Beyond Gorilla and Pongo: Alternative Models for Evaluating Variation and Sexual Dimorphism in Fossil Hominoid Samples

Transcription

1 AMERICAN JOURNAL OF PHYSICAL ANTHROPOLOGY 140: (2009) Beyond Gorilla and Pongo: Alternative Models for Evaluating Variation and Sexual Dimorphism in Fossil Hominoid Samples Jeremiah E. Scott, 1 * Caitlin M. Schrein, 1 and Jay Kelley 2 1 School of Human Evolution and Social Change, Institute of Human Origins, Arizona State University, Tempe, AZ Department of Oral Biology, College of Dentistry, University of Illinois at Chicago, Chicago, IL KEY WORDS bootstrap; dental variation; Lufengpithecus; Ouranopithecus; Sivapithecus ABSTRACT Sexual size dimorphism in the postcanine dentition of the late Miocene hominoid Lufengpithecus lufengensis exceeds that in Pongo pygmaeus, demonstrating that the maximum degree of molar size dimorphism in apes is not represented among the extant Hominoidea. It has not been established, however, that the molars of Pongo are more dimorphic than those of any other living primate. In this study, we used resampling-based methods to compare molar dimorphism in Gorilla, Pongo, andlufengpithecus to that in the papionin Mandrillus leucophaeus to test two hypotheses: (1) Pongo possesses the most size-dimorphic molars among living primates and (2) molar size dimorphism in Lufengpithecus is greater than that in the most dimorphic living primates. Our results show that M. leucophaeus exceeds great apes in its overall level of dimorphism and that L. lufengensis is more dimorphic than the extant species. Using these samples, we also evaluated molar dimorphism and taxonomic composition in two other Miocene ape samples Ouranopithecus macedoniensis from Greece, specimens of which can be sexed based on associated canines and P 3 s, and the Sivapithecus sample from Haritalyangar, India. Ouranopithecus is more dimorphic than the extant taxa but is similar to Lufengpithecus, demonstrating that the level of molar dimorphism required for the Greek fossil sample under the single-species taxonomy is not unprecedented when the comparative framework is expanded to include extinct primates. In contrast, the Haritalyangar Sivapithecus sample, if it represents a single species, exhibits substantially greater molar dimorphism than does Lufengpithecus. Given these results, the taxonomic status of this sample remains equivocal. Am J Phys Anthropol 140: , VC 2009 Wiley-Liss, Inc. A frequently encountered problem in hominoid paleontology is identification of the source of high levels of size variation in a fossil sample (e.g., Kay 1982a,b; Lieberman et al., 1988; Cope and Lacy, 1992; Albrecht and Miller, 1993; Kramer, 1993; Martin and Andrews, 1993; Richmond and Jungers, 1995; Lockwood et al., 1996, 2000; Plavcan and Cope, 2001; Silverman et al., 2001; Scott and Lockwood, 2004; Villmoare, 2005). While some sources are relatively easily identified and controlled (e.g., variation due to ontogeny or pathology), others present greater difficulty. For example, high variation in a single fossil sample can be interpreted as evidence of the presence of multiple species, changes in size over time, or marked sexual dimorphism, or some combination of these factors. Determining which of these alternatives is responsible for the pattern of variation in a given fossil assemblage is important because each has different implications regarding species diversity, modes of evolutionary change (i.e., anagenesis vs. cladogenesis), and social behavior. One perspective on fossil hominoid taxonomy specifies that the degree of variation in extinct species should not be greater than that in Gorilla and Pongo, the most sexually dimorphic extant hominoids, which logically requires rejection of a single-species hypothesis in cases where a temporally and geographically restricted fossil sample is more variable than these great apes (e.g., Kay, 1982a,b; Lieberman et al., 1988; Martin and Andrews et al., 1993; Teaford et al., 1993; Walker et al., 1993; see also Cope and Lacy, 1992; Cope, 1993; Plavcan, 1993). However, adopting a multiple-species taxonomy for a fossil sample solely on the basis of excessive size variation relative to Gorilla and Pongo is problematic for two reasons. First, it is not clear that the upper limit of intraspecific variation in extant primates is represented by these taxa. Among living primates, Gorilla and Pongo are exceeded in body-mass dimorphism (and presumably intraspecific variation in body mass) by the African papionin Mandrillus sphinx (Jungers and Smith, 1997; Setchell et al., 2001). Although it has not been established whether this difference is reflected in aspects of skeletal or dental size variation and dimorphism, data from other papionins, particularly Papio, indicate that at least some of the members of this clade may be more skeletally and dentally dimorphic than the great apes (e.g., Wood, 1976; Uchida, 1996a,b; Plavcan, 2002, 2003). Second, the upper limit of intraspecific variation may not be represented by any extant primate. Among fossil primates, the hominoid sample from the late Miocene *Correspondence to: Jeremiah E. Scott, School of Human Evolution and Social Change, Institute of Human Origins, Arizona State University, Tempe, AZ , USA. jeremiah.scott@asu.edu Received 3 July 2008; accepted 28 January 2009 DOI /ajpa Published online 8 April 2009 in Wiley InterScience ( VC 2009 WILEY-LISS, INC.

2 254 J.E. SCOTT ET AL. site of Lufeng, China, represents a single species Lufengpithecus lufengensis that exceeds Gorilla and Pongo in its degree of postcanine sexual dimorphism (Kelley and Xu, 1991; Kelley, 1993; Kelley and Plavcan, 1998). Establishing that the Lufeng sample represents a single highly dimorphic species was made possible by two key characteristics of the assemblage: the sample is large, comprising hundreds of teeth (e.g., Kelley and Etler, 1989; Wood and Xu, 1991), and a number of postcanine dentitions have been confidently sexed using associated canines and P 3 s (e.g., Kelley and Xu, 1991; Kelley, 1993; Kelley, 1995a,b). Using the sexed specimens (n 16 for each molar position), Kelley and colleagues (Kelley and Xu, 1991; Kelley, 1993; Kelley and Plavcan, 1998) demonstrated that molar dimorphism in L. lufengensis is so high that there is no overlap between male and female individuals in bivariate plots of mesiodistal and buccolingual dimensions. Several researchers have argued that the Lufeng sample contains multiple species (e.g., Wu and Oxnard, 1983a,b; Martin, 1991; Cope and Lacy, 1992; Plavcan, 1993), but a mixture of two or more species is unlikely to have produced the pattern of variation observed in the sample, unless one appeals to highly improbable sampling events (Kelley and Plavcan, 1998). Thus, L. lufengensis extends the known range of intraspecific size variation and sexual dimorphism in the Hominoidea, at least with respect to the postcanine dentition. Despite initial objections based on both ontological and epistemological grounds (e.g., Ruff et al., 1989; Martin, 1991; Cope and Lacy, 1992; Martin and Andrews, 1993; Plavcan, 1993; Teaford et al., 1993; Walker et al., 1993), the idea that some fossil hominoid species were more dimorphic than living great apes has gained wider acceptance, and many researchers now acknowledge extreme dimorphism as a potential source of high measures of variation that must be considered when evaluating fossil samples (e.g., Plavcan, 2001; Plavcan and Cope, 2001; Scott and Lockwood, 2004; Schrein, 2006; Skinner et al., 2006; Simons et al., 2007; Humphrey and Andrews, 2008). 1 This is not to say that extreme dimorphism should be regarded as the null hypothesis for Miocene hominoids; rather, we are suggesting that extreme dimorphism is a viable alternative to the hypothesis that high levels of size variation in a fossil sample indicate the presence of multiple species. Acceptance of L. lufengensis as a single highly dimorphic species highlights the need to incorporate other comparative species in addition to the living great apes when evaluating fossil samples. One option is to use the highly dimorphic papionins as analogues, which some studies have done (e.g., Ruff et al., 1989; Teaford et al., 1993; Uchida 1996b; Harvati et al., 2004; Baab, 2008). Another option is to use L. lufengensis as an analogue (Kelley, 2005). The use of fossil species to model intraspecific variation in other fossil assemblages was suggested by Wood (1991), who used Australopithecus boisei to determine whether variation in A. africanus and A. robustus indicated the presence of multiple species in each of these hypodigms (for other examples of the use of extinct species to evaluate variation in fossil samples, see Kelley, 2005; Skinner et al., 2006; Baab, 2008). Although Wood s (1991) intent in taking this approach was to control for temporal variation, the purpose of using L. lufengensis as an analogue would be to include a reference sample that possesses a level of intraspecific variation not represented among extant hominoids. Although the amount of time represented by the hominoid-bearing deposits at Lufeng is unknown, temporal variation is unlikely to be a major component of the high level of size variation in L. lufengensis, given that intrasexual variation in the sample is within the range of modern species (Kelley and Plavcan, 1998). The fact that L. lufengensis exceeds Pongo in its level of molar dimorphism means that it is potentially more dimorphic in the molar dentition than any extant primate, as Pongo is commonly thought to possess the greatest level of molar dimorphism among living primates (e.g., Mahler, 1973; Kelley and Xu, 1991; Kelley and Plavcan, 1998). If true, then including the Lufeng sample as part of the comparative framework for assessing variation in fossil primate samples becomes even more critical. In fact, it has not been quantitatively verified that Pongo expresses the greatest degree of molar dimorphism among living primates, and therefore the claim that the degree of molar dimorphism documented in L. lufengensis falls outside the range observed in living primate species has not been adequately tested. Thus, in this study, we test two hypotheses regarding molar size dimorphism in primates: (1) Pongo represents the uppermost extreme of molar dimorphism among living primates, and (2) molar dimorphism in L. lufengensis is greater than that in the most dimorphic living primate species. We then apply the results of these analyses to other potential instances of extreme dimorphism in the late Miocene hominoid fossil record the Sivapithecus material from Haritalyangar, India, and the Ouranopithecus macedoniensis material from Greece (Kelley, 2005; Schrein, 2006). Specifically, we evaluate whether levels of apparent sexual dimorphism (i.e., the level of dimorphism required if the distinct large and small size clusters evident in the Sivapithecus and O. macedoniensis molar samples represent conspecific males and females, respectively) in these fossil samples fall within the limits of dimorphism established for living primates and L. lufengensis. MATERIALS AND METHODS Three extant species were included in the analysis: the western lowland gorilla (Gorilla gorilla), the Bornean orangutan (Pongo pygmaeus), and the drill (Mandrillus leucophaeus) (Table 1). The drill was chosen to represent papionins because Plavcan s (1990) data set indicates that Mandrillus probably has the most dimorphic postcanine teeth of any extant papionin. Mandrillus leucophaeus is smaller in body size than M. sphinx and may not be as sexually dimorphic in body mass (Jungers and Smith, 1997), but the two species have similar degrees of postcanine dimorphism. This assessment is based on a comparison of Plavcan s (1990) M. leucophaeus data set to unpublished data for M. sphinx collected by S. Frost, R. Nuger, and M. Singleton. The M. leucophaeus data were used in order to avoid the potential for interobserver error in the M. sphinx data. For each of the extant species, maximum length (mesiodistal, MD) and width (buccolingual, BL) dimensions of the mandibular molars were taken from the literature (for G. gorilla: Mahler, 1973; for P. pygmaeus and M. leucophaeus: Plavcan, 1990). Maxillary molars were not included in the analysis because sample sizes for these teeth are not as large as those for the mandibular molars in the fossil samples.

3 MODELING SEXUAL DIMORPHISM IN MIOCENE APES 255 TABLE 1. Sample sizes for the extant comparative taxa and L. lufengensis M 1 M 2 M 3 Male Female Male Female Male Female G. gorilla P. pygmaeus M. leucophaeus L. lufengensis Data are from the following sources: G. gorilla, Mahler (1973); P. pygmaeus, Plavcan (1990); M. leucophaeus, Plavcan (1990); L. lufengensis, provided by Xu Qinghua. The L. lufengensis sample is identical to the one used by Kelley and Plavcan (1998; Table 1), with the exception of one additional female M 3 (identified by JK after reexamining the Lufeng data). This sample includes only molars from associated dentitions, thus making tooth position (i.e., M 1,M 2,M 3 ) unambiguous and making it possible to sex teeth using associated canines and P 3 s (see Kelley, 1993). These two factors are important for obtaining an accurate estimation of sexual dimorphism in L. lufengensis (Kelley and Plavcan, 1998). Inclusion of isolated and unsexed molars has the potential to bias estimates of dimorphism upwards, either by mixing M 1 s and M 2 s or by including large females in the male sample and small males in the female sample (e.g., Kelley and Etler, 1989; Kelley and Plavcan, 1998). Sexual dimorphism in Lufengpithecus lufengensis and the extant species, quantified using log-transformed (base e) indices of sexual dimorphism (following Smith, 1999), was compared in two ways: (1) by combining the individual molars into a single measure (i.e., multivariate molar size dimorphism) and (2) by analyzing each molar position separately. This allowed us to evaluate overall dimorphism in the molar row and to account for the fact that comparisons among individual teeth are not independent (i.e., species with highly dimorphic M 1 s are also likely to have highly dimorphic M 2 s), while also examining the differences at each molar position. For both analyses, molar size was represented by the geometric mean of the MD and BL dimensions [i.e., (MD 3 BL) 1/2 ]. For the analysis of multivariate molar size dimorphism, we used the geometric-mean-based method developed by Gordon et al. (2008), which is useful for examining the overall dimorphism in a series of variables (molar dimensions in this case) and has the benefit that specimens with missing data (e.g., incomplete fossil specimens) can be included. This approach takes advantage of the fact that the ratio of two geometric means is mathematically equivalent to the geometric mean of the individual ratios for each of the variables that constitute the geometric means (Gordon et al., 2008): GM h 1 x1 ¼ 3 y 1 3 z i 1 1=3 ; GM 2 x 2 y 2 z 2 where GM 1 is the geometric mean of the means (i.e., the cube root of the product of x 1, y 1, and z 1 ) for a set of variables measured on individuals in Group 1 (e.g., males of a particular species) and GM 2 is the geometric mean of the means for the same set of variables (i.e., x 2, y 2, and z 2 ) measured on individuals in Group 2 (e.g., females of the same species). ð1þ Thus, multivariate molar size dimorphism can be calculated in two ways. The first way is to calculate the ratio of GMs. For each sex, a measure of multivariate molar size can be computed as follows: GM #ALL ¼ðGM #1 3 GM # GM #n Þ 1=n ; where GM #1 is the geometric mean of M 1,M 2, and M 3 size [i.e., (M 1 3 M 2 3 M 3 ) 1/3 ] for the first male, GM #2 is the geometric mean of M 1,M 2, and M 3 size for the second male, etc., and thus GM #ALL is the geometric mean of the geometric means of all male individuals. The female geometric mean (GM $ALL ) is computed in the same way. The index of sexual dimorphism (ISD) for multivariate molar size can then be calculated as ISD 5 GM #ALL /GM $ALL (i.e., the ratio of GMs). The second way to calculate multivariate molar size dimorphism is to use the GM of ratios: ISD ¼ðM1 ISD 3 M2 ISD 3 M3 ISD Þ 1=3 ; where M1 ISD is the ISD for M 1 (i.e., mean M1 size for males divided by mean M1 size for females), M2 ISD is the ISD for M 2, and M3 ISD is the ISD for M 3. Note that Equations 3 and 1 are equivalent. When using the ratio of GMs (i.e., GM #ALL /GM $ALL )to calculate multivariate molar size dimorphism, all of the specimens in the analysis must possess each molar; those lacking one or more molars must be excluded. However, using the GM of ratios [i.e., (M1 ISD 3 M2 ISD 3 M3 ISD ) 1/3 ], specimens with missing data can be retained because the ISD for each molar position is calculated independently of the other positions (Gordon et al., 2008), and thus fossil specimens that do not preserve the entire molar row can be included in the analysis. For this study, the GM of ratios was used to calculate ISDs for each sample in order to account for the fact that withinsex sample sizes for each molar position were not equal for any of the species or fossil samples used in the analysis (Table 1). To statistically evaluate differences between sample ISDs, we used the bootstrap (i.e., resampling with replacement) to generate 95% confidence intervals for each pairwise difference as follows: 1. For each molar position, Sample A (e.g., G. gorilla) was resampled with replacement 2000 times, 2 with the sample size and sex ratio for each bootstrap sample being identical to those of the original sample. Note that, because we resampled with replacement, a specimen could be included in each bootstrap sample multiple times or not included at all, and thus each iteration was highly unlikely to produce a sample that was identical to the original sample in specimen composition. 2. The ISD for each bootstrap sample was then computed. For the analysis of multivariate molar size dimorphism, bootstrap samples for each molar position were randomly grouped together (i.e., one bootstrap sample of M 1 s, one bootstrap sample of M 2 s, and one bootstrap sample of M 3 s), and multivariate molar dimorphism was calculated as the GM mean of the ISDs for each molar position. Note that we did not resample entire molar rows at once. Thus, in the case of the G. gorilla sample, for each iteration, an M 1 ISD calculated using 43 males and 43 females was ð2þ ð3þ

4 256 J.E. SCOTT ET AL. between Sample A and Sample B (disregarding the sign of the difference). 3. Finally, we divided the value obtained from Step 2 by the total number of bootstrap samples. The observed difference between Samples A and B was included in the latter calculations, such that P 5 (M 1 1)/(N 1 1), where M is the number of bootstrap differences (absolute values) greater than or equal to the observed difference, N is the total number of bootstrap differences, and one is added to M and N to include the observed difference. Fig. 1. An example of the procedure used to derive P-values for pairwise differences among the extant species and L. lufengensis. The top image shows the distribution of pairwise differences obtained from bootstrapping two samples. The observed difference between the ISDs of the two samples is 0.06; accordingly, the distribution is centered on In the bottom image, the bootstrap distribution has been recentered on zero. The observed difference (0.06), represented by the vertical line, does not fall within the zero-centered distribution. Thus, the P-value for this comparison is P (i.e., 1/2001; see text for further details). combined with an M 2 ISD calculated using 40 males and 40 females, and these were combined with an M 3 ISD calculated using 34 males and 34 females (see Table 1). 3. Steps 1 and 2 were performed for Sample B (e.g., L. lufengensis), with each bootstrap sample containing the same number of males and females as in Sample B. 4. The bootstrapped ISDs for Sample A were then randomly paired with those for Sample B, and the difference between the ISDs for each pairing was calculated, creating a distribution of 2000 ISD differences. The middle 95% of this distribution represents the 95% confidence interval for the pairwise comparison. A pairwise difference with a 95% confidence interval that does not overlap zero (i.e., no difference) can be considered statistically significant at the a level. However, we obtained more precise P-values in the following way: 1. First, we recentered the distribution of pairwise differences between Sample A and Sample B on zero (see Fig. 1), as outlined by Manly (1997, p ). This step was necessary because the distribution of pairwise differences will be centered on the observed difference between Sample A and Sample B. In order to derive a P-value for the observed difference between Samples A and B, the distribution must be recentered on (i.e., the mean of the distribution must equal) zero. According to Manly (1997, p 99), the idea with this approach is to use bootstrapping to approximate the distribution of a suitable test statistic when the null hypothesis is true (i.e., no difference between samples). 2. Next, using the recentered distribution, we counted the number of values that were as extreme as or more extreme than the observed ISD difference Note that this test is two tailed. Although the questions of interest are (1) whether M. leucophaeus exceeds P. pygmaeus and G. gorilla in molar size dimorphism and (2) whether L. lufengensis exceeds all of the extant comparative species in molar size dimorphism, specifying a directional alternative to the statistical null hypothesis of no difference in sexual dimorphism requires a priori justification (i.e., evidence independent of the sample estimates of molar dimorphism; see also Scott and Stroik, 2006). For example, if it were known that postcranial size dimorphism in L. lufengensis is greater than in any extant primate, then one could reasonably hypothesize that other aspects of the Lufeng hominoid are also extremely dimorphic, thus justifying a one-tailed test. This resampling procedure differs from previous applications of the bootstrap (e.g., Lockwood et al., 1996, 2000; Lockwood, 1999; Silverman et al., 2001; Reno et al., 2003; Villmoare, 2005; Harmon, 2006; Schrein, 2006; Gordon et al., 2008; see also Cope and Lacy, 1992) in two important ways. First, the latter studies are generally concerned with determining the probability of obtaining from the comparative samples a sample with characteristics (e.g., size, variation, sexual dimorphism) identical to that of a fossil sample. In the present case, however, because we are dealing with a fossil assemblage (the sexed Lufeng specimens) that is large in comparison to other such assemblages, we are able to use the bootstrap to generate confidence intervals for the fossil ISD, allowing us to make inferences about population parameters. Thus, we are able to incorporate the uncertainty in the sample ISDs for the comparative taxa and for the fossil species, resulting in more robust statistical testing than is typically the case for small fossil samples (see also Gordon et al., 2008). The second notable difference between our resampling procedure and those used in previous studies of variation in fossil samples is that the bootstrap samples obtained from the comparative taxa were not reduced to match the size of the fossil sample. In most of the studies cited above, the statistic for the fossil sample (e.g., coefficient of variation, the index of sexual dimorphism) is used as a point estimate for comparison with distributions generated from the comparative samples. Such distributions are composed of bootstrap samples that are identical in size to the fossil sample. When only a point estimate is used for the fossil sample, it is necessary for the samples produced from resampling the comparative samples to be the same size as the fossil sample because confidence intervals for the fossil sample are not generated. In contrast, because we resampled the comparative samples and the L. lufengensis sample and thus generated confidence intervals for all of the samples matching the sample size of the fossil sample was unnecessary and

5 MODELING SEXUAL DIMORPHISM IN MIOCENE APES 257 Taxon Specimen a TABLE 2. The Sivapithecus and Ouranopithecus samples assignment b C P 3 M 1 M 2 M 3 Sex Tooth size c Ouranopithecus RPl-55 Male** RPl-56 Male RPl-75 Male** RPl-76 Male 13.2 d 18.1 RPl-89* Male NKT-21 Female RPl-54 Female** RPl-79* Female RPl-84* Female RPl-88* Female Sivapithecus GSI D. 197 Male YPM Male ONGC/v/790 Male 13.0 PUA Male 13.9 GSI Male 14.9 GSI Male 11.3 GSI D. 199 Female YPM Female YPM Female GSI Female 10.7 GSI Female 10.3 PUA Female 9.6 a Data for Ouranopithecus were taken from Koufos (1993) and Koufos and de Bonis (2004); specimens marked with an asterisk (*) were not included in Schrein s (2006) analysis. The Sivapithecus data were compiled by JK from various sources. b For Ouranopithecus, sex assignment is based on canine and P 3 size (see also Fig. 2); specimens marked with double asterisks (**) were sexed by Koufos (1995) using canine shape. For Sivapithecus, sex assignment is based on molar size (see text for further discussion). c Tooth size was calculated as the square root of the product of MD and BL diameters. d Only the MD dimension [maximum length, identified by Koufos (1993) as transverse diameter ] is available for this canine. would have actually reduced the power of the test to detect differences, making it overly conservative. After establishing the rank order of molar dimorphism in the extant species and L. lufengensis, we evaluated apparent dimorphism in the Ouranopithecus macedoniensis and Haritalyangar Sivapithecus samples. The O. macedoniensis data used for this study were taken from the literature (Koufos, 1993; Koufos and de Bonis, 2004), while the Sivapithecus data were provided by JK. All of the O. macedoniensis specimens used here can be confidently sexed based on associations with canines and P 3 s (Table 2; Fig. 2; see also Schrein, 2006). This sample includes newly published specimens from Ravin de la Pluie (Koufos and de Bonis, 2004) that were not used in the most recent analysis of variation and sexual dimorphism in this fossil ape (Schrein, 2006), and that expand the sample from n sexed individuals per molar position to n , including the addition of three complete female molar rows (for a total of five) and one complete male molar row (for a total of four) (Table 2). The sample of Sivapithecus mandibular molars from Haritalyangar is smaller, with n individuals per molar position (Table 2), and none of these specimens can be sexed based on associations with canines or P 3 s. However, the M 2 and M 3 samples are each characterized by the presence of two markedly disjunct size clusters. If the Haritalyangar assemblage samples a single species, Fig. 2. Bivariate plots of canine size vs. M 2 size (top) and P 3 size vs. M 2 size (bottom) in Ouranopithecus macedoniensis. Tooth size is the geometric mean of the MD and BL dimensions. Specimens that were sexed based on canine shape by Koufos (1995) are indicated by male (#) and female ($) symbols. Note that (1) large and small canines and P 3 s cluster with the male and female specimens, respectively, and (2) specimens with large molars are associated with large (male) canines and P 3 s, whereas specimens with small molars are associated with small (female) canines and P 3 s. The first and third molars exhibit a similar pattern. then the cluster of large specimens must be composed of males and the cluster of small specimens must be composed of females (Kelley, 2005). In contrast, the distribution of Haritalyangar M 1 s is continuous, making sex assignment more arbitrary. Based on associations with M 2 s, three of the seven M 1 s can be tentatively allocated to male and female clusters, while the largest (ONGC/v/790) and smallest (PUA ) M 1 s can also be assigned to the male and female groupings, respectively. Two teeth GSI and GSI fall in the middle of the M 1 size range. If no overlap between males and females is assumed, then the index of apparent sexual dimorphism (ISD A ) is between 1.19 and 1.21, depending on whether both specimens are assigned to one sex or if the larger GSI is grouped with presumed males and the smaller GSI is grouped with presumed females. If females and males do overlap in size (with the smaller GSI placed with presumed males and the larger GSI placed with presumed females), then the ISD A would be We examined the effect of these alternative assignments and found that they did not substantively influence the results. Thus, we report only the results of the analyses in which GSI was considered male and GSI was considered female. Note that these sex assignments

6 258 J.E. SCOTT ET AL. are identical to those that would have been obtained had we simply used the mean method [i.e., dividing the sample into males and females about the mean (Plavcan, 1994; Gordon et al., 2008)]. The Sivapithecus and O. macedoniensis samples were evaluated using resampling methods, but because these samples are small (n 10 for all molar positions), they were treated as point estimates for the purpose of comparing them to the extant taxa and L. lufengensis. Following previous studies (e.g., Lockwood et al., 1996, 2000; Lockwood, 1999; Silverman et al., 2001; Reno et al., 2003; Villmoare, 2005; Harmon, 2006; Schrein, 2006), we bootstrapped the comparative species (including L. lufengensis) to obtain samples that were identical to the Sivapithecus and O. macedoniensis samples in size and sex ratio, creating distributions for determining the probability of obtaining a sample from G. gorilla, P. pygmaeus, M. leucophaeus, and L. lufengensis with the same level of molar dimorphism observed in the Sivapithecus and O. macedoniensis samples. In the procedure describe above, resampling without replacement can be used instead of bootstrapping (e.g., Gordon et al., 2008). In fact, resampling without replacement is more likely to produce lower P-values than resampling with replacement given that, at small sample sizes (e.g., n in the case of the Sivapithecus analysis), the latter can, in principle, produce samples composed only of multiple entries of the largest male and smallest female, or samples composed only of multiple entries of the smallest male and largest female. Clearly, such samples would produce wider bootstrap distributions (i.e., with very high and very low ISDs) that will be more likely to encompass the fossil value. However, Cope and Lacy (1992, p. 361), in their study of the use of the coefficient of variation (CV) for evaluating variation in fossil samples, noted that a comparative sample of hundreds or thousands is needed to properly simulate CV sample distributions. This problem motivated them to develop a method in which a very large (n 5 10,000) simulated population is generated using descriptive statistics from samples of extant species. This simulated population is then resampled without replacement at a sample size equal to the fossil assemblage in order to determine the probability of sampling the CV observed in the fossil sample from the simulated population. As an alternative for overcoming the intractable problem of limited comparative material, Lockwood et al. (1996) used the bootstrap to generate samples equal in size to fossil samples directly from the comparative samples. In this study, we preferred the bootstrap over resampling without replacement because it is not clear that our comparative samples are sufficiently large. For each molar position, two sets of 1000 bootstrap samples were drawn from each of the extant species samples and from the Lufengpithecus sample, with one set matching the composition of the Sivapithecus sample and the other matching the composition of the O. macedoniensis sample. 3 For example, in the Sivapithecus analysis, each bootstrap sample contained seven M 1 s (four males, three females), six M 2 s (three males, three females), and five M 3 s (two males, three females). For each bootstrap sample, log-transformed (base e) ISDs were calculated as described above. The Sivapithecus and O. macedoniensis ISD A s (log-transformed) were then compared to the bootstrap distributions; if the values for these two fossil samples fell outside the middle 95% of the bootstrap distributions, then the null hypothesis of no difference was considered falsified at the a level. For these tests, two-tailed P-values were obtained by counting the total number of bootstrap ISD values that were as extreme as or more extreme than the Sivapithecus and O. macedoniensis values (including the values for the two fossil samples) and dividing that number by the total number of ISDs (i.e., 1001). In this case, extreme refers to values that when subtracted from the comparative sample s ISD produce a difference (regardless of sign) as large as or larger than the difference produced by subtracting the fossil sample s ISD A from the comparative sample s ISD. Because our division of the Sivapithecus sample into sexes was based solely on size, we also analyzed this sample using a sex-blind statistic the CV to estimate dimorphism. For each molar position, we bootstrapped the comparative samples 1000 times each at a sample size equal to the Sivapithecus sample but without regard to sex (i.e., the sex ratios of the bootstrapped samples did not necessarily match the hypothesized sex ratio of the Sivapithecus sample) and calculated the CV for each. For this part of the analysis, the comparative samples (including the Lufeng sample) were modified so that the sex ratios for each tooth were balanced prior to being bootstrapped. We then compared the CVs for the Sivapithecus molars to the resulting distributions and determined the statistical significance of the sample differences as described above. For this analysis, we only examined the individual molar positions, as there are currently no methods for dealing with missing data in the calculation of the CV. The results of these analyses did not differ substantively from the analyses in which the specimens were sexed, and thus only the ISD-based results are reported. Finally, it is important to note that, because our comparative samples are not composed entirely of complete molar rows, the P-values reported for the analyses of multivariate molar dimorphism should be considered approximate. For example, consider a case in which the molars of a species are identical in their degree of dimorphism. If a sample of 40 males and 40 females is collected in which the ten largest males are missing their M 3 s, then the estimate of dimorphism for the M 3 will be lower than in the other teeth. When the teeth are combined and multivariate molar dimorphism is estimated, the estimate will be biased due to the missing M 3 data. Such a sample will produce a bootstrap distribution that is shifted to the left (i.e., toward monomorphism), resulting in an artificially low P-value and a potential type II error in a pairwise comparison with a fossil sample. However, this problem is unlikely to be an issue in our analysis because the missing teeth in our samples are not size-biased. Thus, the effects of missing data on the P-values for the analysis of multivariate molar dimorphism are likely to be minimal. Therefore, in order to use samples that are as large as possible, we have chosen not to limit the comparative samples to only those individuals that preserve complete molar rows. RESULTS The ISDs for the extant taxa and L. lufengensis are presented in Table 3. For multivariate molar size, sample dimorphism is greatest in L. lufengensis, followed in rank order by M. leucophaeus, P. pygmaeus, and G. gorilla. This pattern is repeated at each individual molar position with one exception: M 3 sample dimorphism is greater in G. gorilla than in P. pygmaeus. The bootstrap tests for multivariate molar size dimorphism

7 MODELING SEXUAL DIMORPHISM IN MIOCENE APES 259 TABLE 3. Indices of sexual dimorphism for the extant taxa and L. lufengensis M ALL a M 1 M 2 M 3 G. gorilla P. pygmaeus M. leucophaeus L. lufengensis a The abbreviation M ALL refers to multivariate molar size here and in subsequent tables. TABLE 4. Results of the bootstrap tests for interspecific differences in multivariate molar size dimorphism Gorilla Pongo Mandrillus Pongo 5 (0.1119) Mandrillus M [ G (0.0005) M [ P (0.004) Lufengpithecus L [ G (0.0005) L [ P (0.0005) L [ M (0.001) Nonsignificant differences are indicated by an equality symbol; greater-than symbols indicate significance and the direction of difference. P-values for each comparison are given in parentheses (probabilities are two-tailed). Abbreviations: G, Gorilla; P, Pongo; M, Mandrillus; L, Lufengpithecus. reveal that G. gorilla and P. pygmaeus are not significantly different, whereas M. leucophaeus is significantly more dimorphic than the two extant apes, and L. lufengensis is significantly more dimorphic than all of the living species (Table 4, Fig. 3). When dimorphism is examined by molar position (Table 5), P. pygmaeus and G. gorilla differ statistically only at M 2, with P. pygmaeus being more dimorphic at this position. Mandrillus leucophaeus is significantly more dimorphic than G. gorilla at M 1 and M 2 but not at M 3 and is significantly different from P. pygmaeus only at M 3. Lufengpithecus lufengensis is significantly more dimorphic than the living apes at all molar positions, but is significantly more dimorphic than M. leucophaeus only at M 1 and M 2. Some of these differences are nonsignificant after adjusting a-levels for multiple comparisons using the sequential Bonferroni method (e.g., Rice, 1989); the results that remain significant are: M. leucophaeus more dimorphic than G. gorilla for M 1 and M 2, L. lufengensis more dimorphic than the living great apes for all molar positions, and L. lufengensis more dimorphic than M. leucophaeus for M 1. Application of the sequential Bonferroni adjustment to the comparisons of multivariate molar size dimorphism does not alter the results. The results for the analysis of the O. macedoniensis sample are given in Table 6. For multivariate molar size, O. macedoniensis is more dimorphic than all of the extant taxa, but it is not significantly different from L. lufengensis (Fig. 4A). The results for M 1 and M 3 are similar to those for multivariate molar size, but for M 2, O. macedoniensis is only significantly more dimorphic than G. gorilla. Sequential Bonferroni adjustment renders only the latter difference nonsignificant. In contrast to the O. macedoniensis sample, apparent dimorphism in the Sivapithecus assemblage is significantly greater than in any of the comparative taxa, including L. lufengensis, even after sequential Bonferroni adjustment (Table 7, Fig. 4B). The only exception is apparent dimorphism in the M 1 sample, which cannot be statistically distinguished from M 1 dimorphism in L. lufengensis. Thus, the Haritalyangar M 2 and M 3 Fig. 3. Bootstrap distributions for multivariate molar size dimorphism for the extant species and L. lufengensis. The middle 95% of each distribution is equivalent to the 95% confidence interval for multivariate molar size dimorphism. Gorilla gorilla and P. pygmaeus are not significantly different, M. leucophaeus is more dimorphic than the living apes, and L. lufengensis is more dimorphic than all of the extant taxa. dimensions indicate that if this assemblage samples a single species, then the distal molars of that species are even more dimorphic than those of L. lufengensis. This is true for multivariate molar dimorphism as well. DISCUSSION Identifying levels of sexual dimorphism in fossil species that are extreme in comparison to living species requires knowledge of the limits of dimorphism in extant taxa. In the case of L. lufengensis, previous studies have used extant Pongo to represent the upper limit of molar dimorphism in living primates (Kelley and Xu, 1991; Kelley, 1993: Kelley and Plavcan, 1998). However, the results of this study demonstrate that the molars of P. pygmaeus are not the most size-dimorphic among extant primates; in fact, our samples do not allow us to unequivocally establish that P. pygmaeus is even the most dimorphic living hominoid in this respect. Mandrillus leucophaeus emerges as the most dimorphic extant primate when the molar row is considered in its entirety, but the drill cannot be consistently distinguished statistically from either P. pygmaeus or G. gorilla at individual molar positions (though sample dimorphism is always greatest in the drill among the extant taxa). While the general lack of statistical differences between G. gorilla and P. pygmaeus in this study challenges the conventional assumption that the molars of the orangu-

8 260 J.E. SCOTT ET AL. TABLE 5. Results of the bootstrap tests for interspecific differences at individual molar positions M1 M2 M3 Gorilla Pongo Mandrillus Gorilla Pongo Mandrillus Gorilla Pongo Mandrillus Pongo 5 (0.1009) P [ G* (0.0325) 5 (0.4448) Mandrillus M [ G (0.0015) 5 (0.1339) M [ G (0.0005) 5 (0.2214) 5 (0.1149) M [ P* (0.0305) Lufengpithecus L [ G (0.0005) L [ P (0.0005) L [ M (0.012) L [ G (0.0005) L [ P (0.0005) L [ M* (0.0195) L [ G (0.0005) L [ P (0.001) 5 (0.0885) Nonsignificant differences are indicated by an equality symbol; greater-than symbols indicate significance and the direction of difference. P-values for each comparison are given in parentheses (probabilities are two-tailed). Abbreviations: G, Gorilla; P, Pongo; M, Mandrillus; L, Lufengpithecus. An asterisk (*) indicates that the comparison is not significant after sequential Bonferroni adjustment (applied within each variable). tan are more dimorphic than those of the gorilla, Uchida (1996a) documented intrageneric variation in molar dimorphism in both Pongo and Gorilla. Thus, a more complete analysis that includes material from eastern lowland gorillas, mountain gorillas, and Sumatran orangutans could reveal differences between these two genera. It is also important to point out that our ability to detect differences among the living taxa is hampered in some respects. First, ratios such as the ISD generally have wider confidence intervals than the variables from which they are derived due to the fact that there is measurement error in both the numerator (male mean) and denominator (female mean) (see discussion in Smith, 1999). Second, because males and females constitute separate components of the ISD, the effective sample sizes for each species are about half the total sample sizes. Thus, given that the differences in molar sample ISDs among G. gorilla, P. pygmaeus, and M. leucophaeus are relatively small, especially when compared to differences in canine, craniofacial, and body-mass dimorphism across the Anthropoidea (Plavcan, 2001), it is not surprising that many of the comparisons in this study are nonsignificant. Future studies will require larger samples to establish whether the differences in sample ISDs observed between G. gorilla and P. pygmaeus and between P. pygmaeus and M. leucophaeus at individual molar positions reflect true population differences. In spite of the conservative nature of the statistical tests, our results confirm that the molars of L. lufengensis are more dimorphic than those of living apes. Our analysis of the Lufeng hominoid also demonstrates that it was more dimorphic than M. leucophaeus (at least with respect to the molar row in its entirety and M 1, and probably M 2 as well). That we were able to detect significant differences between L. lufengensis and the extant species despite the fact that the Lufeng sample contains only n individuals per molar position highlights just how much greater dimorphism is in the molar teeth of this fossil ape. Notably, the analysis of multivariate molar size dimorphism indicates that M. leucophaeus is intermediate between the great apes and L. lufengensis (see Fig. 3). Thus, although molar dimorphism in L. lufengensis is extreme relative to living primates, comparison to the drill reveals that it is not as extreme as it appears to be when only extant great apes are considered. Despite the drill s higher level of overall molar dimorphism compared to extant apes, its inclusion in our analysis of the O. macedoniensis material does not alter previous conclusions that a single-species interpretation of this fossil assemblage necessitates a level of dimorphism that exceeds that observed in living primates (Schrein, 2006). Although the addition of the more recent Ravin de la Pluie specimens (Koufos and de Bonis, 2004) results in slightly lower indices of apparent sexual dimorphism (ISD A s) for O. macedoniensis in comparison to those for the smaller sample used by Schrein (2006), there is still no overlap in size between specimens identified as male and those identified as female on the basis of canine or P 3 size/morphology (see Table 2, Fig. 2), an attribute also evident in the L. lufengensis sample (Kelley and Plavcan, 1998). In our analysis, M 2 is the only O. macedoniensis variable for which size dimorphism cannot be statistically distinguished from that of any of the extant species, except for G. gorilla prior to sequential Bonferroni adjustment. Schrein (2006) obtained somewhat different results for her comparisons

9 MODELING SEXUAL DIMORPHISM IN MIOCENE APES 261 TABLE 6. Bootstrap results for the Ouranopithecus comparisons Ouranopithecus M ALL (ISD ) M 1 (ISD ) M 2 (ISD ) M 3 (ISD ) Gorilla O [ G (0.001) O [ G (0.002) O [ G* (0.018) O [ G (0.004) Pongo O [ P (0.001) O [ P (0.002) 5 (0.1339) O [ P (0.001) Mandrillus O [ M (0.001) O [ M (0.006) 5 (0.4336) O [ M (0.004) Lufengpithecus 5 (0.3407) 5 (0.3996) 5 (0.2987) 5 (0.0939) Nonsignificant differences are indicated by an equality symbol; greater-than symbols indicate significance and the direction of difference. P-values for each comparison are given in parentheses (probabilities are two-tailed). Abbreviations: G, Gorilla; P, Pongo; M, Mandrillus; O, Ouranopithecus. An asterisk (*) indicates that the comparison is not significant after sequential Bonferroni adjustment (applied within each variable). between O. macedoniensis and the great apes (i.e., only the M 1 of Ouranopithecus could be confidently identified as being more dimorphic than in gorillas and orangutans), which is probably attributable to the smaller sample of O. macedoniensis used in her study and the fact that the compositions of the Gorilla and Pongo samples were different from those used in the present analysis. Note also that Schrein (2006) examined MD and BL dimensions separately, whereas we combined them [i.e., (MD 3 BL) 1/2 ]. Nevertheless, the pattern of results in the two studies is broadly similar. Under the assumption that fossil species could not have been more dimorphic than extant species, our results could be interpreted as supporting the presence of multiple species in the O. macedoniensis sample (e.g., Kay, 1982a). However, the fact that molar size dimorphism in O. macedoniensis cannot be statistically distinguished from that in L. lufengensis demonstrates that the level of dimorphism required for O. macedoniensis under a single-species taxonomy is not unprecedented among late Miocene hominoids. Thus, the inclusion of L. lufengensis in the comparative framework demonstrates that, although O. macedoniensis does not fit expectations regarding sexual dimorphism among extant taxa, it can be accommodated within known models of intraspecific variation and sexual dimorphism when other fossil species are used as analogues. Schrein (2006) reviewed several lines of evidence pointing to the existence of only a single species within the O. macedoniensis dental sample: large specimens are male, small specimens are female; the Ravin de la Pluie specimens, which constitute the bulk of the sample, are from individuals that were sympatric and probably synchronic; and molar morphology is homogeneous. The results presented here strengthen the case for recognizing a single species among the hominoid remains from Ravin de la Pluie, Xirochori, and Nikiti, one characterized by an extreme degree of molar dimorphism relative to living primates. The results of the Sivapithecus analysis, on the other hand, do not lend themselves to easy interpretation. Kelley (2005) found that measures of variation for the Haritalyangar M 2 s and M 3 s are statistically significantly higher than those for L. lufengensis, suggesting an even greater level of sexual dimorphism than that exhibited by L. lufengensis if these specimens represent a single species. Our results confirm that the level of apparent dimorphism in the Haritalyangar M 2 and M 3 samples is highly unlikely to have come from a species as dimorphic as L. lufengensis; none of the samples bootstrapped from the Lufeng sample produced ISDs as high as the ISD A s for the Haritalyangar M 2, M 3, or multivariate molar size. In contrast, the ISD A for M 1 falls well within the Fig. 4. Bootstrap distributions for multivariate molar size dimorphism generated by bootstrapping the extant taxa at a sample size and sex ratio identical to that of the (A) Ouranopithecus and (B) Sivapithecus samples. Only the M. leucophaeus and L. lufengensis distributions are shown; the G. gorilla and P. pygmaeus distributions would be to the left of the Mandrillus distribution. The solid vertical line in A indicates the ISD A for the O. macedoniensis sample; the dashed vertical line in B indicates the ISD A for the Sivapithecus sample. Lufeng distribution, and Kelley s (2005) analysis of variation in the Haritalyangar maxillary molars indicates that the same would certainly be true for M 1,M 2, and M 3, as levels of apparent dimorphism in these teeth are slightly lower than or similar to those of L. lufengensis (M 2 ISD A ; M 3 ISD A ; an ISD A cannot be calculated for M 1 due to the fact that the specimens