Facial appearance is a cue to oestrogen levels in women
2. Material and methods
Participants were 59 white women from the student undergraduate population at the University of St Andrews (age, M=20.4, s.d.=1.5, range 18–24). No participants were currently using the contraceptive pill or had been in last 90 days. All received monetary payment for participation.
Participants were photographed each time they came to the laboratory, weekly for four to six weeks. Participants were photographed in a neutral expression, under standard conditions with diffuse flash lighting from two lateral flashguns. Images were captured on a digital camera at a resolution of 1200×1000 pixels in uncompressed TIFF format using 24 bit RGB encoding. No restrictions were made for make-up use during photography, however the use of make-up was recorded in self-reports. Consequently, 32 participants were not wearing any make-up when photographed and 27 were wearing make-up. The first photograph taken (week 1) was used for ratings if the participant had either always worn make-up (n=27) or always not worn make-up (n=14) in all the photographs. If there was a combination of no make-up and make-up photographs (n=18), the first photograph with no make-up was used. For presentation to raters, the faces were aligned on interpupillary distance and masked around the face line, so cues to hair and clothing were reduced.
Average faces were created in order to visualize the differences in facial appearance between women with high and low reproductive hormones. Composites were constructed from the faces of the females with the highest 10 and lowest 10 oestrogen using the methods outlined in Benson & Perrett (1993) and Tiddeman et al. (2001). See figure 1 for composite faces. Separate composites were not constructed for progesterone levels, because they were highly intercorrelated with oestrogen levels (see §3b); therefore, the composites would have contained the majority of the same faces. Composites were created from oestrogen rather than progesterone as the latter was not as strongly related to the face ratings (see §3b).
(c) Hormone measurement
Participants were instructed to provide their sample of urine from the midstream of the first urination of the morning of each day of testing. Participants collected samples once a week for four to six weeks, in order to cover all stages of the menstrual cycle. All samples were stored at −20 °C until assays were performed.
The assays involved a direct competitive ELISA 96-well plate system to assess oestrone-3-glucuronide (E1G) and pregnanediol-3-glucuronide (P3G) (major metabolites of oestradiol and progesterone, respectively). Urine samples, diluted in assay buffer, were incubated with labelled antigen (E1G or P3G conjugated to horseradish peroxidase) in the presence of rabbit anti-steroid antibody (anti-P3G antibody (RAB F 27/7/87) or anti-E1G antibody (RAB 1), respectively). Bound and free antigens were separated using solid-phase goat anti-rabbit IgG. The plates were washed and bound antigen was detected by incubation with the substrate o-phenylenediamine and the developed reaction was detected using a plate reader at 492 nm. For full methods, see Joseph-Horne et al. (2002). The intra-assay variation for both was less than 10%. Hormone level results were expressed as steroid : creatinine ratio. Assays were not available for five subjects' late follicular oestrogen and two subjects' luteal progesterone.
Menstrual cycle information was collected via self-report (diary data). To determine day of menstruation and length of menstrual cycle, participants reported the number of days since the onset of their last period of menstrual bleeding and their average menstrual cycle length. Date of onset of period following study completion was also collected via email. Cycle day was calculated by the backwards counting method, previously used by Jones et al. (2005). The levels of oestrogen in the late follicular stage of the menstrual cycle (14–21 days before next period) were used for comparison as this is the stage at which females are most likely to conceive. This stage may show greatest variation in fertility and thus be most likely to show the greatest associations with physical condition and attractiveness. Previous research reported heightened attractiveness at the follicular (fertile) stage of the menstrual cycle (Roberts et al. 2004). For progesterone, an average of the luteal (non-fertile) stage was analysed (13–1 day before next period), as progesterone levels are very low until ovulation and then rise until onset of menses. Previous research with progesterone and WHR has used average luteal levels (Jasieńska et al. 2004).
Participants completed the following questionnaire on menstrual status and make-up use at each testing session.
As best you can recollect, when was the date of the beginning of your last menstrual period—i.e. the first day of bleeding?
When do you expect your next menstrual period to begin?
Is your menstrual cycle fairly regular? Yes/No
Normally, how long is your menstrual cycle? (i.e. 28 days, 30 days)
Are you currently wearing any make-up? Yes/No
(e) Face ratings
Participants for the ratings task were 15 female and 14 male students from the University of St Andrews (age, M=20.1, s.d.=2.6, range 18–25). All received payment for their participation. All participants rated the masked faces for femininity, attractiveness and health. The masked faces were rated individually on a 7-point scale from 1=not feminine to 7=very feminine. Faces were presented in random order. This procedure was repeated rating the original faces for attractiveness (1=not attractive to 7=very attractive) and apparent health (1=not healthy to 7=very healthy). Blocks for the different ratings were presented in random order. The task was self-paced.
The composite faces were rated in a forced-choice paradigm by 11 female and 10 male students from the University of St Andrews (age, M=22.3, s.d.=1.6, range 19–25). All participants rated the pair of composite faces for which was more attractive along the 8-point preference scale; much more attractive (left image), more attractive (left image), slightly more attractive (left image), guess (left image), guess (right image), slightly more attractive (right image), more attractive (right image), much more attractive (right image). The composites were rated in the same way for femininity and health.
Progesterone and oestrogen metabolite levels and each of the three facial ratings were normally distributed (Kolmogorov–Smirnov: progesterone, p=0.83; oestrogen, p=0.09; femininity, p=0.58; attractiveness, p=0.28; health, p=0.63; all zs<1.2). Therefore, parametric statistics were used in subsequent analyses. Ratings of faces were highly consistent across raters (all Cronbach's α>0.9).
(a) Controlling for potential confounds
(i) Make-up use
To determine if women who always choose to wear make-up differ from those that choose not to wear make-up, independent samples t-tests were used to compare hormone levels and age. There was no significant difference between make-up wearers and non-make-up wearers in age (M=20.5, s.d.=1.2; M=20.3, s.d.=1.7; t=−0.64, d.f.=57, p=0.53), oestrogen levels (M=13.3, s.d.=6.5; M=13.1, s.d.=7.9; t=−0.14, d.f.=52, p=0.89) or progesterone levels (M=0.27, s.d.=0.10; M=0.25, s.d.=0.13; t=−0.52, d.f.=55, p=0.60).
Faces wearing make-up were rated as significantly more feminine (M=4.3, s.d.=1.0), attractive (M=3.40, s.d.=0.73) and healthy (M=4.36, s.d.=0.68) than those not wearing make-up (M=3.5, s.d.=1.1; M=3.01, s.d.=0.92, M=3.99, s.d.=1.12), t=−2.84, p=0.006; t=−2.34, p=0.023; t=−2.13, p=0.038; respectively, all d.f.=57.
There was no significant difference in day of cycle when the rated photograph was taken (make-up, M=18.7, s.d.=5.7; non-make-up, M=14.9, s.d.=8.7) t=−1.30, d.f.=57, p=0.21. Therefore, rating differences between make-up and no make-up images cannot be due to cyclic change in attractiveness (see Roberts et al. 2004) as a result of make-up use being potentially biased to a particular part of the cycle.
(ii) Effect of age
As there was no difference in age or hormones between the make-up and non-make-up wearers, the following correlations between hormones and age were conducted with the total sample, using Pearson's product moment correlation.
Age did not correlate with either of the hormone levels (oestrogen, r=0.11, p=0.44, n=54; progesterone, r=0.18, p=0.21, n=57), or any of the face ratings (femininity, r=0.11, p=0.40; attractiveness, r=0.09, p=0.51; health, r=0.19, p=0.16; all n=59). Therefore, age was not controlled for in any of the following analyses.
(b) Hormone levels and facial attributions
As the use of make-up influenced attributions, the following analyses investigating the hormone appearance relationship were conducted for make-up and non-make-up wearers separately.
For those wearing no make-up, late follicular oestrogen levels were significantly positively correlated with femininity (r=0.48, p=0.007, n=30), attractiveness (r=0.48, p=0.007, n=30) and health ratings (r=0.52, p=0.003, n=30). For those wearing make-up, however, oestrogen levels were not related to femininity (r=0.003, p=0.99, n=24), attractiveness (r=−0.08, p=0.71, n=24) or health ratings (r=0.07, p=0.74, n=24).
For females wearing no make-up, luteal progesterone levels were not related to femininity (r=0.28, p=0.13, n=30). There was a trend for progesterone to positively correlate with attractiveness (r=0.33, p=0.075, n=30) and health ratings (r=0.35, p=0.055, n=30). For those wearing make-up, however, progesterone levels were not related to femininity (r=0.09, p=0.66, n=27), attractiveness (r=0.04, p=0.83, n=27) or health ratings (r=0.17, p=0.41, n=27).
As the hormone levels did not relate to attributions for those wearing make-up, the following further analysis of hormone appearance relationships was restricted to the non-make-up group.
The three face ratings are highly interrelated (attractiveness and femininity, r=0.84; attractiveness and health, r=0.81; femininity and health, r=0.60; all p<0.001, n=31). The three face ratings were entered into a principal component analysis (PCA). One factor with eigenvalue greater than 1 was extracted (eigenvalue=2.51, accounting for 83.69% of the variance) on which all face ratings loaded (femininity, r=0.89; attractiveness, r=0.97; health, r=0.88). This factor was interpreted as a general ‘quality’ factor. Oestrogen was significantly positively correlated with the quality factor (r=0.54, p=0.002, n=30) and there was a trend for progesterone to positively correlate with the quality factor (r=0.35, p=0.058, n=30). As progesterone is highly correlated with oestrogen (r=0.67, p<0.001), a linear regression was performed to determine the extent to which both hormones were independently related to the quality factor. A significant regression model was produced for predicting quality rating (adj r=0.28, p=0.017) with only oestrogen as a significant predictor (β=0.51, p=0.035). Progesterone was a non-significant predictor (β=0.03, p=0.89).
(i) Effect of sex of rater
To determine if there was any effect of sex of the rater on the relationship between oestrogen levels and the face ratings, the faces were divided into three equal groups (high, mid and low hormone levels) for a mixed ANOVA, with sex of rater as a between-subjects factor (two levels), oestrogen level as the within-subjects factor (three levels) and face ratings as the dependent variable.
There was a highly significant main effect of oestrogen level with the faces corresponding to high oestrogen levels receiving higher ratings (all F2,50>38.98, p<0.001). There was no main effect of sex of rater on any of the ratings (all F1,25<0.63, p>0.44). There were no interactions between rater sex and oestrogen level (all F2,50<0.93, p>0.40). Thus, rater sex did not affect ratings or qualify the relationship between oestrogen and face ratings.
(ii) Composite faces
The high oestrogen face was rated as much more feminine, attractive and healthy than the low reproductive hormone face (all t>6.31, p<0.001, d.f.=20) using a one-sample t-test on strength of preference. These results cannot be due to a mediating effect of age on facial appearance and hormones as there was no significant difference in age of the composites (low, M=20.8, s.d.=1.5; high, M=20.4, s.d.=1.8; t=0.54, p=0.60, d.f.=18). All face ratings were highly consistent (α>0.9).
Although many accounts of facial attractiveness propose that femininity in women's faces indicates high levels of oestrogen, there is little empirical evidence in support of this assumption. Here, we used assays for urinary metabolites of oestrogen (oestrone-3-glucuronide, E1G) and progesterone (pregnanediol-3-glucuronide, P3G) to investigate the relationship between circulating gonadal hormones and ratings of the femininity, attractiveness and apparent health of women's faces. Positive correlations were observed between late follicular oestrogen and ratings of femininity, attractiveness and health. Positive correlations of luteal progesterone and health and attractiveness ratings were marginally significant. Ratings of facial attributions did not relate to hormone levels for women wearing make-up when photographed. There was no effect of sex of rater on the relationships between oestrogen and ratings of facial appearance. These findings demonstrate that female facial appearance holds detectable cues to reproductive health that are considered attractive by other people.
This work was supported by BBSRC, MRC and Unilever Research. The authors thank Lesley Ferrier for recruiting and scheduling of participants; Alexandra Boyden, William Calderhead, Carolyn Cheetham, Catherine Dutton, Fiona Elder, Suzanne Hall, Jennifer Hardingham, Laura Harper, Laura Johnson, Jamie Lawson, Gillian McBride, Jennifer McChesney, Anne-Marie Morgan, Joanna Souter and Katherine Timmins for assistance with data collection. We also thank the Reproductive Medicine Laboratory Staff of the University of Edinburgh and Martha Urquhart for assistance in hormone analysis.
Present address: School of Psychology, University of Aberdeen, Aberdeen AB24 3FX, UK.
Present address: School of Psychology, Durham University, Durham DH1 3LE, UK.
Present address: School of Psychology, University of Southampton, Southampton SO17 1BJ, UK.