AMERICAN JOURNAL OF PHYSICAL ANTHROPOLOGY 140:253–264 (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 Kelley2
1
School of Human Evolution and Social Change, Institute of Human Origins,
Arizona State University, Tempe, AZ 85287-4101
2
Department of Oral Biology, College of Dentistry, University of Illinois at Chicago, Chicago, IL 60612
KEY WORDS bootstrap; dental variation; Lufengpithecus; Ouranopithecus; Sivapithecus
ABSTRACT Sexual size dimorphism in the postca-
nine dentition of the late Miocene hominoid Lufengpithe-
cus lufengensis exceeds that in Pongo pygmaeus, demon-
strating that the maximum degree of molar size dimor-
phism 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 resam-
pling-based methods to compare molar dimorphism in
Gorilla, Pongo, and Lufengpithecus to that in the papio-
nin 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 spe-
A frequently encountered problem in hominoid paleon-
tology is identification of the source of high levels of size
variation in a fossil sample (e.g., Kay 1982a,b; Lieber-
man 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 combina-
tion of these factors. Determining which of these alterna-
tives is responsible for the pattern of variation in a given
fossil assemblage is important because each has different
implications regarding species diversity, modes of evolu-
tionary 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 sex-
ually 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).
C 2009
V WILEY-LISS, INC.
cies. Using these samples, we also evaluated molar dimor-
phism and taxonomic composition in two other Miocene
ape samples—Ouranopithecus macedoniensis from
Greece, specimens of which can be sexed based on associ-
ated canines and P3s, 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-spe-
cies taxonomy is not unprecedented when the compara-
tive 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 140:253–264,
2009. V 2009 Wiley-Liss, Inc.
C
However, adopting a multiple-species taxonomy for a fos-
sil sample solely on the basis of excessive size variation
relative to Gorilla and Pongo is problematic for two rea-
sons. First, it is not clear that the upper limit of intra-
specific 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 estab-
lished 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 Univer-
sity, Tempe, AZ 85287-4101, USA. E-mail: jeremiah.scott@asu.edu
Received 3 July 2008; accepted 28 January 2009
DOI 10.1002/ajpa.21059
Published online 8 April 2009 in Wiley InterScience
(www.interscience.wiley.com).
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 post-
canine dentitions have been confidently sexed using
associated canines and P3s (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. lufen-
gensis 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 varia-
tion observed in the sample, unless one appeals to highly
improbable sampling events (Kelley and Plavcan, 1998).
Thus, L. lufengensis extends the known range of intra-
specific 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 accep-
tance, 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. lufen-
gensis 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 varia-
tion in other fossil assemblages was suggested by Wood
(1991), who used Australopithecus boisei to determine
whether variation in A. africanus and A. robustus indi-
cated the presence of multiple species in each of these
hypodigms (for other examples of the use of extinct spe-
cies 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 repre-
sented among extant hominoids. Although the amount of
American Journal of Physical Anthropology
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 dimor-
phism among living primates, and therefore the claim
that the degree of molar dimorphism documented in L.
lufengensis falls outside the range observed in living pri-
mate species has not been adequately tested. Thus, in
this study, we test two hypotheses regarding molar size
dimorphism in primates: (1) Pongo represents the upper-
most extreme of molar dimorphism among living pri-
mates, 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 Ouranopi-
thecus 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. macedonien-
sis 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 Bor-
nean orangutan (Pongo pygmaeus), and the drill (Man-
drillus leucophaeus) (Table 1). The drill was chosen to
represent papionins because Plavcan’s (1990) data set
indicates that Mandrillus probably has the most dimor-
phic 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. leuco-
phaeus data set to unpublished data for M. sphinx col-
lected 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 (mesio-
distal, 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. leu-
cophaeus: 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.
 MODELING SEXUAL DIMORPHISM IN MIOCENE APES
TABLE 1. Sample sizes for the extant comparative taxa and
L. lufengensis
M1 M2 M3
Male Female Male Female Male Female
G. gorilla 43 43 40 40 34
P. pygmaeus 20 20 20 20 18
M. leucophaeus 23 18 24 18 24
L. lufengensis 10 12 11 11 6 10
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 excep-
tion of one additional female M3 (identified by JK after
reexamining the Lufeng data). This sample includes only
molars from associated dentitions, thus making tooth
position (i.e., M1, M2, M3) unambiguous and making it
possible to sex teeth using associated canines and P3s
(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 M1s
and M2s or by including large females in the male sam-
ple 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., multivari-
ate 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 M1s are
also likely to have highly dimorphic M2s), while also
examining the differences at each molar position. For
both analyses, molar size was represented by the geo-
metric mean of the MD and BL dimensions [i.e., (MD 3
BL)1/2].
For the analysis of multivariate molar size dimor-
phism, we used the geometric-mean-based method devel-
oped by Gordon et al. (2008), which is useful for examin-
ing 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 advan-
tage 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):
GM1 h x1 y1 z1 i1=3
¼ 3 3 ;
GM2 x2 y2 z2
where GM1 is the geometric mean of the means (i.e., the
cube root of the product of x1, y1, and z1) for a set of vari-
ables measured on individuals in Group 1 (e.g., males of
a particular species) and GM2 is the geometric mean of
the means for the same set of variables (i.e., x2, y2, and
z2) measured on individuals in Group 2 (e.g., females of
the same species).
255
Thus, multivariate molar size dimorphism can be cal-
culated in two ways. The first way is to calculate the ra-
tio of GMs. For each sex, a measure of multivariate
molar size can be computed as follows:
34
18
GM#ALL ¼ ðGM#1 3 GM#2 3 . . . 3 GM#n Þ1=n ; ð2Þ
16
where GM#1 is the geometric mean of M1, M2, and M3
size [i.e., (M1 3 M2 3 M3)1/3] for the first male, GM#2 is
the geometric mean of M1, M2, and M3 size for the sec-
ond 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 ¼ ðM1ISD 3 M2ISD 3 M3ISD Þ1=3 ; ð3Þ
where M1ISD is the ISD for M1 (i.e., mean M1 size for
males divided by mean M1 size for females), M2ISD is
the ISD for M2, and M3ISD is the ISD for M3. 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., (M1ISD 3 M2ISD 3
M3ISD)1/3], specimens with missing data can be retained
because the ISD for each molar position is calculated in-
dependently 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 within-
sex sample sizes for each molar position were not equal
for any of the species or fossil samples used in the analy-
sis (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 sam-
ple 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 com-
ð1Þ puted. For the analysis of multivariate molar size
dimorphism, bootstrap samples for each molar posi-
tion were randomly grouped together (i.e., one boot-
strap sample of M1s, one bootstrap sample of M2s,
and one bootstrap sample of M3s), 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
M1 ISD calculated using 43 males and 43 females was
American Journal of Physical Anthropology
256 J.E. SCOTT ET AL.
Fig. 1. An example of the procedure used to derive P-values
for pairwise differences among the extant species and L. lufen-
gensis. The top image shows the distribution of pairwise differ-
ences obtained from bootstrapping two samples. The observed
difference between the ISDs of the two samples is 0.06; accord-
ingly, the distribution is centered on 0.06. 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 5 0.0005 (i.e., 1/2001; see text for fur-
ther details).
combined with an M2 ISD calculated using 40 males
and 40 females, and these were combined with an M3
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 ran-
domly paired with those for Sample B, and the differ-
ence between the ISDs for each pairing was calcu-
lated, 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 con-
sidered statistically significant at the a 5 0.05 level.
However, we obtained more precise P-values in the fol-
lowing way:
1. First, we recentered the distribution of pairwise dif-
ferences between Sample A and Sample B on zero
(see Fig. 1), as outlined by Manly (1997, p 99–100).
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 ap-
proximate 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
American Journal of Physical Anthropology
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 (abso-
lute values) greater than or equal to the observed dif-
ference, N is the total number of bootstrap differences,
and one is added to M and N to include the observed
difference.
Note that this test is two tailed. Although the ques-
tions 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 post-
cranial 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 appli-
cations 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 character-
istics (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, allow-
ing 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 gener-
ated 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 con-
trast, because we resampled the comparative samples
and the L. lufengensis sample—and thus generated con-
fidence intervals for all of the samples—matching the
sample size of the fossil sample was unnecessary and
 MODELING SEXUAL DIMORPHISM IN MIOCENE APES 257
TABLE 2. The Sivapithecus and Ouranopithecus samples

Sex Tooth sizec

a
Taxon Specimen assignmentb C P3 M1 M2 M3
Ouranopithecus
RPl-55 Male** 11.7 11.6 14.2 15.9 17.8
RPl-56 Male 10.9 11.3 14.6 15.6 16.6
RPl-75 Male** 12.7 11.6 15.3 16.7 17.4
RPl-76 Male 13.2d – – – 18.1
RPl-89* Male 13.0 11.7 14.5 16.0 18.0
NKT-21 Female 8.5 9.8 11.6 13.6 13.8
RPl-54 Female** 8.5 10.1 12.3 13.6 14.2
RPl-79* Female 9 10.1 12.2 14.1 14.5
RPl-84* Female – 9.6 11.6 13.7 14.1
RPl-88* Female 8.8 9.5 12.8 14.0 14.3
Sivapithecus
GSI D. 197 Male – – 11.7 14.5 –
YPM 13828 Male – – 12.6 14.6 14.9
ONGC/v/790 Male – – 13.0 – –
PUA 1052-69 Male – – – – 13.9
GSI 18039 Male – – 14.9
GSI 18042 Male – – 11.3 – –
GSI D. 199 Female – – – 11.5 12.0
YPM 13806 Female – – – 10.6 9.9
YPM 13825 Female – – 10.1 11.3 –
GSI 18041 Female – – 10.7 – –
GSI 18067 Female – – – – 10.3
PUA 760-69 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 Sivapi-
thecus data were compiled by JK from various sources.
b
For Ouranopithecus, sex assignment is based on canine and Fig. 2. Bivariate plots of canine size vs. M2 size (top) and P3
P3 size (see also Fig. 2); specimens marked with double aster- size vs. M2 size (bottom) in Ouranopithecus macedoniensis.
isks (**) were sexed by Koufos (1995) using canine shape. For Tooth size is the geometric mean of the MD and BL dimensions.
Sivapithecus, sex assignment is based on molar size (see text Specimens that were sexed based on canine shape by Koufos
for further discussion). (1995) are indicated by male (#) and female ($) symbols. Note
c
Tooth size was calculated as the square root of the product of that (1) large and small canines and P3s cluster with the male
MD and BL diameters. and female specimens, respectively, and (2) specimens with
d
Only the MD dimension [maximum length, identified by Kou- large molars are associated with large (male) canines and P3s,
fos (1993) as ‘‘transverse diameter’’] is available for this canine. whereas specimens with small molars are associated with small
(female) canines and P3s. The first and third molars exhibit a
similar pattern.
would have actually reduced the power of the test to
detect differences, making it overly conservative.
After establishing the rank order of molar dimorphism then the cluster of large specimens must be composed of
in the extant species and L. lufengensis, we evaluated males and the cluster of small specimens must be com-
apparent dimorphism in the Ouranopithecus macedo- posed of females (Kelley, 2005). In contrast, the distribu-
niensis and Haritalyangar Sivapithecus samples. The O. tion of Haritalyangar M1s is continuous, making sex
macedoniensis data used for this study were taken from assignment more arbitrary. Based on associations with
the literature (Koufos, 1993; Koufos and de Bonis, 2004), M2s, three of the seven M1s can be tentatively allocated
while the Sivapithecus data were provided by JK. All of to ‘‘male’’ and ‘‘female’’ clusters, while the largest
the O. macedoniensis specimens used here can be confi- (ONGC/v/790) and smallest (PUA 760-69) M1s can also
dently sexed based on associations with canines and P3s be assigned to the ‘‘male’’ and ‘‘female’’ groupings,
(Table 2; Fig. 2; see also Schrein, 2006). This sample respectively. Two teeth—GSI 18041 and GSI 18042—fall
includes newly published specimens from Ravin de la in the middle of the M1 size range. If no overlap between
Pluie (Koufos and de Bonis, 2004) that were not used in males and females is assumed, then the index of ap-
the most recent analysis of variation and sexual dimor- parent sexual dimorphism (ISDA) is between 1.19 and
phism in this fossil ape (Schrein, 2006), and that expand 1.21, depending on whether both specimens are assigned
the sample from n 5 5–6 sexed individuals per molar to one sex or if the larger GSI 18042 is grouped with
position to n 5 9–10, including the addition of three presumed males and the smaller GSI 18041 is grouped
complete female molar rows (for a total of five) and one with presumed females. If females and males do overlap
complete male molar row (for a total of four) (Table 2). in size (with the smaller GSI 18041 placed with pre-
The sample of Sivapithecus mandibular molars from sumed males and the larger GSI 18042 placed with pre-
Haritalyangar is smaller, with n 5 5–7 individuals per sumed females), then the ISDA would be 1.16. We exam-
molar position (Table 2), and none of these specimens ined the effect of these alternative assignments and
can be sexed based on associations with canines or P3s. found that they did not substantively influence the
However, the M2 and M3 samples are each characterized results. Thus, we report only the results of the analyses
by the presence of two markedly disjunct size clusters. If in which GSI 18042 was considered male and GSI 18041
the Haritalyangar assemblage samples a single species, was considered female. Note that these sex assignments

American Journal of Physical Anthropology

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 sam-
ple 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 com-
paring them to the extant taxa and L. lufengensis. Fol-
lowing 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 sam-
ples in size and sex ratio, creating distributions for
determining the probability of obtaining a sample from
G. gorilla, P. pygmaeus, M. leucophaeus, and L. lufengen-
sis 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 replace-
ment is more likely to produce lower P-values than
resampling with replacement given that, at small sample
sizes (e.g., n 5 5–7 in the case of the Sivapithecus analy-
sis), the latter can, in principle, produce samples com-
posed 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 distribu-
tions (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 com-
parative 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. mace-
doniensis sample.3 For example, in the Sivapithecus
analysis, each bootstrap sample contained seven M1s
(four males, three females), six M2s (three males, three
females), and five M3s (two males, three females). For
each bootstrap sample, log-transformed (base e) ISDs
were calculated as described above. The Sivapithecus
and O. macedoniensis ISDAs (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
American Journal of Physical Anthropology
no difference was considered falsified at the a 5 0.05
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 Siva-
pithecus and O. macedoniensis values (including the val-
ues 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 differ-
ence produced by subtracting the fossil sample’s ISDA
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 sam-
ples 1000 times each at a sample size equal to the Sivapithe-
cus 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 cal-
culated 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 differ-
ences as described above. For this analysis, we only exam-
ined 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 sub-
stantively 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 com-
parative samples are not composed entirely of complete
molar rows, the P-values reported for the analyses of
multivariate molar dimorphism should be considered ap-
proximate. For example, consider a case in which the
molars of a species are identical in their degree of dimor-
phism. If a sample of 40 males and 40 females is col-
lected in which the ten largest males are missing their
M3s, then the estimate of dimorphism for the M3 will be
lower than in the other teeth. When the teeth are com-
bined and multivariate molar dimorphism is estimated,
the estimate will be biased due to the missing M3 data.
Such a sample will produce a bootstrap distribution that
is shifted to the left (i.e., toward monomorphism), result-
ing 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 dimor-
phism are likely to be minimal. Therefore, in order to
use samples that are as large as possible, we have cho-
sen 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, sam-
ple 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: M3 sample dimor-
phism is greater in G. gorilla than in P. pygmaeus. The
bootstrap tests for multivariate molar size dimorphism
 MODELING SEXUAL DIMORPHISM IN MIOCENE APES
TABLE 3. Indices of sexual dimorphism for the extant taxa
and L. lufengensis
MALLa M1 M2
G. gorilla 1.08 1.06 1.07
P. pygmaeus 1.10 1.09 1.11
M. leucophaeus 1.13 1.12 1.14
L. lufengensis 1.19 1.18 1.19
a
The abbreviation ‘‘MALL’’ 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 parenthe-
ses (probabilities are two-tailed). Abbreviations: G, Gorilla; P,
Pongo; M, Mandrillus; L, Lufengpithecus.
reveal that G. gorilla and P. pygmaeus are not signifi-
cantly different, whereas M. leucophaeus is significantly
more dimorphic than the two extant apes, and L. lufen-
gensis is significantly more dimorphic than all of the liv-
ing species (Table 4, Fig. 3).
When dimorphism is examined by molar position
(Table 5), P. pygmaeus and G. gorilla differ statistically
only at M2, with P. pygmaeus being more dimorphic at
this position. Mandrillus leucophaeus is significantly
more dimorphic than G. gorilla at M1 and M2—but not
at M3—and is significantly different from P. pygmaeus
only at M3. Lufengpithecus lufengensis is significantly
more dimorphic than the living apes at all molar posi-
tions, but is significantly more dimorphic than M. leuco-
phaeus only at M1 and M2. Some of these differences are
nonsignificant after adjusting a-levels for multiple com-
parisons using the sequential Bonferroni method (e.g.,
Rice, 1989); the results that remain significant are: M.
leucophaeus more dimorphic than G. gorilla for M1 and
M2, L. lufengensis more dimorphic than the living great
apes for all molar positions, and L. lufengensis more
dimorphic than M. leucophaeus for M1. 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 M1 and M3 are sim-
ilar to those for multivariate molar size, but for M2, 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 signifi-
cantly greater than in any of the comparative taxa,
including L. lufengensis, even after sequential Bonfer-
roni adjustment (Table 7, Fig. 4B). The only exception is
apparent dimorphism in the M1 sample, which cannot be
statistically distinguished from M1 dimorphism in
L. lufengensis. Thus, the Haritalyangar M2 and M3
259
M3
1.11
1.09
1.15
1.19
Fig. 3. Bootstrap distributions for multivariate molar size
dimorphism for the extant species and L. lufengensis. The mid-
dle 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 spe-
cies 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. Mandril-
lus leucophaeus emerges as the most dimorphic extant
primate when the molar row is considered in its entirety,
but the drill cannot be consistently distinguished statis-
tically from either P. pygmaeus or G. gorilla at individ-
ual 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-
American Journal of Physical Anthropology
260 J.E. SCOTT ET AL.

Mandrillus

in parentheses (probabilities are two-tailed). Abbreviations: G, Gorilla; P, Pongo; M, Mandrillus; L, Lufengpithecus. An asterisk (*) indicates that the comparison is not significant
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
tan are more dimorphic than those of the gorilla, Uchida

5 (0.0885)
(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

M [ P* (0.0305)
genera.

L [ P (0.001)
It is also important to point out that our ability to
Pongo
detect differences among the living taxa is hampered in
M3

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 mea-
surement error in both the numerator (male mean) and
L [ G (0.0005)

denominator (female mean) (see discussion in Smith,

5 (0.4448)
5 (0.1149)

1999). Second, because males and females constitute sep-

Gorilla

arate components of the ISD, the effective sample sizes

TABLE 5. Results of the bootstrap tests for interspecific differences at individual molar positions

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
L [ M* (0.0195)

in canine, craniofacial, and body-mass dimorphism

Mandrillus

across the Anthropoidea (Plavcan, 2001), it is not sur-

prising that many of the comparisons in this study are
nonsignificant. Future studies will require larger sam-
ples to establish whether the differences in sample ISDs
observed between G. gorilla and P. pygmaeus and
between P. pygmaeus and M. leucophaeus at individual
L [ P (0.0005)

molar positions reflect true population differences.

5 (0.2214)

In spite of the conservative nature of the statistical

Pongo
M2

tests, our results confirm that the molars of L. lufengensis

are more dimorphic than those of living apes. Our analy-
sis 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 M1, and prob-
P [ G* (0.0325)
M [ G (0.0005)
L [ G (0.0005)

ably M2 as well). That we were able to detect significant

differences between L. lufengensis and the extant species
Gorilla

despite the fact that the Lufeng sample contains only n 5

16–22 individuals per molar position highlights just how
much greater dimorphism is in the molar teeth of this fos-
sil ape. Notably, the analysis of multivariate molar size
dimorphism indicates that M. leucophaeus is intermediate
after sequential Bonferroni adjustment (applied within each variable).
L [ M (0.012)

between the great apes and L. lufengensis (see Fig. 3).

Mandrillus

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 dimor-
phism compared to extant apes, its inclusion in our anal-
L [ P (0.0005)
5 (0.1339)

ysis of the O. macedoniensis material does not alter pre-

Pongo

vious conclusions that a single-species interpretation of

M1

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)
M [ G (0.0015)

results in slightly lower indices of apparent sexual

L [ G (0.0005)
5 (0.1009)

dimorphism (ISDAs) for O. macedoniensis in comparison

Gorilla

to those for the smaller sample used by Schrein (2006),

there is still no overlap in size between specimens identi-
fied as male and those identified as female on the basis
of canine or P3 size/morphology (see Table 2, Fig. 2), an
attribute also evident in the L. lufengensis sample
Lufengpithecus

(Kelley and Plavcan, 1998). In our analysis, M2 is the

only O. macedoniensis variable for which size dimor-
Mandrillus

phism cannot be statistically distinguished from that of

any of the extant species, except for G. gorilla prior to
Pongo

sequential Bonferroni adjustment. Schrein (2006)

obtained somewhat different results for her comparisons

American Journal of Physical Anthropology

 MODELING SEXUAL DIMORPHISM IN MIOCENE APES 261
TABLE 6. Bootstrap results for the Ouranopithecus comparisons
Ouranopithecus
MALL (ISD 5 1.20) M1 (ISD 5 1.21) M2 (ISD 5 1.16) M3 (ISD 5 1.24)
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 M1 of Ouranopithecus could be confidently identified
as being more dimorphic than in gorillas and orangu-
tans), which is probably attributable to the smaller sam-
ple 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 dimor-
phism in O. macedoniensis cannot be statistically distin-
guished 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 demon-
strates 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 exis-
tence of only a single species within the O. macedonien-
sis dental sample: large specimens are male, small speci-
mens 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, Fig. 4. Bootstrap distributions for multivariate molar size
dimorphism generated by bootstrapping the extant taxa at a
Xirochori, and Nikiti, one characterized by an extreme sample size and sex ratio identical to that of the (A) Ouranopi-
degree of molar dimorphism relative to living primates. thecus and (B) Sivapithecus samples. Only the M. leucophaeus
The results of the Sivapithecus analysis, on the other and L. lufengensis distributions are shown; the G. gorilla and
hand, do not lend themselves to easy interpretation. Kel- P. pygmaeus distributions would be to the left of the Mandrillus
ley (2005) found that measures of variation for the distribution. The solid vertical line in A indicates the ISDA for
Haritalyangar M2s and M3s are statistically significantly the O. macedoniensis sample; the dashed vertical line in B indi-
higher than those for L. lufengensis, suggesting an even cates the ISDA for the Sivapithecus sample.
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 Lufeng distribution, and Kelley’s (2005) analysis of vari-
dimorphism in the Haritalyangar M2 and M3 samples is ation in the Haritalyangar maxillary molars indicates
highly unlikely to have come from a species as dimorphic that the same would certainly be true for M1, M2, and
as L. lufengensis; none of the samples bootstrapped from M3, as levels of apparent dimorphism in these teeth are
the Lufeng sample produced ISDs as high as the ISDAs slightly lower than or similar to those of L. lufengensis
for the Haritalyangar M2, M3, or multivariate molar (M2 ISDA 5 1.16; M3 ISDA 5 1.19; an ISDA cannot be
size. In contrast, the ISDA for M1 falls well within the calculated for M1 due to the fact that the specimens

American Journal of Physical Anthropology

262 J.E. SCOTT ET AL.
TABLE 7. Bootstrap results for the Sivapithecus comparisons
Sivapithecus
MALL (ISDA 5 1.29) M1 (ISDA 5 1.20) M2 (ISDA 5 1.32) M3 (ISDA 5 1.34)
Gorilla S [ G (0.001) S [ G (0.001) S [ G (0.001) S [ G (0.003)
Pongo S [ P (0.001) S [ P (0.004) S [ P (0.001) S [ P (0.001)
Mandrillus S [ M (0.001) S [ M (0.01) S [ M (0.001) S [ M (0.001)
Lufengpithecus S [ L (0.001) 5 (0.6533) S [ L (0.001) S [ L (0.001)

Nonsignificant differences are indicated by an equality symbol; greater-than symbols indicate significance and the direction of dif-
ference. P-values for each comparison are given in parentheses (probabilities are two-tailed). Abbreviations: G, Gorilla; P, Pongo; M,
Mandrillus; L, Lufengpithecus; S, Sivapithecus.

form a fairly tight cluster, precluding the use of size as a
criterion for sex assignment; see Fig. 8.3 in Kelley,
2005).
If we accept that fossil species are not constrained by
currently known limits of intraspecific variation and sex-
ual dimorphism (Kelley, 1993; Kelley and Plavcan, 1998;
Plavcan and Cope, 2001; Schrein, 2006), then the single-
species hypothesis cannot be definitively ruled out for
the Haritalyangar sample, despite the fact that such a
species would have to be even more dimorphic than
L. lufengensis. While the high levels of size variation
and apparent sexual dimorphism in the Haritalyangar
second and third mandibular molars can be used to
argue for the presence of two species (Kelley, 2005), such
evidence cannot be used as the sole basis for rejecting
the single-species hypothesis (Kelley and Plavcan, 1998;
Plavcan and Cope, 2001; Schrein, 2006). Presumably,
there is a limit to the amount of molar dimorphism that
can be expressed in primates, but whether or not
L. lufengensis (and O. macedoniensis) represents that
limit is currently unknown. Other recently reported cases
of extreme dimorphism, including the middle Miocene
hominoid Griphopithecus alpani from Pas  alar, Turkey
[apparent even though a second species has been identi-
fied in the Pas  alar assemblage (Humphrey and Andrews,
2008; Kelley et al., 2008)], and the Oligocene early catar-
rhine Aegyptopithecus zeuxis from the Fayum, Egypt
(Simons et al., 2007), could provide insight into this issue.
On the other hand, given that the Haritalyangar sec-
ond and third mandibular molars do not fit any of our
comparative models of sexual dimorphism, additional
lines of evidence are required before accepting a single-
species taxonomy for this sample, as the two-species tax-
onomy cannot be rejected based on current evidence
either. In this context, demonstration that the size clus-
ters evident in the M2 and M3 samples are truly homoge-
neous with respect to sex using canine and/or P3 size
and morphology, as was done for the L. lufengensis and
O. macedoniensis samples (Kelley and Xu, 1991; Kelley,
1993; Schrein, 2006; see Fig. 2), would provide support
for the single-species hypothesis. Conversely, a two-spe-
cies hypothesis—with each size cluster representing a
different species—would be supported if it is shown that
males and females are present in both clusters. Unfortu-
nately, the current sex assignments are based on size
because there are no associated canines or P3s, which
precludes unequivocally linking the high levels of varia-
tion in the sample to sex differences.
Finally, it is important to note that the temporal span
for Sivapithecus at Haritalyangar, at approximately
400,000 years (Pillans et al., 2005), is certainly greater
than for either the Lufengpithecus or Ouranopithecus
samples, but how much of this range is represented in
the portion of the Haritalyangar sample analyzed here is
unknown. Thus, based on the current evidence, the tax-
onomy of the Haritalyangar Sivapithecus material
remains ambiguous (see also Kelley, 2005).
CONCLUSIONS
Our analysis of sexual dimorphism in molar size in
great apes and the drill demonstrates that the latter spe-
cies exceeds the extant hominoids in some aspects of
molar dimorphism. Thus, Pongo does not possess the
most dimorphic molars among living primates, though
they are among the most dimorphic. These results indi-
cate that Mandrillus leucophaeus (and perhaps other
papionins) should be included in extant comparative
samples when evaluating variation in fossil hominoid
assemblages, particularly when variation appears to
exceed that in gorillas and orangutans. We confirmed
the extreme degree of molar dimorphism in the Lufeng-
pithecus lufengensis sample relative to extant species,
thereby establishing the importance of using this species
as a comparative analogue to represent levels of intra-
specific variation and sexual dimorphism not sampled
among living primates (e.g., Kelley, 2005).
Using the drill and L. lufengensis samples as part of
our comparative framework, we analyzed apparent size
dimorphism in the molars of two other late Miocene
hominoid assemblages that have been identified as possi-
bly comprising single highly dimorphic species. Our
results for the Haritalyangar Sivapithecus sample show
that, if only one species is present at this site, then it is
more dimorphic than L. lufengensis for at least some
teeth. Given the uncertainties concerning the sex of the
individuals in this sample, and because the sample as a
whole does not fit any of our comparative models, we are
unable to choose with confidence between the hypotheses
that the sample contains (1) a single, extremely size-
dimorphic species or (2) two species, differing primarily
in size. On the other hand, in the case of Ouranopithecus
macedoniensis, our analysis shows that the level of
molar dimorphism required under a single-species taxon-
omy is fully compatible with the degree of sexual dimor-
phism in the molars of L. lufengensis, thus demonstrat-
ing that, while a single-species taxonomy for the O. mac-
edoniensis sample requires a degree of molar
dimorphism that is extreme compared to that in living
anthropoids, it is not extreme in comparison to at least
one other hominoid species from the late Miocene of
Eurasia.
ACKNOWLEDGMENTS
We thank Xu Qinghua of the Institute of Vertebrate
Paleontology and Paleoanthropology, Beijing, China, for

American Journal of Physical Anthropology

 MODELING SEXUAL DIMORPHISM IN MIOCENE APES
providing access to the Lufengpithecus dental data while
sponsored by grants from the National Academy of Sci-
ences and the Leakey Foundation to JK. We are grateful
to Michael Plavcan for answering questions regarding
his Mandrillus leucophaeus data, and Stephen Frost
kindly allowed us to examine Mandrillus sphinx data
that he, Rachel Nuger, and Michelle Singleton collected.
Dennis Young provided invaluable statistical advice.
This manuscript benefitted from thoughtful comments
and suggestions by Editor-in-Chief Christopher Ruff, the
associate editor, and two anonymous reviewers. The
bootstrap tests were performed using Microsoft Excel
macros written by Charles Lockwood—friend, mentor,
and colleague. He is missed.
LITERATURE CITED
Albrecht GH, Miller JMA. 1993. Geographic variation in pri-
mates: a review with implications for interpreting fossils. In:
Kimbel WH, Martin LB, editors. Species, species concepts,
and primate evolution. New York: Plenum. p 123–161.
Baab KL. 2008. The taxonomic implications of cranial shape
variation in Homo erectus. J Hum Evol 54:827–847.
Bunce M, Worthy TH, Ford T, Hoppit W, Willerslev E, Drum-
mond A, Cooper A. 2003. Extreme reversed sexual size dimor-
phism in the extinct New Zealand moa Dinornis. Nature
425:172–175.
Cope DA. 1993. Measures of dental variation as indicators of
multiple taxa in samples of sympatric Cercopithecus species.
In: Kimbel WH, Martin LB, editors. Species, species concepts,
and primate evolution. New York: Plenum. p 211–237.
Cope DA, Lacy MG. 1992. Falsification of a single species hy-
pothesis using the coefficient of variation: a simulation
approach. Am J Phys Anthropol 89:359–378.
Gordon AD, Green DJ, Richmond BG. 2008. Strong postcranial
size dimorphism in Australopithecus afarensis: results from
two new resampling methods for multivariate data sets with
missing data. Am J Phys Anthropol 135:311–328.
Harmon EH. 2006. Size and shape variation in Australopithecus
afarensis proximal femora. J Hum Evol 51:217–227.
Harvati K, Frost SR, McNulty KP. 2004. Neanderthal taxonomy
reconsidered: implications of 3D primate models of intra- and
interspecific differences. Proc Natl Acad Sci USA 101:1147–
1152.
Humphrey LT, Andrews P. 2008. Metric variation in the postca-
nine teeth from Pas alar. Turkey. J Hum Evol 54:503–517.
Huynen L, Millar CD, Scofield RP, Lambert DM. 2003. Nuclear
DNA sequences detect species limits in ancient moa. Nature
425:175–178.
Kay RF. 1982a. Sexual dimorphism in Ramapithecinae. Proc
Natl Acad Sci USA 79:209–212.
Kay RF. 1982b. Sivapithecus simonsi, a new species of Miocene
hominoid, with comments on the phylogenetic status of the
Ramapithecinae. Intl J Primatol 3:113–173.
Kelley J. 1986. Species recognition and sexual dimorphism in
Proconsul and Rangwapithecus. J Hum Evol 15:461–495.
Kelley J. 1993. Taxonomic implications of sexual dimorphism in
Lufengpithecus. In: Kimbel WH, Martin LB, editors. Species,
species concepts, and primate evolution. New York: Plenum.
p 429–457.
Kelley J. 1995a. Sexual dimorphism in canine shape among
extant great apes. Am J Phys Anthropol 96:365–389.
Kelley J. 1995b. Sex determination in Miocene catarrhine pri-
mates. Am J Phys Anthropol 96:391–417.
Kelley J. 2005. Twenty-five years contemplating Sivapithecus
taxonomy. In: Lieberman DE, Smith RJ, Kelley J, editors.
Interpreting the past: essays on human, primate, and mam-
mal evolution in honor of David Pilbeam. Boston: Brill Aca-
demic. p 123–143.
Kelley J, Andrews P, Alpagut B. 2008. A new hominoid species
from the middle Miocene site of Pas  alar. Turkey. J Hum Evol
263
54:455–479.
Kelley J, Etler D. 1989. Hominoid dental variability and species
number at the late Miocene site of Lufeng, China. Am J Pri-
matol 18:15–34.
Kelley J, Plavcan JM. 1998. A simulation test of hominoid species
number at Lufeng, China: implications for the use of the coeffi-
cient of variation in paleotaxonomy. J Hum Evol 35:577–596.
Koufos GD. 1993. Mandible of Ouranopithecus macedoniensis
(Hominidae, Primates) from a new late Miocene locality of
Macedonia (Greece). Am J Phys Anthropol 91:225–234.
Koufos GD. 1995. The first female maxilla of the hominoid Our-
anopithecus macedoniensis from the late Miocene of Macedo-
nia, Greece. J Hum Evol 29:385–399.
Koufos GD, de Bonis L. 2004. New material of Ouranopithecus
macedoniensis from late Miocene of Macedonia (Greece) and
study of its dental attrition. Geobios 39:223–243.
Kramer A. 1993. Human taxonomic diversity in the Pleistocene:
does Homo erectus represent multiple hominid species? Am J
Phys Anthropol 91:161–171.
Lieberman DE, Pilbeam DR, Wood BA. 1988. A probabilistic
approach to the problem of sexual dimorphism in Homo
habilis: a comparison of KNM-ER 1470 and KNM-ER 1813. J
Hum Evol 17:503–511.
Lockwood CA. 1999. Sexual dimorphism in the face of
Australopithecus africanus. Am J Phys Anthropol 108:97–
127.
Lockwood CA, Kimbel WH, Johanson DC. 2000. Temporal
trends and metric variation in the mandibles and dentition of
Australopithecus afarensis. J Hum Evol 39:23–55.
Lockwood CA, Richmond BG, Jungers WL, Kimbel WH. 1996.
Randomization procedures and sexual dimorphism in Austral-
opithecus afarensis. J Hum Evol 31:537–548.
Mahler P. 1973. Metric variation in the pongid dentition. PhD
dissertation, University of Michigan, Ann Arbor.
Manly BFJ. 1997. Randomization, bootstrap and Monte Carlo
methods in evolutionary biology, 2nd ed. London: Chapman &
Hall.
Martin L. 1991. Teeth, sex and species. Nature 352:111–112.
Martin LB, Andrews P. 1993. Species recognition in middle Mio-
cene hominoids. In: Kimbel WH, Martin LB, editors. Species,
species concepts, and primate evolution. New York: Plenum.
p 393–427.
Pillans B, Williams M, Cameron D, Patnaik R, Hogarth J, Sahni
A, Sharma JC, Williams F, Bernor RL. 2005. Revised correla-
tion of the Haritalyangar magnetostratigraphy, Indian Siwa-
liks: implications for the age of the Miocene hominids Indopi-
thecus and Sivapithecus, with a note on a new hominid tooth.
J Hum Evol 48:507–515.
Plavcan JM. 1990. Sexual dimorphism in the dentition of extant
anthropoid primates. PhD Dissertation, Duke University.
Plavcan JM. 1993. Catarrhine dental variability and species rec-
ognition in the fossil record. In: Kimbel WH, Martin LB, edi-
tors. Species, species concepts, and primate evolution. New
York: Plenum. p 239–263.
Plavcan JM. 1994. Comparison of four simple methods for esti-
mating sexual dimorphism in fossils. Am J Phys Anthropol
94:465–476.
Plavcan JM. 2001. Sexual dimorphism in primate evolution.
Yearb Phys Anthropol 44:25–53.
Plavcan JM. 2002. Taxonomic variation in the patterns of
craniofacial dimorphism in primates. J Hum Evol 42:579–
608.
Plavcan JM. 2003. Scaling relationships between craniofacial
sexual dimorphism and body mass dimorphism in primates:
implications for the fossil record. Am J Phys Anthropol
120:38–60.
Plavcan JM, Cope DA. 2001. Metric variation and species recog-
nition in the fossil record. Evol Anthropol 10:204–222.
Reno PL, Meindl RS, McCollum MA, Lovejoy CO. 2003. Sexual
dimorphism in Australopithecus afarensis was similar to
that of modern humans. Proc Natl Acad Sci USA 100:9404–
9409.
Rice WR. 1989. Analyzing tables of statistical tests. Evolution
43:223–225.
American Journal of Physical Anthropology
264 J.E. SCOTT ET AL.
Richmond BG, Jungers WL. 1995. Size variation and sexual
dimorphism in Australopithecus afarensis and living homi-
noids. J Hum Evol 29:229–245.
Ruff CB, Walker A, Teaford MF. 1989. Body mass, sexual dimor-
phism and femoral proportions of Proconsul from Rusinga
and Mfangano islands, Kenya. J Hum Evol 18:515–536.
Schrein CM. 2006. Metric variation and sexual dimorphism in
the dentition of Ouranopithecus macedoniensis. J Hum Evol
50:460–468.
Scott JE, Lockwood CA. 2004. Patterns of crown size and shape
variation in great apes and humans and species recognition
in the hominid fossil record. Am J Phys Anthropol 125:303–
319.
Scott JE, Stroik LK. 2006. Bootstrap tests of significance and
the case for humanlike skeletal-size dimorphism in Australo-
pithecus afarensis. J Hum Evol 51:422–428.
Setchell JM, Lee PC, Wickings EJ, Dixson AF. 2001. Growth
and ontogeny of sexual size dimorphism in the mandrill
(Mandrillus sphinx). Am J Phys Anthropol 115:349–360.
Silverman N, Richmond B, Wood B. 2001. Testing the taxonomic
integrity of Paranthropus boisei sensu strico. Am J Phys
Anthropol 115:167–178.
Simons EL, Seiffert ER, Ryan TM, Attia Y. 2007. A remarkable
female cranium of the early Oligocene anthropoid Aegyptopi-
thecus zeuxis. Proc Natl Acad Sci USA 21:8731–8736.
Skinner MM, Gordon AD, Collard NJ. 2006. Mandibular size
and shape variation in the hominins at Dmanisi, Republic of
Georgia. J Hum Evol 51:36–49.
Smith RJ. 1999. Statistics of sexual size dimorphism. J Hum
Evol 36:423–459.
American Journal of Physical Anthropology
Smith RJ, Jungers WL. 1997. Body mass in comparative prima-
tology. J Hum Evol 32:523–559.
Teaford MF, Walker A, Mugaisi GS. 1993. Species discrimina-
tion in Proconsul from Rusinga and Mfangano Islands, Kenya.
In: Kimbel WH, Martin LB, editors. Species, species concepts,
and primate evolution. New York: Plenum. p 373–392.
Uchida A. 1996a. Craniodental variation among the great apes.
Cambridge, MA: Peabody Museum of Archaeology and Eth-
nology, Harvard University.
Uchida A. 1996b. Dental variation of Proconsul from the Tin-
deret region. Kenya. J Hum Evol 31:489–497.
Villmoare B. 2005. Metric and non-metric randomization meth-
ods, geographic variation, and the single-species hypothesis
for Asian and African Homo erectus. J Hum Evol 49:680–701.
Walker A, Teaford MF, Martin L, Andrews P. 1993. A new spe-
cies of Proconsul from the early Miocene of Rusinga/Mfangano
islands, Kenya. J Hum Evol 25:43–56.
Wood BA. 1976. The nature and basis of sexual dimorphism in
the primate skeleton. J Zool Lond 180:15–34.
Wood BA. 1991. A palaeontological model for determining the
limits of early hominid taxonomic variability. Paleontol Afr
28:71–77.
Wood BA, Xu Q. 1991. Variation in the Lufeng dental remains.
J Hum Evol 20:291–311.
Wu R, Oxnard CE. 1983a. Ramapithecines from China: evidence
from tooth dimensions. Nature 306:258–260.
Wu R, Oxnard CE. 1983b. Ramapithecus and Sivapithecus from
China: some implications for higher primate evolution. Am J
Primatol 5:303–344.