenkins, 1973; Workman et al., 1973). The higher the value ofFST, the greater the variation around the contemporary allele frequencies, indicating greater differentiation. F S T can be estimated from quantitative traits using an equal and additive effects model of polygenic inheritance (WilliamsBlangero and Blangero, 1989; Relethford and Blangero, 1990). Following standard quantitative genetics theory for loci with two alleles (e.g., Falconer, 19811, let a, 0, and -a represent the genotypic values of the three genotypes a t any given locus. This model also assumes that these genotypic values are the same across all loci (Falconer, 1981; Rogers and Harpending, 1983). The phenotypic means can then be written as xi= 2api for population i, and i = 2aJ However, this value cannot be estimated directly from data (Relethford and Blangero, 1990) and must be estimated instead from the relationship (61 whereg, is the pooled within-group additive genetic variance (Rogers and Harpending, 1983). Substituting equations (3) and (4) for the numerator in equation (2), and substituting equations (5) and (6) for the denominator in equation (2), gives where The total mean (x) and the pooled withingroup additive genetic variance &), are both obtained as weighted estimates over all samples, and the weighting is by population size. (Note: The definition of cii given here differs from that originally given by Relethford and Blangero [19901). The multivariate extension of this method, based on g groups and t traits, presented by Relethford and Blangero (1990), is used here. First, all data are converted into standardized scores by variable. Second, a g by t matrix is computed consisting of deviations of group means from the total means pooled over all populations (A). Third, the pooled within-group additive genetic vari(3) ance-covariance matrix is computed (G). Fourth, a codivergence matrix ( C ) is computed as (4) for the total sample. The total additive genetic variance (under the assumption of panmixia) can also be written in terms of the allele frequencies, as c = AG-~A', (9) where G is the pooled within-group additive genetic variance<ovariance matrix, and the prime ('1 indicates matrix transposition. This matrix is then divided by t to give an average value over all traits (C/t).Since FST is defined as CRANIOMETRIC VARIATION AMONG MODERN HUMANS 57 FST The lowest possible value of FsToccurs substitution of the diagonal elements of Clt for ciiin equation (7)and combining with equation (10) gives Solving for F S T gives The weighting factor, w i ,is usually defined in terms of relative census size, but in the case of the present analysis these values are unknown during recent human evolution. Thus, this value is set equal to llg (equal weighting of all populations). While this might not be an accurate reflection of past population sizes, it does offer comparability with estimates obtained from genetic markers. The weighting factor is also used in computing the variance-covariance matrix and the matrix of pooled means (Relethford and Blangero, 1990). Note that FSTas used here (and in studies of genetic markers) refers to variation around a contemporary array of allele frequencies, and not a hypothetical array of ancestral allele frequencies, which are never known. Computation of the pooled within-group additive genetic variance-covariance matrix (G) requires information on heritabilities. If these are not available, the pooled withingroup phenotypic variance-covariance matrix (V) can be substituted to provide an estimate of the minimum FST value (WilliamsBlangero and Blangero, 1989; Relethford and Blangero, 1990). This method assumes that all heritabilities are equal to 1(G = V), and that the additive genetic covariance matrix is proportional to the phenotypic covariance matrix. That this value is a minimum value is clear from consideration of equations (9) and (12). For any given trait, G = h v where h2 is the heritability of the trait. As h2 increases C decreases, and so does when h2 = 1(G = V). Minimum FST is a conservative statistic; it implies that actual genetic differentiation is at least as great as that estimated under the assumption that G = V. As such, the minimum F S T should be less than an estimate ofFsT computed from genetic markers. The amount less reflects the average heritability over all traits. A rough approximation of the actual F S T value can be made using estimates of average heritability of craniometric traits as reported in the literature (e.g., Devor, 1987). This rough approximation assumes a high correspondence between genetic and phenotypic correlation matrices, which is supported in a review of quantitative genetics analyses (Cheverud, 1988). The relationship between the number of traits and the number of loci is affected by the intercorrelation of variables. Rogers and Harpending (1983) show that one completely heritable quantitative trait is expected t o provide the same information as a single locus trait with two alleles. If traits are highly intercorrelated the effective number of loci decreases. While methods exist for estimating the effective number of loci (Williams-Blangero and Blangero, 1992), they require detailed pedigree data, which are not available for this study. The methods described here were applied to Howells’ data to determine the minimum FST values for major geographic regions. Comparative data from genetic markers for major geographic groups usually focus on six groups (roughly similar to the ones used here) or the three “major races” of humans“Caucasoid,” “Negroid,” and “Mongoloid.” For comparability, I have computed the minimum FST for all six of Howells’ core areas as well as for the three geographic regions usually associated with the “major races”-Europe, Sub-Saharan Africa, and the Far East. The match between populations across studies is not perfect, but it is certainly useful enough for the purpose of rough comparison of craniometrics and genetic markers. Preliminary analyses were performed separately by sex. Analyses were also performed pooling the two sexes, which pro- J.H. RELETHFORD 58 TABLE 3. Minimum FST values for craniometr‘ic traits Number of regions Regions included Male Female Europe Sub-Saharan Africa Far East Europe Sub-SaharanAfrica Australasia Polynesia Americas Far East 0.076 (0.002) 0.063 (0.002) 0.065 (0.001) 0.093 (0.002) 0.085 (0.002) 0.085 (0.001) Pooled These are minimum possible FST values obtained using complete heritability. For comparison with other FST values, such as from genetic markers, these values must be adjusted for average heritability as described in the text. Standard errors are in parentheses. DISCUSSION How do these estimates compare to genetic studies? For maximum comparison, I selected studies sampling many loci and closely equivalent in terms of geographic regions, both in number and composition. Several genetic studies have looked at variation among the three geographic regions of Europe, Sub-Saharan Africa, and East Asia. Nei and Roychoudhury (1982) examined diversity among the three “major races.” These groupings correspond fairly closely with geographically defined units, except for some exceptions such as the inclusion of samples from India in the “Caucasoid (= European) group. They found a value of RESULTS 0.088 based on 62 protein loci and a value of The minimum FSTvalues and their stan- 0.109 based on 23 blood groups. The average dard errors are reported in Table 3, sepa- of these two data sets is 0.099. Livshits and rately by sex and pooled, for both three-re- Nei (1990) performed a similar analysis degion and six-region analyses. The standard fined more in terms of geographic regions errors are all very small relative to the FST than traditional racial groupings. On the bavalues, indicating highly significant among- sis of 86 protein and enzyme loci, and 33 group variation. Depending on sex and the blood group loci, they found an FSTvalue of number of regions in the analysis, the mini- 0.114.Averaging the similar results of these mum FST values range from 0.063 to 0.085. two studies gives a mean estimate of 0.107. Contrary to some opinion, craniometric vari- The minimum FST value from craniometrics ation across regions within our species is is less than this, as expected if heritabilities fairly limited. These estimates confirm are less than 1. Howells’ (1989) impression of limited variHow much less? Given estimates of miniability in modern humans. The minimum mum FsdMin FsT)from metric traits and FST values are somewhat higher for males. an estimate of average heritability, the estiIn any case, the pooled minimum FST values mate of genetic FsT is equal to do not differ much from either sex, particularly the females. All discussions for the reMin FST mainder of this paper will focus on the (13) FST = Min FsT + h2(1 - Min FST) pooled minimum FsT values. vides better correspondence with genetic markers that are most frequently collected for both sexes. To avoid problems in sexrelated size variation, all data were first standardized within each sex prior t o pooling, a common approach in studies of quantitative differentiation (Williams-Blangero and Blangero, 1989, 1990). The FST values were corrected for sample size bias as described by Relethford (1991). Standard errors for minimum FST were derived according to analytic formulae developed recently by John Blangero (personal communication). CRANIOMETRIC VARIATION AMONG MODERN HUMANS [derived from formulae presented by Relethford and Blangero (199011. Devor (1987) reports average heritabilities for craniometric traits from four populations based on path analysis. The average heritability ranges from 0.48 to 0.61 across these populations, with an overall average of 0.55. This estimate is conservative since a number of craniometric traits measured on living subjects involve soft tissue that is apt to be affected more by environmental influences than hard tissue (e.g., nose breadth), so that heritabilities for skeletal measures are likely to be higher (Devor, 1987). Using an estimate of h2 = 0.55 for the three-region analysis (Min FST = 0.065), the estimated value of F S T is 0.112, which is virtually identical with the estimate from genetic markers using three geographic regions (0.107). For comparison with the six-region study, I selected two genetic studies that focused on six roughly comparable samples of populations. Ryman et al. (1983) looked at genetic diversity for 25 loci (including proteins, blood groups, and HLA markers). Their groupings are roughly similar to those used here, except for the use of a “Caucasoid group for Europe and a more general sample from Oceania rather than one exclusively from Polynesia. Their estimate of FST is 0.099. Another six-region analysis was performed by Latter (1980),who pooled Australasia and Polynesia into an Oceanic grouping, and also added a separate group representing the Middle East and India that is not represented in the craniometric data set. Latter used 18 loci (10 blood groups, 3 proteins, and 5 enzymes) giving an estimate of 0.104. The average of the two studies is 0.102, virtually identical to the average of the three-region genetic analyses. This value is greater than the minimum FST value obtained from craniometrics (0.085). Using an average heritability of 0.55 gives an estimate of FsT = 0.144, which is higher than that found for genetic markers, but still indicates a low level of among-group variation. Of course, the estimated FST values depend on specific choice of average heritability. For the three-region analysis, for example, the estimate of FST ranges from 0.148 59 for an average heritability of 0.4 to 0.090 for an average heritability of 0.7. Using different heritabilities for different traits would change matters as well, although the overall estimate of F S T would still fall within this range for reasonable values of craniometric heritabilities. While we cannot provide an exact estimate Of F S T without greater information on heritability, it is important to note that regardless of the specific average value chosen, all of the above F S T estimates are in rough agreement with the range from genetic studies, and all support the idea of a relatively low degree of among-group variation. The estimates presented here (both minimum FST and estimated FST) are based on a large data set with many variables. However, we must remember that the 57 variables used here are not independent and do not represent 57 independent loci. Howells (1973a) applied factor analysis to a portion of these data and found 18 separate morphometric factors. While I am not claiming any kind of direct correspondence between factors and loci, his analyses do suggest that the large number of traits actually provide information on a fair number of independent traits. In spite of potential problems in sample composition and other methodological concerns, the overall results are quite striking. In both the three-region and six-region analyses the craniometric data show minimum FSTvalues less than found in studies of genetic markers, as expected under the hypothesis that both types of data reflect approximately the same microevolutionary forces. Rough estimates of heritability show the fit of the three-region analysis in particular to be excellent. The results are also in agreement with studies using mitochondria1DNA. Takahata (1993) reports estimates of F S T of 0.31 and 0.46. Since mtDNA is haploid and maternally inherited, these estimates must be divided by 4 for comparison with genetic markers and craniometrics. This division gives F S T = 0.078 and 0.115 (average = 0.097), remarkably similar to the approximate estimate of 10% from both genetic markers and craniometrics, and particu- 60 J.H. RELETHFORD larly interesting since mtDNA has a higher mutation rate than nuclear DNA and is apparently unaffected by selection (Stoneking, 1993). There are several important implications from these results for both contemporary human variation and patterns of human evolution. As noted by Lewontin (1972) and subsequent studies, the low degree of among-group genetic variation relative to total species variation argues strongly against traditional typological racial classifications. While among-group variation is statistically significant in all cases, the overall degree of among-group variation is too low to produce any substantial accuracy in racial classifications. This finding has been suggested by some to be different from the degree of among-group variation in morphological traits (Nei and Roychoudhury, 1982; Stringer and Andrews, 1988), although no formal analysis has accompanied these suggestions. The results presented here indicate that there is limited variation among major geographic or “racial” clusters in modern humans for both genetic and craniometric measures. It might be expected that craniometric variation would also be affected by natural selection andor developmental acclimitization, since a number of studies have shown significant correlations of craniometric traits with various climatic indices (e.g., Beals, 1972; Guglielmino et al., 1977; Beals et al., 1984).While suggestive, these studies are not proof of selection. Guglielmino et al.’s study used a subset of Howells’ data, and noted that a distinction between Africa and Australia, on one hand, and Europe and the Far East, on the other, suggesting size differences corresponding to latitude. Recent analyses, however, suggest that these relationships may reflect differences in population size and genetic drift across geographic regions (Relethford and Harpending, 1994). In any case, the real issue is the meaning of the close correspondence in FSTestimates from genetic markers, mitochondrial DNA, and craniometrics. Does this correspondence reflect similar patterns of population history and structure (gene flow and genetic drift), and/or coincidental correspondence due to natural selection? Gene flow and ge- netic drift are expected to have the same impact on all loci, whereas the effects of selection may vary across loci. In the absence of loci that show particularly large effects of selection, FSTis generally taken as an average index of population relationships brought about by a balance between gene flow and genetic drift. The low level of average among-group variation in modern humans might reflect selection, although it is difficult to construct scenarios by which selection will produce the same degree of variation in so many diverse traits, including mitochondrial DNA, which is considered essentially neutral (Harpending et al., 1993; Stoneking, 1993). Not all traits should be expected to necessarily show the same levels of among-group variation. In particular, individual traits affected strongly by selection might show divergent estimates of FST. For example, I have looked at mean male skin reflectance at 685 nanometers from published reports on 56 populations in Europe, Sub-Saharan Africa, and the Far East. Using these data, the minimum FST is 0.647, which is considerably higher than the level of among-group variation exhibited by craniometrics. Preliminary analyses using averages over additional wavelengths show similar results. The most obvious, and hardly surprising, conclusion is that skin color has been strongly affected by natural selection in such a way that variation among geographic regions is large. The close correspondence in FST estimates across diverse data sets is not necessarily ammunition for the continued selectionistneutralist debate. The issue is not whether genetic markers or craniometrics are affected by selection, but rather what can be inferred from the average level of amonggroup variation. The level of variation among major geographic regions relates to the ongoing controversy regarding the origin of anatomically modern Homo sapiens. Which of the current models, if any, are most compatible with low levels of amonggroup variation? While some authors see low FST values as indicative of a recent divergence from an initial African population, others see those same values as reflecting continued gene flow between geographic regions following a much earlier (1,000,000 CRANIOMETRIC VARIATION AMONG MODERN HUMANS years BP) African origin. At present, there are no definitive conclusions regarding which model best explains the low FST values, although some preliminary analyses have argued against the multiregional model on the grounds that interregional gene flow would have had to be extremely high, andor that effective population size would have to have been much larger than estimated from genetic surveys (e.g., Rouhani, 1989; Harpending et al., 1993; Takahata, 1993). Another criticism is that the high levels of gene flow required by the multiregional model could make regional continuity less likely (Stoneking, 1993). This debate will continue. The results of the present study do not provide a clear choice between the competing models of human origins. They do, however, provide further support for the continued finding of low levels of among-group variation. Any future discussion of the origin of modern humans must deal with this similarity. Contrary to some earlier views, the level of morphological variation in modern humans is low. ACKNOWLEDGMENTS I am most grateful to W.W. Howells for the use of his data. I thank Henry Harpending, W.W. Howells, and Mark Stoneking for their valuable insights. I also thank John Blangero for providing analytic formula for the computation of the standard errors of minimum FST This research was supported in part by a SUNY Graduate Research Initiative Grant awarded to JHR. LITERATURE CITED Beals KL (1972) Head form and climatic stress. Am. J . Phys. Anthropol. 37235-92. Beals KL, Smith CL, and Dodd SM (1984) Brain size, cranial morphology, climate, and time machines. Curr. Anthropol. 25:303-330. Cann RL, Stoneking M, and Wilson AC (1987) Mitochondrial DNA and human evolution. Nature 32531-36. Cheverud J M (1988) A comparison of genetic and phenotypic correlations. Evol. 421958-968. Devor E J (1987) Transmission of human craniofacial dimensions. J. Craniofacial Genet. Dev. Biol. 7:95106. Falconer DS (1981) Introduction to Quantitative Genetics, 2nd ed. London: Longman. Franciscus RG, and Long J C (1991) Variation in human nasal height and breadth. Am. J . Phys. Anthropol. 85r419-427. 61 Guglielmino-Matessi CR, Gluckman P, and CavalliSforza LL (1977) Climate and the evolution of skull metrics in man. Am. J . Phys. Anthropol. 50:549-564. Harpending HC, and Jenkins T (1973) Genetic distance among Southern African populations. In MH Crawford and PL Workman (eds.): Methods and Theories of Anthropological Genetics. Albuquerque: University of New Mexico Press, pp. 177-199. Harpending HC, Sherry ST, Rogers AR,and Stoneking M (1993) Genetic structure of ancient human populations. Curr. Anthropol. 34:483496. Howells WW (1973a) Cranial Variation in Man: A Study by Multivariate Analysis of Patterns of Difference among Recent Human Populations. Papers of the Peabody Museum No. 67. Cambridge, MA: Peabody Museum, Harvard University. Howells WW (1973b) Measures of population distance. In MH Crawford and PL Workman (eds.): Methods and Theories of Anthropological Genetics. Albuquerque: University of New Mexico Press, pp. 159-176. Howells WW (1989) Skull Shapes and the Map: Craniometric Analyses in the Dispersion of Modem Homo. Papers of the Peabody Museum No. 79. Cambridge, MA: Peabody Museum, Harvard University. Konigsberg LW, and Blangero J (1993) Multivariate quantitative genetic simulations in anthropology with an example from the south Pacific. Hum. Biol. 65: 897-9 15. Kramer A (1993) Human taxonomic diversity in the Pleistocene: Does Homo erectus represent multiple hominid species. Am. J. Phys. Anthropol. 91:161171. Latter BDH (19801 Genetic differences within and between populations of the major human subgroups. Am. Natural. 11 6:220-237. Lewontin RC (1972) The apportionment of human diversity. Evol. Biol. 6~381-398. Livshits G, and Nei M (1990) Relationships between intrapopulational and interpopulational genetic diversity in man. Ann. Hum. Biol. 6:501-513. Lynch M (1989) Phylogenetic hypotheses under the assumption of neutral quantitative-genetic variation. Evol. 43:l-17. Nei M, and Roychoudhury AK (1982) Genetic relationship and evolution of human races. Evol. Biol. 14:l59. Relethford J H (1991) Genetic drift and anthropometric variation in Ireland. Hum. Biol. 63r155-165. Relethford JH, and Blangero J (1990) Detection of differential gene flow from patterns of quantitative variation. Hum. Biol. 62:5-25. Relethford JH, and Harpending HC (1994) Craniometric variation, genetic theory, and modern human origins. Submitted MS. Relethford JH, and Lees FC (1982) The use of quantitative traits in the study of human population structure. Yrbk. Phys. Anthropol. 25r113-132. Rogers AR, and Harpending HC (1983) Population structure and quantitative characters. Genetics 105: 985-1002. Rouhani S (1989) Molecular genetics and the pattern of human evolution: Plausible and implausible models. In P Mellars and C Stringer (eds.): The Human Revolution: Behavioural and Biological Perspectives on the 62 J.H. RELETHFORD Origins of Modern Humans. Princeton, NJ: Princeton University Press, pp. 47-61. Ryman N, Chakraborty R, and Nei M (1983) Differences in the relative distribution of human gene diversity between electrophoretic and red and white cell antigen loci. Hum. Hered. 33:93-102. Stoneking M (1993) DNA and recent human evolution. Evol. Anthropol. 260-73. Stringer CB, and Andrews P (1988) Genetic and fossil evidence for the origin of modern humans. Science 239:1263-1268. Takahata N (1993) Allelic genealogy and human evolution. Mol. Biol. Evol. 10:2-22. Vigilant L, Stoneking M, Harpending H, Hawkes K, and Wilson AC (1991) African populations and the evolution of human mitochondria1 DNA. Science 253:15031507. Williams-Blangero S and Blangero J (1989) Anthropometric variation and the genetic structure of the Jirels of Nepal. Hum. Biol. 61:l-12. Williams-Blangero S, and Blangero J (1990) Effects of population structure on within-group variation in the Jirels of Nepal. Hum. Biol. 62131-146. Williams-Blangero S, and Blangero J (1992) Quantitative genetic analysis of skin reflectance: A multivariate approach. Hum. Biol. 64:3549. Williams-Blangero S, Blangero J, and Towne B (1990) Quantitative traits and population structure: Introduction. Hum. Biol62:1-4. Wolpoff MH (1989) Multiregional evolution: The fossil alternative to Eden. In P Mellars and C Stringer (eds.): The Human Revolution: Behavioural and Biological Perspectives on the Origins of Modern Humans. Princeton, NJ: Princeton University Press, pp. 62-108. Workman PL, Harpending HC, Lalouel JM, Lynch C, Niswander JD, and Singleton R (1973) Population studies on southwestern Indian tribes. VI. Papago population structure: A comparison of genetic and migration analyses. In NE Morton (ed.): Genetic Structure of Populations. Honolulu: University Press of Hawaii, pp. 323-354. Wright S (1951) The genetical structure of populations. Ann. Eugen. 15:323-354.