A Comparison of Masculinity Facial Preference Among Naturally Cycling, Pregnant, Lactating, and Post-Menopausal Women

Women show cyclical shifts in preferences for physical male traits. Here we investigated how fertility status influences women’s facial masculinity preference in men by analyzing a large sample of heterosexual women (N = 3720). Women were regularly either cycling (in both low- and high-conception probability groups), lactating or were currently in a non-fertile state (pregnant or post-menopausal). Analyses simultaneously controlled for women’s age and sexual openness. Participants via two alternative forced choice questions judged attractiveness of masculinized and feminized men’s faces. After controlling for the effect of age and sociosexuality, regularly cycling and pregnant women showed a stronger preference for masculinity than lactating and post-menopausal women. However, there was no significant difference in masculinity preference between women in the low- and high-conception probability groups. Women’s sociosexuality showed a positive, but very weak association with men’s facial masculinity preference. We suggest that women’s overall, long-term hormonal state (cycling, post-menopausal) is a stronger predictor of preference for sexual dimorphism than changes in hormonal levels through the cycle.

On the other hand, Boothroyd et al. (2017) showed that intermediate, rather than high, levels of men's masculinity were associated with offspring survival, which does not support the idea & Urszula M. Marcinkowska ummarcinkowska@gmail.com that women prefer more masculine males in order to confer heritable immunity on their offspring. According to this view, women in reproductive age prefer average levels of masculinity (Scott, Pound, Stephen, Clark, & Penton-Voak, 2010;Stephen et al., 2012) which provides higher genetic benefits to their offspring (Foo, Simmons, & Rhodes, 2017;Lie, Rhodes, & Simmons, 2008).

Aims
In our study, we aimed to replicate findings on variation in masculinity preference among women of various age groups based on a new sample of women,enhance the existing pool of evidence onmenstrualcyclicalpreferenceshiftsonanew,large,anddiverse sample, and, most importantly, compare preferences between groups of women of varying fertility (cycling, lactating, pregnant, and menopausal). A significant addition that we made in comparison with previous studies was controlling for participant's age and sociosexuality.

Method Participants
Women were recruited via online forums, mailing lists, and via personal communication. Responses were collected through a web-based survey, as it has been shown that online and laboratory studies of variation in preference for sexual dimorphism produce comparable patterns of results (Welling et al., 2008). Entering the study was conditioned by participant's age (minimum age = 18 years old) and not using hormonal contraceptives, as hormonal contraception can influence women's preference (Roberts et al., 2014;Welling, Puts, Roberts, Little, & Burriss, 2012). A total of 3720 heterosexual women completed the survey. Sexual orientation was based on the Kinsey scale (Kinsey, Pomeroy, & Martin, 1948). Only participants scoring 2 or lower were included in the study (exclusively heterosexual, predominantly heterosexual only incidentally homosexual or predominantly heterosexual, but more than incidentally homosexual). Participants reported age (in years), their current hormonal status (regularly menstruating, pregnant, lactating, post-menopausal), and their average length of the menstrual cycle and days since the beginning of the last menstrual bleeding.

Procedure
Participants were presented with 20 slides (shown in a random order), and they selected via forced choice the more attractive of two stimuli pictures by answering the question''Which of the following faces is more sexually attractive?''The forced choice method is more appropriate for this kind of research compared with ratings of single pictures (Leivers, Simmons, & Rhodes, 2015). Each slide depicted two versions of the same facial picture modified to be more or less masculine. Individual preference for masculinity was calculated as the proportion of masculinized pictures being selected among the 20 pairs of pictures. This index varied from 0 (20 feminized pictures selected) to 1 (20 masculinized pictures selected).

Measures
In this study, a subset of base pictures from a previous study examining correlates of men's facial masculinity was used (Rantala et al., 2012). All pictures were taken using standard background and light conditions. Facial expression of the photographed person was neutral. All photographed men were Caucasian. Base pictures were transformed on a femininity/masculinity scale by using the linear difference between a composite (average) of 40 adult males and a composite of 40 adult females following established methods (Perrett et al., 1998). From each base picture, we created two stimuli pictures by adding or subtracting 50 percent of the difference between male and female composites to the base picture. What is crucial, these stimuli pictures within a pair differed only in the shape of the face and not in any other aspects (such as color, texture, symmetry), which can influence the choice (DeBruine et al., 2010). All manipulations were made with PsychoMorph program (Tiddeman, Burt, & Perrett, 2001) in a way consistent with earlier studies .

Sociosexuality
To assess attitudes toward sexual behavior, the Revised Socio-sexualOrientationInventory(SOI-R;(Penke&Asendorpf,2008); Cronbach's a = 0.73) was used. This is a nine-item scale which provides an overall measure of sociosexual orientation (e.g.,''How manydifferent partners have youhad sexual intercourse with on one and only one occasion?''1 = 0 partners, 9 = 20 partners and more) as well as three subdivisions: the Behavior subscale that measures the number of casual sex partners and the frequency of change in partners; the Attitude subscale that measures the participant's disposition toward short-term sexual encounters; and the Desire subscale that measures the frequency of sexual fantasies or arousal in relation to potential mates with whom the individual is currently not in a committed relationship. A high SOI-R score indicates a propensity to engage in more short-term sexual relationships. The mean SOI score in this study was M = 3.21 (SD = 1.62, absolute range, 1-9).

Fertility Groups
Participants were divided into five fertility groups: (1) naturally menstruating women who were in the high-conception probability phase of their menstrual cycle, (2) naturally menstruating women who were in the low-conception probability phase, (3) pregnant, (4) lactating, and (5) post-menopausal women (Table 1).Withinthe naturally menstruating womengroup,based onthereversecountofdays(deductingdayofthecyclewhencompleting the survey from stated average cycle length), those who were in 19-14 days prior to the next menses were defined as the high-conception probability group, and all other participants were defined as low-conception probability (Roney, Simmons, & Gray, 2011).

Statistical Analyses
Initially, analysis of covariance (ANCOVA) was used to test differencesofmasculinitypreferenceamongthefivestudygroups. SOI and age of participants were treated as covariates. We also tested the assumption that there was no interaction between categorical and continuous predictors with homogeneity-of-slopes ANCOVA. The homogeneity-of-slopes model yielded nonsignificant results (all p[.14) implying that the homogeneity of regression slopes assumption was met. For the purpose of preliminary analysis, women were clustered into two groups-overall high fertility (high-and low-conception probability groups) and overalllowfertility(lactating,pregnantand post-menopausal).Aslactating women resume ovulating on average 32 weeks after the labor, we assumed that fertility in the lactating group was significantly lower than fertility in regularly menstruating group (Howie,McNeilly,Houston,Cook,&Boyle,1982;Labbok,2015). Levene's test of homogeneity of samples showed, however, that the samples of participants involved here were unequal, F(4, 3715) = 4.77, p = .001, which prevents the use of ANCOVA (Levene, 1960). Various types of transformation of masculinity scores did notyieldbetterresults. We thereforefollowed recommendations of Quade (1967) and regressed the dependent variable (masculinity score) against covariates (SOI and age). Residuals from regression (dependent variable) were finally analyzed with ANOVA where five fertility groups were treated as categorical predictor. Fisher post hoc test was used for pair-wise comparison between means following Quade (1967). Effect sizes (partial g 2 ) were calculated according to Huberty (2002), where values around 0.01 are considered small, 0.04 moderate, and 0.10 a large effect.

Results
Mean preference for each fertility group was computed (Table 2). An ANOVA comparing mean preference among the five groups showed statistically significant differences, F(4, 3715) = 5.69, p\.001, albeit the effect size was low (partial g 2 = 0.006). Planned comparisons showed that the overall high-fertility group (i.e., women in high-and low-conception probability phases) had significantly stronger preferences for masculinity compared with the overall low-fertility group (pregnant, lactating, and menopausal women), F(1, 3715) = 4.26, p\.05.
Masculinity preferences differed significantly among the fivefertilitygroups (Fig. 1).Fisherposthoctestshowedthatwomen who were in both the fertile and non-fertile phases of the menstrual cycle and pregnant women had a significantly higher masculinity preference score than the post-menopausal women (all p\.0001). Lactating women showed no significantly different preferences than all other fertility groups of women. Other differences were not statistically significant (all p[.07). Both SOI score and age positively correlated with masculinity preference (Spearman r = .17 and .23, both p\.0001, respectively). When both SOI and masculinity preferences were controlled for age, correlation between these variables was low, albeit statistically significant (Spearman r = .14, p\.0001).

Discussion
Consistent with previous studies, our results showed that current fertility status of women influenced their preference for sexualdimorphisminmen'sfaces,althoughtheeffectsizeswerelow. We found that masculinity preference of women who were naturally cycling at the time of completing the survey was stronger than that of women whose actual fertility status would prevent conceiving (post-menopausal). This finding follows a general assumption that higher probability of conceiving is related to higher preference for masculinity, because this allows women to obtain good genes for their offspring (Feinberg et al., 2006;Gangestad & Thornhill, 2008;Penton-Voak & Perrett, 2000;Penton-Voak et al., 1999;however, see Havlíček, Cobey, Barrett, Klapilova, & Roberts, 2015).
On the other hand, preferences for masculinity among fertile women were low (values about 0.5) which provides support for preferences of average levels of masculinity (Boothroyd et al., 2017;Scott et al., 2010;Stephen et al., 2012). There are at least three explanations for this finding. First, average, but not masculine male faces are cues of major histocompatibility complex (MHC) heterozygosity that is linked to immunocompetence (Lie & Simmons, 2008) and better perceived health (Foo et al., 2017). Second,it maybe that our measure of masculinitypreference was too narrow, because men's masculinity is expressed not only in faces, but in a number of additional physical features like voice (Cartei, Bond, & Reby, 2014;Feinberg et al., 2008), putative male pheromones (Saxton, Lyndon, Little, & Roberts, 2008), and bodies (Little, Jones, & Burriss, 2007). Little et al. (2011) showed, however, significant consistency in women's preferences for masculinity across all mentioned stimulus types, which suggests that this explanation is less likely. Third, men's masculinity is associated with sexual aggression toward women (Lackie & de Man, 1997), lower trustworthiness (Smith et al., 2009), unrestricted sociosexuality (Boothroyd et al., 2008), and reduced paternal investment . Thus, possible benefits from mating with masculine men are traded against costs associated with men's masculinity which may result in lower preference of masculine men's faces by women.
In our study, we did not find a difference in preferences between women in high-conception and low-conception probability phases, as found in a few previous studies (Harris, 2011(Harris, , 2013Peters, Simmons, & Rhodes, 2009). It has been proposed that there are pronounced differences in women's preferences depending on conception probability, based on their hormonal state (for a meta-analysis, see Gildersleeve, Haselton, & Fales, 2014). Preference for masculinity and good genes was proposed to be highest around ovulation when the conception is most likely (Gangestad & Thornhill, 2008). In contrast, a preference for increased paternal investment would increase during the low-conception period, and especially during the luteal phase, when the hormonal profile somewhat resembles a beginning of pregnancy (Jones et al., 2005). We did not find a support for the lowered masculinity preference among women in their low-conception probability phase. We ascertain that the backwards cycle day counting method used in this sample was not precise enough to actuallyallowus toclassify correctlywomen into low-andhighconception phase and more objective indicators of cycle status are required (Gangestad et al., 2016). It has also been suggested that participants can recall the dates of the menses onset in a faulty manner(Lukaszewski&Roney, 2009) and,mostimportantly, hormonal levels in menstrual cycles vary among women and among cyclesofasinglewoman (Jasienska&Jasienski,2008).Thismeans that when the cycle day counting method is used, some women classified as being in''non-fertile''cycle phase may have higher levels of ovarian steroid hormones than women classified as being in''fertile''cycle phase.
In addition, regularly cycling women often have cycles that are unovulatory or cycles with low progesterone levels (Ellison, 2003;Jasienska,2013); thus,these women,in fact,should be classified as''non-fertile''regardless of cycle phase. Due to the large number of participants and being an Internet-based study, we were unable to use methods for detecting ovulation or to measure levels of hormones. Differences between high-and low-fertility phases can be very subtle and could be better tracked by a within-participant design, rather than a between-participant one. It is close tounachievable,however,duetomethodologicalobstaclestofacilitate a within-subject design in such large data samples. We believe that between-subject, grand scale studies complement withinsubject smaller sample studies. Hence, the lack of cyclical shift in masculinity preference in our data does not exclude theory that there is a difference between women of varying fertility status.
As pregnancy and menopause signal a long-term state of nonfertility, we could expect that from an evolutionary point of view women's preference should be directed to resources and parenting skills, rather than good genes (Cobey, Little, & Roberts, 2015;Little et al., 2010). Preferences of post-menopausal women for more feminine men's faces could be caused by a shift from mating-oriented behavior to family-oriented behavior (Hawkes, O'Connell, Jones, Alvarez, & Charnov, 1998). More feminine men have apparently lower testosterone levels (Schaefer, Fink, Mitteroecker, Neave, & Bookstein, 2005) that can be associated with higher involvement in paternal care (Muller et al., 2009). Itmay be that after menopause, a woman's preference may change toward better parental and/or grandparental care (Rantala, Polkki, & Rantala, 2010). Cobey et al. (2015) found that postpartum women (up to 12 weeks after birth) showed lower masculinity preference than pregnantwomen.Similarly,wefoundthatpregnantwomenshowed strongerpreferenceformasculinitycomparedwithlactatingwomen albeit the difference was short of statistical significance (Fisher post hoc test, p = .077), perhaps because our sample consisted of exclusively breastfeeding women. This difference, albeit not statistically significant, could be explained by hormonal changes associated with transition to parenthood, during which baseline testosterone level is decreasing (Kuzawa, Gettler, Huang, & McDade, 2010). Indeed, Alder, Cook, Davidson, West, and Bancroft (1986) found that testosterone and androstenedione levels were significantly lower in lactating women who reported severe reduction in sexual interest. Such physiological change would be adaptive, because lowered attraction to men's facial cues associated with sexual attractiveness may enhance maternal behavior .
Notably, our results could stem from the reproductive ambition of participants (i.e., desire to become pregnant), which is positively correlated with preference for masculinity in men's faces (Watkins, 2012). It is possible that reproductive ambition would not change over the cycle but rather result from the reproductive history of a woman-hence, significant difference between cycling and not cycling women, and a lack of difference between high-and low-conception probability phases.
One possible confounding factor in our research could be men's age on facial stimuli, because the age itself changes and is related to preferences for partners as well. Buss and Schmitt (1993), in their classic paper on mating preferences, showed that women in 37 cultures preferred older men. The age of a pre-ferredmanwasonaverage3.5 yearsolderthantheageofawoman. We cannot exclude the possibility that preferences for masculinity were confounded by higher age differences between older, post-menopausal women and male facial stimuli. Older women, however, showed similar masculinity preferences as lactating women and both groups show similar androgen decline (Alder et al., 1986;Davison, Bell, Donath, Montalto, & Davis, 2005) supporting the idea that women's long-term hormonal changes influences mating preferences (Havlíček et al., 2015;Little et al., 2010). Future research can examine whether age differences between raters and facial stimuli influence mating preferences.
Several studies showed that women's sociosexuality was positively associated with preferences for masculine men's faces (e.g., Burt et al., 2007;Waynforth et al., 2005). The present study confirmed this relationship, but the correlation was very weak. Most possibly, these associations are influenced by several other variables that were not controlled in this study. More attractivewomenshow,forexample,highersociosexuality (Clark, 2004) and stronger preferences formasculinemale faces (Little, Burt,Penton-Voak,&Perrett,2001).Unpartneredwomenshowed higher sociosexuality scores that significantly correlated with preference for men's facial masculinity compared with partnered women (Sacco et al., 2012). Some personality traits, such as extraversion, correlate with women's sociosexuality (Wright & Reise, 1997) and, in turn, extraversion was found to correlate with women's preferences for masculinity in men's faces (Welling, DeBruine, ). Future research on sociosexuality and masculinity preferences should take more factors influencing masculinity preference into account before firm conclusions can be made.

Conclusions
To conclude, we found an effect of overall fertility status on facial sexual dimorphism preference in women. It appears that the overall lowered fertility state caused by menopause affects the masculinity preference. Preferences for masculinity in naturally cycling women were, however, low, which can be explained by preferences for average, rather than masculine faces thatprovide health benefits to children. We did not find differences in masculinity preference depending on varying conception probabilitythroughoutthe menstrualcycle though(basedonthe backward counting days method). Women's sociosexuality showed positive, but very weak influence on preferences for masculine men's faces. Based on our results, we suggest that women's longtermhormonalstateisastrongerpredictorofpreferenceforsexual dimorphismthanchangesinhormonallevelsthroughoutthecycle.