1. Introduction

Recent research into the components of human male facial attractiveness has not produced agreement on the role of facial masculinity. On the one hand, a number of researchers have found female preference for masculine facial structure (Cunningham et al., 2004, Grammer & Thornhill, 1994, Johnston et al., 2001, Scheib et al., 1999). Others have found either no preference or a preference for feminine faces (Little & Hancock, 2002, Penton-Voak, et al., 1999, Perrett et al., 1998, Swaddle & Reierson, 2002). Some of the discrepancy may be due to shifts in female preference for masculine faces over the menstrual cycle: Penton-Voak and Perrett (2000) revealed that women at high conception risk, but not low conception risk, preferred masculine faces. Penton-Voak et al. (2003) found that less physically attractive women showed reduced preference for masculine male faces, suggesting that preference may vary as a function of individual phenotypic quality. Another potential reason for inconsistent results could be that masculinity may be preferred only in context of short-term relationships: Little, Jones, Penton-Voak, Burt, and Perrett (2002) asked women to judge male faces for long- and short-term sexual relationships and found that women showed increased preference for masculinity when judging faces with a short-term relationship in mind. Condition-dependent choice for facial masculinity is also the focus of the present study, in which women's mating strategies were assessed using the Sociosexuality Index (Simpson & Gangestad, 1991).

Androgens promote cortical bone growth and bone turnover (e.g., Kung, 2003, Mayo et al., 2004). Experimental manipulation of steroid or testosterone levels affects craniofacial bone growth in laboratory animals (Noda et al., 1994, Verdonck et al., 1998). In humans, sexual dimorphism in craniofacial features becomes evident enough to distinguish between male and female faces after puberty (Scheuer & Black, 2000, p. 147). At puberty, the depth of the mandible increases in males, as well as the size of the vertical ramus (Knight, 1996). Changes in the vertical ramus are responsible for the square jaw that is characteristic of male faces. Mandible size has been shown to respond to low doses of testosterone when boys are administered testosterone to treat delayed puberty (Verdonck, Gaethofs, Carels, & de Zegher, 1999), suggesting direct testosterone influence. A second craniofacial trait likely to be caused by the presence of testosterone is the development of supraorbital ridges, which results in squarer and smaller eye sockets relative to skull size. These eye and mandible features are routinely used for sex determination in forensic analyses (see Knight, 1996). Measurements taken from frontal facial photographs have shown significant sex differences on eye and jaw dimensions (Gangestad & Thornhill, 2003). There is evidence that facial masculinity is correlated with salivary testosterone levels in men (Penton-Voak & Chen, 2004), despite the salivary measures in the research representing only a single point in time, whereas facial masculinity should represent the cumulative presence of circulating testosterone.

Facial masculinity research has employed a number of different methods of masculinity assessment. Often, sex differences in facial structure have been of more interest to the researchers than are facial features likely to be testosterone-influenced. This has led to masculinity being defined in part by lack of feminine features, such as larger lips, rather than only the presence of masculine features (e.g., Gangestad & Thornhill, 2003). Similarly, in most studies in which faces have been digitally manipulated, overall facial sexual dimorphism has been manipulated (e.g., Little et al., 2002, Penton-Voak et al., 2003, Penton-Voak, et al., 1999, Perrett et al., 1998). The present study represents an attempt to rely only on craniofacial measures of facial masculinity that are most likely to reflect the presence of testosterone, taken from unmanipulated facial photographs.

The physical attractiveness of masculine facial structure has received attention from researchers largely because the testosterone presumed to underlie it is known to be physiologically costly. Testosterone is an immunosuppressant (e.g., Al-Afaleq & Homeida, 1998, Folstad & Karter, 1992, Kanda et al., 1996). Therefore, men able to maintain high testosterone levels are advertising that they can offset its physiological costs and, hence, are displaying high genetic quality (Thornhill & Gangestad, 1999; and see Hamilton & Zuk, 1982, Zahavi, 1975). Evidence that facial masculinity displays genetic quality has been suggested in research showing that higher facial masculinity correlates with higher bilateral symmetry, which has been argued to be a component of genetic quality (Gangestad & Thornhill, 2003). Second, Rhodes, Chan, Zebrowitz, and Simmons (2003) found that women rated more masculine faces as appearing healthier.

In addition to a likely role as a cue to genetic quality, high testosterone levels have been linked with increased male mating effort and pair-bond dissolution (Booth & Dabbs, 1993, Dabbs & Morris, 1990, Udry, 1988). High facial masculinity has been shown to be associated with lower self-rated paternal orientation and lower relationship maintenance effort (Waynforth, 2002). Facial masculinity may thus present a problem for women when choosing mates: By selecting for high genetic quality based on testosterone display, women could lose potential benefits of male parental care. This conflict may underlie findings of preference for feminine faces, as well as shifts across the menstrual cycle, as women may primarily select for good genes in short-term sexual relationships or extra-pair copulations when at high conception risk (see Penton-Voak & Perrett, 2000). Given that women seeking short-term relationships do not face the trade-off between genetic quality (as evidenced by facial masculinity) and male parental investment, women seeking short-term sexual relationships should focus on facial masculinity to a greater degree than women seeking a long-term relationship.

2. Methods

Frontal facial photographs were taken of 47 Anglo and 17 Hispanic males, as well as 65 Anglo and 29 Hispanic females, aged 18–30 years. Participants were students enrolled in an introductory anthropology course at the University of New Mexico. Photographs were taken against a neutral background. Participants were instructed to push their hair back away from the face, maintain a neutral expression, and keep the mouth closed. Photographs were viewed on the digital camera's screen immediately after each shot and were repeated in cases where the eyes were closed or partly closed or the head was tilted. Thirty-three photographs of women and 19 photographs of men were later removed from the sample when further inspection revealed slight head tilting likely to affect masculinity measures, leaving 61 women and 45 men in the sample.

In the present study, two measures of mandible size were assessed: chin depth and ramus size. One measure of eye orbit size was assessed: eye height. Measurements (in pixels) were taken from facial photographs using Scion Image computer software. Chin depth was recorded as the distance from the mouth to the chin, divided by interpupillary distance. Eye height was measured as the distance from the top of the eyelid to the bottom. The average of the left and right eyes was then divided by interpupillary distance. Ramus size was calculated as the width of the face at its widest point below the eye minus the width of the face at the mouth, divided by interpupillary distance. Eye height and ramus size were subtracted from one, so that like chin depth, higher numbers represent higher facial masculinity.

Sixty undergraduate women at the University of Durham rated the photographs. The facial photographs were randomly paired, and 10 pairs of faces were presented to each female rater, producing a total of 600 pairs of photographs rated. The raters were asked to select the more attractive of the two faces in each pair. Because the 600 observations were not completely independent, but instead were repeated observations by 60 women, a multilevel modeling approach was used. A mixed-effects binary regression model was employed using Mixor statistical software (Hedeker & Gibbons, 1996).

Women's mating strategies were assessed using the sociosexuality inventory (SOI) developed by Simpson and Gangestad (1991). The SOI inventory includes questionnaire items addressing, via questionnaire items addressing sexual experience and whether a desire for multiple future sex partners is expressed desired sex partner number.

3. Results

As expected, there were significant sex differences in all three measures for the samples of 45 men and 61 women: Men had longer chins, and women had larger eyes and more triangular faces (chin depth, t=6.39, p=.0001; eye height, t=8.95, p=.0001; ramus size, t=3.39, p=.001). This is consistent with data collected on similar masculinity measures by Gangestad and Thornhill (2003), and with the assertion that the differences associate with sex differences in hormone levels.

Principal components analysis was used to create an overall masculinity variable from the three masculinity measures and to reduce the seven-item SOI index to a single variable. The masculinity variables were positively correlated with each other in the 45 male faces: for eye height and ramus size, r=.62, p=.0001; chin depth and ramus size, r=.36, p=.014; and chin depth and eye height, r=.07, p=.64. The principal components analysis revealed one component with an eigenvalue above 1 (1.75), which accounted for 58.2% of the variance. Correlations between the seven SOI items were all positive and significant beyond the p=.05 level, with the exception of one item asking the frequency of sexual fantasy outside of the participant's long-term relationship, which correlated negatively with three of the remaining six items. This item was removed from further analyses. A principal components analysis on the remaining six items revealed eigenvalues of 2.42 and 1.19 for the first and second principal components, respectively. The first principal component accounted for 40.3% of the variance. The meaning of the second principal component was unclear, although it appeared to be loaded more heavily on the sexual history items of the SOI. It was not used in further analyses.

Table 1 summarizes the results of the mixed-effects logistic regression analyses of SOI and women's preferences for facial masculinity. Women chose the more feminine face slightly more than 50% of the time for three of the four masculinity measures. Higher SOI was significantly associated with increased preference for the more masculine face in the pair for chin depth, ramus size, and the three traits together. The direction of the trend was as predicted for eye height, but did not reach statistical significance. The masculinity difference between the two stimuli was statistically controlled for in each logistic regression analysis. Table 1 also includes intraclass correlations to assess the degree of dependency in the data due to the clustering of the 600 observations among 60 women. The intraclass correlations were modest when compared with those reported in other studies (see Hedeker, Gibbons, & Flay, 1994).

Table 1. Summary of the results of mixed-effects logistic regression models testing associations between sociosexuality and preference for the more masculine of two male faces presented simultaneously

	Masculinity measure	Percent (%) of participants choosing masculine image	p value for SOI as a predictor of preference for masculinity	Z value for SOI as a predictor of preference for masculinity	Intracluster correlation
	Chin size	47.5	.005	2.81	.058
	Ramus size	46.5	.040	2.05	.030
	Eye height	51.0	.104	1.63	.025
	All combined using PCA	46.8	.008	2.66	.013

Sixty women assessed facial attractiveness in 10 randomly selected pairs of faces, resulting in a sample size of 600 in each analysis.

4. Discussion

These findings on preference for facial masculinity are in accordance with previous work showing no overall preference for masculine male faces (Little & Hancock, 2002, Penton-Voak, et al., 1999, Perrett et al., 1998, Swaddle & Reierson, 2002). However, women who reported more sexual experience and interest in short-term sexual relationships were more likely to prefer more masculine male faces. Given the aforementioned research showing that high testosterone is associated with increased male mating effort, a possible explanation is that women's mating decisions must strike a balance between the choice for testosterone expression and the likelihood of future male desertion: When male parental investment is being sought, women may be able to decrease the risk of male desertion by choosing a mate with feminine facial features. If a short-term relationship is being sought, women should maximize their reproductive success by placing more emphasis on indicators of good genes, hence, their greater emphasis on facial masculinity.

The idea that women trade good genes for an increased chance of long-term paternal investment by relying on facial masculinity as a cue is not necessarily consistent with some of the existing research on testosterone and men's relationships. Gray et al. (2004) found that men in sexual relationships had lower salivary testosterone levels than did single men. If this is because testosterone production drops substantially in the context of long-term sexual relationships, women would not need to avoid men with large facial testosterone markers. Gray et al. were not able to establish the causal direction of the testosterone and relationship status link, but earlier longitudinal research suggested that it works in both directions: Mazur and Michalek (1998) measured testosterone levels in a cohort of military men. They showed that testosterone levels drop around the time of marriage, suggesting that the formation of a long-term sexual relationship influences testosterone levels. However, their results simultaneously showed the opposite causal pattern: that higher testosterone leads to less stable marriage.

From the perspective of male fitness maximization, the results of the present research suggest that the degree of facial masculinity development will impact the likelihood of being chosen by females interested in short-term sexual relationships. Given that change in masculine facial structure due to bone growth is likely to be most profound in the period following adrenarche, investment in masculine facial structure during adolescence may determine the future reproductive returns to seeking long- versus short-term sexual relationships.

In conclusion, the results presented here imply that, although the handicap principle may have played an important role in the evolution of masculine facial display, female requirements for male parental care are likely to result in an opposing force driving selection for feminine male faces. The fitness consequences of masculine facial structure and underlying hormonal causes need to be addressed in future research. In addition, studies of male physical attractiveness and facial masculinity have used widely differing methodologies and measures of facial masculinity. Here, three measures of masculinity that strongly differentiate the sexes and are well-established measures used in forensic analyses were analysed (see Knight, 1996). Second, traits most likely to indicate the presence of testosterone, rather that those that differentiate the sexes more generally were used.