Visualizing Facial Shape Regression upon 2^} n^ d to 4^ t^{ h Digit Ratio and Testosterone

Full text

Coll. Antropol. 29 (2005) 2: 415–419
UDC 572.54.087-055.1:612.616.31
Original scientific paper
Visualizing Facial Shape Regression upon
2nd to 4th Digit Ratio and Testosterone
Katrin Schaefer1, Bernhard Fink2, Philipp Mitteroecker1, Nick Neave3and Fred L. Bookstein1,4
1Department of Anthropology, University of Vienna, Vienna, Austria
2Department for Sociobiology/Anthropology, University of Goettingen, Goettingen, Germany
3Human Cognitive Neuroscience Unit, Northumbria University, Newcastle upon Tyne, UK
4Department of Statistics, University of Washington, Seattle, USA
ABSTRACT
Sex steroids are supposed to moderate the differences between male and female facial characteristics. Studies on
women’s preferences for male faces reported increased preferences for facial architecture developed under the influence
of testosterone as this may indicate masculinity, dominance and social status. Recent research demonstrates that facial
sexual dimorphism does not only develop at puberty but may be organized much earlier in ontogeny. However, the ac-
tual cause and timing of variation in facial shape due to sex-steroids remains speculative. This study uses data from
Neave and colleagues1who measured digit ratio (2D:4D) as a proxy to prenatal testosterone and also salivary testoster-
one samples in order to study differential effects of androgens on perceived male facial shape. Male facial shape was re-
gressed upon 2D:4D ratio and circulating levels of testosterone by means of geometric morphometric methods. We
found some evidence for opposite effects of early androgen action (via 2D:4D ratio) on the upper and the lower face re-
spectively (i.e. low 2D:4D ratio results in a relatively robust and prominent lower face), whereas circulating testoster-
one seems to cause a rather uniform elongation of the face. Local deformations primarily show pronounced and medi-
ally tailed eyebrows for the shapes associated with increasing salivary testosterone. These preliminary results suggest
that prenatal and pubertal testosterone have differential effects on male facial shape that should be considered in fu-
ture studies on women’s preferences towards male facial appearance.
Key words: 2D:4D, digit ratio, testosterone, males, facial shape, geometric morphometrics
Introduction
Some proportions of the human face are sexually di-
morphic. Typical male traits develop under the influ-
ence of testosterone whereas female traits are formed
under the absence of high testosterone. In pubertal
males, a high testosterone/estrogen ratio is supposed to
facilitate the lateral growth of the cheekbones, mandi-
bles and chin, the forward growth of the bones of the
eyebrow ridges, and the lengthening of the lower face2,3.
Evolutionary psychology suggests that females should
have evolved preferences for such facial characteristics
and consider them attractive because they signal health,
developmental stability, dominance and masculinity4.
There are several studies that lend support to the view
that women’s preferences for facial 'hormone markers'
may reflect an index of mate quality5. However, evi-
dence on the causal relation between testosterone and
male facial characteristics still remains speculative,
though a recent study by Penton-Voak and Chen6re-
ports preliminary findings of a link between high salivary
testosterone and 'masculine' facial appearance.
Neave, Laing, Fink and Manning1found that the rel-
ative length of the 2nd (index) to 4th (ring) finger (2D:4D
ratio), which is known to be negatively correlated with
prenatal testosterone7,8, was significantly negatively
correlated with ratings of perceived male dominance
and masculinity, but not attractiveness. In contrast to
Penton-Voak and Chen6, circulating levels of testoster-
one analysed from salivary samples did not correlate
with ratings of dominance, masculinity or attractive-
ness.
A more recent study by Fink and colleagues9demon-
strates that prenatal sex steroids (as measured via
2D:4D ratio), particularly testosterone, affect facial ar-
415
Received for publication August 29, 2005

chitecture. Clearly, characteristics that are considered
typically 'male' were found to correspond to low 2D:4D
ratios, whereas some typically 'female' features corre-
sponded to high 2D:4D ratios. These authors further
showed that 2D:4D affects male and female face shape
by similar patterns, but was three times more intense in
men than in women. Hence, testosterone seems to be a
powerful hormone with regard to its effects on face
shape and this may apply not only at puberty but much
earlier in ontogeny. However, this suggestion contrasts
earlier findings about associations between actual lev-
els of testosterone and facial characteristics.
There is, to the best of our knowledge, no data avail-
able that deal with the action of both early and adult cir-
culating testosterone in more detail. In the present
study we examine adult male facial shape by means of
geometric morphometric methods10 (see methods sec-
tion) based on the data of Neave and colleagues1and
present preliminary results and visualization of facial
shape in relation to 2D:4D ratio and circulating testos-
terone. On the basis of the recent results by Fink and
others9, namely the fact that effects of 2D:4D on face
shape show a different pattern from actual facial sexual
dimorphism, we hypothesize that the inconsistency in
literature about the testosterone and face shape associ-
ation may be due to testosterone determining dissimilar
shape patterns at different developmental stages.
Material and Methods
We used the data set as published in Neave et al.1
Therefore we only provide basic information concerning
data recording, and focus on the novel methodology of
the present study, i.e., the analysis and visualization of
prenatal and adult testosterone effects upon male facial
shape by means of geometric morphometric measure-
ment.
Data recording
Our data set comprised 48 heterosexual, right-hand-
ed male participants aged 18–33 years (mean age= 21.3,
SD= 3.4). Measurements of the lengths of the 2nd (index
finger, 2D) and 4th (ring finger, 4D) digits were avail-
able as were digital facial images. Calculated repeat-
abilities for successive measurements were reported in
the original paper to be high. Forty-six participants pro-
vided a salivary sample from which circulating levels of
free testosterone were measured with Coat-A-Count to-
tal testosterone kits (Euro/DPC, Wales, UK) (for details
on the recoding procedure and assay methodology see
Neave et al.1).
Soft tissue measurement points
Fifty-one somatometric landmarks11 were selected to
cover all facial regions (Figure 1) and digitized within
the ’Beautynet’ software environment as two-dimen-
sional (2D) coordinates on each of the photographs (see
for details Grammer et al.12) by one of the authors (B.F.)
experienced in this task.
Geometric morphometric analyses
For the analyses of facial shape, we use geometric
morphometrics, an approach deriving from multivariate
statistical analyses of Cartesian coordinate data10,13.
For the registration of the 46 configurations with 51 2D
landmark points each, we applied the standard Procrus-
tes analysis method14–16, which results in a set of vectors
of 102 shape coordinates.
In order to determine the shape change associated
with a parameter of interest, a linear regression func-
tion is calculated for every shape coordinate separately,
where the slopes of the functions predict the shape
change that occurs within one unit of the independent
variable. The alpha level of significance of the regres-
sions was calculated by a Monte Carlo permutation
test17 with the generalized shape variance explained by
the regression (sum of the variances explained by 2D:4D
or testosterone over all the shape coordinates sepa-
rately) as the test statistic. Thin-plate spline (TPS) de-
formation grids are used for visualization. Bookstein10
proposed the TPS interpolation formalism to visualize
differences in the positions of landmarks by modelling
the deformations taking place between the landmarks,
i.e. in all regions without landmark points. Shape re-
gressions are visualized by computing the mean form
(consensus) and adding the respective slope to the corre-
sponding shape variable. Since the changes within one
unit are often small, the slopes are usually multiplied
by an arbitrary factor to enhance visualization. All sta-
tistical computations were done in Mathematica 5.0
(Wolfram Research, Inc.).
K. Schaefer et al.: Facial Shape and Testosterone, Coll. Antropol. 29 (2005) 2: 415–419
416
Fig. 1. Fifty-one somatometric landmarks were digitized as two-
-dimensional (2D) coordinates on each of the facial photographs.

Results
The 2D:4D ratios were similar for both hands (left hand:
X±SD=0.96±0.03, range 0.89–1.02; right hand: X±SD=
0.96±0.03, range 0.88–1.01), and significantly positively
correlated (Pearson’s r=0.639, p<0.001). As such, they
were averaged among both hands and the analyses per-
formed with the resulting composite ratio.
A two-block Partial Least Squares analysis10 of the
shape coordinates (first block) against testosterone and
2D:4D ratio (second block), determining the import of ei-
ther block on the other, revealed that facial shape is pre-
dicted by both factors with almost equal power (singular
values 0.010 and 0.008). The two shape regression vec-
tors of 2D:4D ratio and testosterone, however, are al-
most orthogonal to each other (angle of 80.6 degree), in-
dicating no interrelation among them.
In order to investigate to what extent and into which
direction the 2D:4D ratio systematically affects male fa-
cial features we regressed the 46 vectors of the 102
shape coordinates (resulting from the 51 somatometric
landmarks per image after the Procrustes fit) upon the
46 respective 2D:4D ratio values. This shape regression
is visualized in Figure 2a by means of TPS deformation
grids. The middle panel shows the mean landmark con-
figuration of the sample, characterized by the rectangu-
lar grid with uniform quadratic units, the neighbouring
panels represent both directions of the predicted defor-
mation. The panel on the left side shows the deforma-
tion associated with higher 2D:4D ratio (i.e. lower levels
of prenatal testosterone). The main effect is a grid con-
traction in the lower face and a widening of the meshes
in the upper face, indicating a smaller chin, a narrowing
of the midface, and a relative expansion of the forehead.
The opposite holds for the image on the right, visualiz-
ing the shape changes with lower 2D:4D ratio (i.e.
higher levels of prenatal testosterone): the grid indi-
cates a relative enlargement of the whole lower face
achieved by a broadening of the zygomatic arch and a
more prominent chin. The eyebrows follow this pattern.
Salivary testosterone levels of the sample ranged
from 4.49 to 15.95 ng/dL (X=8.94, SD=2.67). In order to
investigate to what extent and into which direction tes-
tosterone systematically affects male facial features, we
performed a shape regression upon these values. Figure
2b visualizes that regression using TPS deformation
grids for the predicted transformation in both directions
from the average (middle panel). The left grid shows the
facial shape change associated with decreasing salivary
K. Schaefer et al.: Facial Shape and Testosterone, Coll. Antropol. 29 (2005) 2: 415–419
417
Fig. 2. (a) Thin plate spline visualization grids of the shape regression upon 2D:4D ratio. The left deformation grid shows the consen-
sus (middle panel) deformation towards lower prenatal testosterone levels (higher 2D:4D ratio), while the right grid corresponds to
shape changes associated with higher prenatal testosterone levels (both splines multiplied by factor of 0.25 for visibility). Some land-
marks (black dots) are connected by lines (cubic splines) for illustration. (b) Thin plate spline visualization grids of the shape regres-
sion upon salivary testosterone. The left deformation grid shows the consensus deformation towards lower salivary testosterone levels,
while the right grid corresponds to shape changes associated with increasing testosterone levels (splines exaggerated by factor of 20
for visibility).

testosterone levels whereas the right grid illustrates the
deformation corresponding to increasing levels. The
most obvious and general difference is the primarily
horizontal grid compression in the former as opposed to
a more vertical stretching in the grid describing higher
testosterone levels: a rounded versus an elongated face.
A second, more local distinction is found in the region of
the eyes: while the shapes that correspond to lower tes-
tosterone show small eyebrows and narrow eyes, the de-
formation towards higher testosterone levels shows
rounder eyes with eyebrows expanding laterally and, es-
pecially, medially. In both shape regressions the ex-
plained variance does not exceed 5% (2.5% for 2D:4D,
2.1 % for salivary testosterone), and does not yield sig-
nificance by permutation test17 after 10,000 permuta-
tions (p>0.05, p=0.6 for 2D:4D, p=0.3 for circulating tes-
tosterone).
Discussion
The present study examined male facial shape in re-
lation to indicators of prenatal and circulating levels of
testosterone (as measured via 2D:4D ratio and saliva re-
spectively). We found some evidence that early andro-
gen action operates in contrasting ways on the upper
and the lower face while circulating levels of testoster-
one seem to cause a rather uniform elongation of the
face. Neave et al.1suggested that high prenatal levels of
testosterone (resulting in a low 2D:4D ratio) 'organise'
male facial features (presumably activated during pu-
berty) that may subsequently reflect dominance/mascu-
line characteristics. In light of the present analysis, the
visualization of face shape regressions upon 2D:4D ratio
indicates that low 2D:4D faces are evidently more ro-
bust than high 2D:4D faces. Our data further indicate
that circulating levels of testosterone in turn do not
cause this effect on male facial shape.
The pattern for the effect of 2D:4D on male facial
shape demonstrated here is in accord with the recent
study by Fink et al.9who also found low 2D:4D ratios to
be associated with the same robust facial shape in both
men and women. Interestingly enough, this connection
was found to be significant only in men which points to
testosterone as the responsible candidate rather than to
the testosterone/estrogen ratio9. Moreover, this pattern
supports the notion that 2D:4D ratio manifests itself in
a broadening of the os zygomaticum and the mandible,
facial characteristics that are known to be assessed as
dominant and masculine1,18,19. Taken together it seems
that there is accumulating evidence for the effect of pre-
natal testosterone upon facial shape: higher levels of
testosterone in utero presumably promote the develop-
ment of robust facial architecture. These effects cer-
tainly contribute to the stamping of the adult male face
but cannot be simply equated with total sexual dimor-
phism.
High circulating testosterone levels, in turn, seem to
determine a facial elongation pattern, maybe more re-
lated to extended growth. In addition, the pronounced
and medially almost fusing eyebrows are likewise ab-
sent in the low 2D:4D shapes. It has been proposed that
the eyebrows receive the highest rank in the order of
magnitude of perceived gender information carried by
parts of the face20. Evidently, such facial characteristics
define the scope of a different domain than the one per-
taining to prenatal testosterone. However, there is more
research needed to yield the appropriate fine-grained
picture of the hormonal operations at different ages (in-
cluding growth) composing the final form of the adult
male and female human face. The apparent limitation
of the present study is its sample size (N=46), which re-
stricts the analysis to be exploratory and permits state-
ments only about trends, though there is a picture
emerging about the morphological effects of prenatal,
pubertal and adult testosterone (and their respective
perceptions) that seems worth pursuing in future stud-
ies. It remains to be explained why women seem to find
the high-testosterone traits attractive.
In conclusion, the contributions of the present study
are twofold. First, although numerous studies of adult
male facial appearance have assumed an association
with hormone levels, there was hitherto no data avail-
able that has actually visualized, detailed, and com-
pared variation in facial shape in relation to circulating
and prenatal testosterone. Second, our visualisations
support the claims of Neave et al.1that some of the
adult male facial characteristics (known to be perceived
as dominant and masculine) might be determined very
early in ontogeny.
Acknowledgements
We are indebted to Philipp Gunz for rendering assis-
tance with the analysis. We are also grateful to John T.
Manning, Dennis E. Slice, and two anonymous reviewers,
who made valuable comments on earlier versions of the
manuscript. Support by the Austrian Ministry of Cul-
ture, Science and Education, and the Austrian Council
for Science and Technology grant P200.093/I–VI/2004 to
Horst Seidler.
REFERENCES
1. NEAVE, N., S. LAING, B. FINK, J. T. MANNING, Proc. R. Soc.
B., 270 (2003) 2167. — 2. ENLOW, D. H.: Essential of facial growth. (W.
B. Saunders, Philadelphia, 1996). — 3. FARKAS, L. G.: Anthropometry
or the head and face in medicine. (Elsevier, New York, 1981). — 4. FINK,
B., I. S. Penton-Voak, Curr. Dir. Psychol. Sci., 11 (2002) 154. — 5. GRAM-
MER, K., B. FINK, A. P. MØLLER, R. THORNHILL, Biol. Rev., 78 (2003)
385. — 6. PENTON-VOAK, I. S., J. Y. CHEN, Evol. Hum. Behav., 25
(2004) 229. — 7. MANNING, J. T.: Digit ratio: a pointer to fertility, be-
haviour, and health. (Rutgers University Press, New Brunswick, 2002).
— 8. LUTCHMAYA, S., S. BARON-COHEN, P. RAGGATT, R. KNICK-
MEYER, J. T. MANNING, Early Hum. Dev., 77 (2004) 23. — 9. FINK,
B., K. GRAMMER, P. MITTEROECKER, P. GUNZ, K. SCHAEFER, F.
L. BOOKSTEIN, J. T. MANNING, Proc. R. Soc. B., 272 (2005) 1995. —
10. BOOKSTEIN, F. L.: Morphometric tools for landmark data: geome-
try and biology. (Cambridge University Press, Cambridge, 1991). — 11.
KNUSSMANN, R.: Handbuch der vergleichenden Biologie des Menschen
K. Schaefer et al.: Facial Shape and Testosterone, Coll. Antropol. 29 (2005) 2: 415–419
418

(Fischer, Stuttgart, 1988). — 12. GRAMMER, K., B. FINK, A. JUETTE,
G. RONZAL, R. THORNHILL, Female faces and bodies: n-dimensional
feature space and attractiveness. In: RHODES, G., L. ZEBROBWITZ
(Eds.): Advances in visual cognition: Volume I: Facial attractiveness –
evolutionary, cognitive, cultural and motivational perspectives. (Ablex,
Westport, CT, 2001). — 13. MARCUS, L. F., M. CORTI, A. LOY, G. J. P.
NAYLOR, D. E. SLICE: Advances in morphometrics. (Plenum Press,
New York, 1996). — 14. ROHLF, F. J., D. E. SLICE, Syst. Zool., 39 (1990)
40. — 15. DRYDEN, I. L, K. V. MARDIA: Statistical shape analysis.
(Wiley, Chichester, UK, 1998). — 16. BOOKSTEIN, F. L., Acta. Zool.
Acad. Sci. H., 44 (1998) 7. — 17. GOOD, P. I.: Permutation tests: a prac-
tical guide to resampling methods for testing hypotheses. (Springer, New
York, 2000). — 18. MUELLER, U., A. MAZUR, Behav. Ecol., 8 (1997)
569. — 19. SWADDLE, J. P., G. W. REIERSON, Proc. R. Soc. B., 269
(2002) 2285. — 20. BROWN, E., D. I. PERRETT, Perception, 22 (1993)
829.
K. Schaefer et al.: Facial Shape and Testosterone, Coll. Antropol. 29 (2005) 2: 415–419
419
K. Schaefer
Department of Anthropology, University of Vienna, Althanstrasse 14, 1090 Vienna, Austria
e-mail: katrin.schaefer@univie.ac.at
VIZUALIZACIJA ODNOSA OBLIKA LICA PREMA OMJERU VELI^INE DRUGOG I ^ETVRTOG
PRSTA RUKE I RAZINI TESTOSTERONA
SA@ETAK
Pretpostavlja se da spolni steroidi moderiraju razlike izme|u mu{kih i `enskih obilje`ja lica. Dosada{nja istra-
`ivanja su pokazala ve}u sklonost `ena za arhitekturu mu{kih lica razvijenu pod utjecajem testosterona jer ona
indicira mu`evnost, dominaciju i dru{tveni status. Studija koja je nedavno provedena pokazala je da se spolni dimor-
fizam lica ne razvija samo u vrijeme puberteta ve} mo`e biti uspostavljen puno ranije u ontogenetskom razvoju.
Me|utim, stvarni uzrok kao i vrijeme nastanka varijacija oblika lica koje su pod utjecajem spolnih steroida jo{ uvijek
su samo spekulativni. U svrhu istra`ivanja diferencijalnog utjecaja androgena na do`ivljaj mu`evnosti lica, ova studija
koristi podatke Neave i suradnika1koji su prikupili uzorke testosterona sline ali su i mjerili omjer duljine drugog i
~etvrtog prsta ruke (2D:4D) koji odra`ava u~inke testosterona u prenatalno doba. Geometrijsko-morfometrijskim me-
todama napravljena je regresija mu{kog oblika lica prema omjeru 2D:4D i razini cirkuliraju}eg testosterona. Re-
zultati su pokazali da rano djelovanje androgena (putem 2D:4D omjera) ima suprotne u~inke na gornji u odnosu na
donji dio lica. Tako nizak 2D:4D omjer odgovara relativno robusnom i upadljivom donjem dijelu lica dok izgleda da
cirkuliraju}i testosteron uzrokuje prili~no ravnomjernu elongaciju lica. Izra`ene i medijalno povezane obrve nisu
pokazale odnos s omjerom 2D:4D ve} su povezane samo s pove}anim vrijednostima testosterona u slini. Ovi pre-
liminarni rezultati sugeriraju da vrijednosti testosterona u prenatalno doba i u vrijeme puberteta imaju razli~ite
u~inke na formiranje mu{kih obilje`ja lica. Ovu ~injenicu trebalo bi uzeti u obzir prilikom budu}ih istra`ivanja sklo-
nosti `ena prema izgledu mu{kih lica.