- visible asymmetries are more important to attractiveness ratings than are non visible asymmetries F1,37=7.55 (p=.01)
- funnel plot analyses indicate a substantial publication bias in the literature
- studies with large sample sizes show a near zero relationship between attractiveness ratings and asymmetry F1,36=6.97 (p=.01)
Visible vs Not Visible Asymmetries
The distinction here is straightforward: if raters can see the measured asymmetry it is assumed that the asymmetry is visible. For example, if facial asymmetries are measured and faces are evaluated for attractiveness, then the study is categorized as visible. On the other hand, if the asymmetry is in the body and photos of faces are rated, the study is categorized as not visible. This visible/not visible distinction is relevant to why humans find symmetry attractive – for example in comparing a good genes interpretation of symmetry to a processing fluency interpretation. Van Dongen makes the point that the models underlying the good genes interpretation of the role of symmetry on attractiveness require that non-observable symmetry be significantly related to attractiveness ratings.
Since the value of a comprehensive data analysis is only as good as the data it uses, a check on the quality of the included sources is highly desirable. One such check is a funnel plot. In a funnel plot, each study is plotted for effect size and sample size. Since variation resulting from chance is more likely to be larger in studies with small sample sizes, a visual inspection of the plotted data points for the included studies should show a symmetric distribution around the typical effect size, with more variation in effect size expected for studies using smaller samples (hence the name, funnel plot). The published studies exploring the relationship between attractiveness and symmetry do not reveal the expected funnel-like symmetric shape. While a number of causes for this undesirable result are possible, publication bias is the most likely. A statistical technique used to minimize the effect of publication bias is the trim and fill method (especially useful in cases of publication bias). Van Dongen uses this technique to more accurately estimate the actual effect size of symmetry on beauty ratings.
A meta analysis that shows a decreasing effect size in studies with increasing sample sizes is another indicator of publication bias. The rationale for this conclusion is that there is a preference for publishing research that has found support for a particular hypothesis over research that reports finding no relationship. A manuscript that supports the null hypothesis is typically more interesting when the sample size is larger and thus gets published. Manuscripts that do not show a relationship are left ‘in the file drawer.’
Van Dongen’s overall study results, after accounting for bias, found that there was a significant effect of visible asymmetry on visual attractiveness ratings (r=.15 with a 95% confidence interval of 0.07-0.23). This degree of effect size is typically categorized as small/medium. What that means is that a person of average attractiveness (left side of the highlighted area below – the 50th percentile) who suddenly became more symmetric (by the typical amount of variation in symmetry found in human faces naturally) would now be rated more attractive than 62 percent of other people (the right side of he highlighted area below).
Another way to illustrate this degree of change is via a beauty rating scale. For ease of estimation, imagine an 8-point rating scale (from 0-8) that is normally distributed with 4 as the average, typical score. This degree of change would take the average person’s rating from a 4 to a 4.3. On this same scale, a person who is in all other respects rated average for attractiveness, but who had an exceptionally high degree of symmetry, would likely be rated a 5 rather than a 4 (this degree of symmetry is expected only in about 1/1000 people. It is important to note that for the illustrations above to hold, the changes in symmetry must be visible. Van Dongen’s meta analysis found no relationship between attractiveness ratings and the symmetry measures of features that are not visible to the person doing the rating.
One significant limitation in the symmetry/attractiveness literature is that the data primarily come from western, college student samples – limiting our ability to generalize these conclusions to other populations.
Van Dongen, S. (2011). Associations between asymmetry and human attractiveness: Possible direct effects of asymmetry and signatures of publication bias Annals of Human Biology, 38 (3), 317-323 DOI: 10.3109/03014460.2010.544676
The Louis Vuitton fall 2010 collection, presented by Marc Jacobs in Paris last month, showcased models with more size, shape, and age diversity than is characteristic in runway shows. Both MJ and LV deserve some acknowledgement for challenging beauty stereotypes. At the time of this writing, shots of the models on the runway can be seen here, and sans fards head shots (duplicated to the right) can be seen here. I suspect that the point of releasing the head shots image is to underscore that ordinary women are beautiful by showing how ordinary beautiful women can look. Comments on the images at BuzzFeed are mixed, but, mostly in the omg (‘oh my god’), wtf (‘what the f**k’), and ew (‘expression of disgust’) categories. Most of the models are not wearing makeup, though, for example, Elle Mcpherson is (bottom right). But, of course, there is more to this image’s impact than just a lack of make-up on the models’ faces. The deadpan expressions on most of the models and the ‘bad-hair’ contribute substantially to the super-ordinary appearance of these models (some of whom earn millions of dollars a year as models – and note: if make-up was all that mattered, supermodels could not command these kinds of salaries).
In addition to the subject dependent variables of make-up, hair, and expression; technical decisions about how to produce these images also contribute to their impact. Three stand-out:
- perspective distortion
- unflattering lighting
- unflattering exposure/contrast/levels or curves adjustments
Perspective distortion can result from the use of a wide angle lens and is illustrated here:
The top image illustrates the distorting effects of the use of a wide angle lens while the bottom image shows a distortion-free representation. This type of distortion appears visible in the models’ heads/faces and contributes to the “alienesque” appearance of some of the models.
Unflattering shadows exist on each face. It appears that models posed in front of a wall, with a window to their left front. This sort of lighting is not used when a photographer is attempting to take a flattering image.
The exposure/contrast/levels or curves adjustments vary with each portrait: most wash-out features/details in unflattering ways.
All-in-all, rather than being a ‘sans fards’ (without artifice) image, it appears that pre-photoshop, old-school photographic rules/techniques were intentionally ignored in order to make these supermodels appear super-ordinary.
Gunn et.al. (2009), comparing a number of aged/aging twinned and non-twinned subjects (some of the non-twins were of different ages), have concluded that the primary indicators of aging in women are:
- skin wrinkling
- hair graying
- lip height (measured from the “vermillion border on the philtral crest” (the high points of the upper lips spaced around the philtral groove [below the center of the nose]) to the lowest point on the lower lip – in this case, adjusted for face height due to the use of non-standard distances from face-to-camera in the making of the stimulus photos
These differences are visible in the composite photos below.
- thinning hair
- uneven skin tone/pigmented spotting
- more prominent nasolabial folds (the creases that run from the corners of the nose to the corners of the mouth – primarily resulting from changes in fat deposition associated with aging)
- possibly: receding hair
Interestingly, heritability analyses of this data indicate that signs of aging in skin are influenced equally by genetic and environmental differences; that lip height, hair graying and recession are primarily influenced by genetic factors; and that hair thinning was influenced primarily by environmental factors.
All-in-all, an interesting study and a solid contribution to the literature on aging with some implications for the psychology of beauty.
- Subjects are all caucasian/northern european
Gunn, D., Rexbye, H., Griffiths, C., Murray, P., Fereday, A., Catt, S., Tomlin, C., Strongitharm, B., Perrett, D., Catt, M., Mayes, A., Messenger, A., Green, M., van der Ouderaa, F., Vaupel, J., & Christensen, K. (2009). Why Some Women Look Young for Their Age PLoS ONE, 4 (12) DOI: 10.1371/journal.pone.0008021
BMI – the ratio of body mass to height, typically correlates well with ratings of body attractiveness. WHR – a direct comparison of waist and hip measurements – also correlates with attractiveness. Recent research that compares the relative strengths of the two ratios generally finds that variation in BMI accounts for a greater proportion of variation in attractiveness ratings than does variation in WHR. The implication is that, at least in contemporary industrial/technological societies, levels of body fat matter more than how that body fat is distributed. I found myself reflecting on these ratios in relation to women’s body attractiveness today, and wondered how WHR would influence ratings of body attractiveness if BMI was held constant? My guess was that WHR would be more strongly correlated with attractiveness ratings when controlling for BMI in this way. (I couldn’t recall a study that explored this possibility and I also could not find one in the literature – if you know of one please post a link or citation.) My rationale was that if subjects are matched for BMI, then WHR variation would likely result from variation in estrogen efficacy. My hypothesis was that, other things being equal, curviness resulting from estrogen efficacy would more strongly influence attractiveness ratings.
So far my thinking has been pretty predictable. Then I reflected on estrogens’ role in developing the sexually dimorphic features that are found attractive in women’s faces (Smith, et.al, 2006). That’s when I realized that, to date, comparisons of WHR and BMI are done on ratings of body attractiveness alone. This practice is sensible, since cognitively, evaluations of faces and bodies are separate processes. But, since estrogens significantly influence both facial attractiveness and body attractiveness, these two ratings should be related. [There is some support for this relationship (Thornhill & Grammer, 1999).]
These musings leave me wondering: might WHR be a better predictor of overall attractiveness than BMI in women?
Image of the 3rd century Bikini Girls mosaic from the Villa Romana in Sicily courtesy of Roundtheworld. Wikipedia Commons.
Law Smith, M., Perrett, D., Jones, B., Cornwell, R., Moore, F., Feinberg, D., Boothroyd, L., Durrani, S., Stirrat, M., Whiten, S., Pitman, R., & Hillier, S. (2006). Facial appearance is a cue to oestrogen levels in women Proceedings of the Royal Society B: Biological Sciences, 273 (1583), 135-140 DOI: 10.1098/rspb.2005.3296
Thornhill, R. (1999). The Body and Face of Woman One Ornament that Signals Quality? Evolution and Human Behavior, 20 (2), 105-120 DOI: 10.1016/S1090-5138(98)00044-0
Using digitally manipulated levels of sexual dimorphism in human male and female faces (like the ones to the right), Glassenberg et.al. (2009) found that, compared to heterosexual women, homosexual women preferred greater masculinization in female faces [Brown-Forsythe t(303.38) = -2.92, p<.01] while heterosexual women preferred greater masculinization in male faces [t(375) = 6.77, p<.001]. Compared to heterosexual males, homosexual males preferred masculinization in both male and female faces [t(520) = -7.42, p<.001 and t(520) = -6.72, p<.001 respectively]. Calculations based on sociosexual orientation were mostly non-significant, though relatively small, significant, positive correlations were found in heterosexual males between unrestricted SO and a preference for feminization in female faces [R(125) = .20, p<.05] while homosexual males showed a positive correlation between unrestricted SO and a preference for masculinized male faces [R(259) = .17, p<.001]. These specifics aside, all raters preferred feminized female faces to masculinized female faces.
- large sample
- rated stimuli consisted of 3 face composites to ensure recognizable individuality. There was no effort to match stimuli for attractiveness prior to manipulating sexual dimorphism, so an attractiveness x dimorphism interaction would be missed in this design.
This study suggests that homosexuals’ preferences are neither identical to nor mirror-images of heterosexuals’ preferences. This data also suggests that researchers should control for sexual orientation when conducting attractiveness studies in which sex/gender are relevant variables.
Glassenberg, A., Feinberg, D., Jones, B., Little, A., & DeBruine, L. (2009). Sex-Dimorphic Face Shape Preference in Heterosexual and Homosexual Men and Women Archives of Sexual Behavior DOI: 10.1007/s10508-009-9559-6
Egan & Cordan (2008) digitally altered the faces of 17-year-old girls (n=10) to look either younger (morphed to appear similar to the prototype of 10-year-old girls – top row) or older (similar to the prototype of 20-year-old women – bottom row). Additionally, some stimuli were altered by adding digital make-up (right column). The authors had forensic interests and were exploring the effect of alcohol consumption on judgments of age and attractiveness. As a result, they did not report the specific data on attractiveness ratings alone, but, did conclude that faces that appear younger are found more attractive. Raters consisted of an equal number of adult women and men between the ages of 18-70.
The faces were manipulated using proprietary software, Psychomorph. To my eye, the morphed images look good, though, there appear to be distortions in the ears of the “older” faces. Eye size, distance between eyes, lips, forehead height, hair, and clothes do not appear different (to my eye) between the “younger” and the “older” sample stimuli. Though there may be a sense of greater protrusion in the “older” forehead…. The primary apparent differences are a larger nose and longer lower face in the “older” version.
While we don’t have the specific data reported in this paper, the conclusions are consistent with what is generally asserted: looking young is attractive in human females. At least one contributing reason for this attraction to youthful appearance in female faces is the increased rate of development in male faces at puberty, relative to female faces. That is, men’s and women’s faces show the same growth spurt: but males show this growth more markedly. This variation results in larger noses, mandibles, and sinuses (along with brows and cheekbones) in men.
Since these areas are larger in men, larger features become masculine features. Since these facial features are smaller in women, smaller ones become feminine. Another way to conceptualize this: looking younger looks less masculine. To my way of thinking, this explains what might appear to be a disturbing preference in both men and women for female faces with some prepubescent structural characteristics.
Egan, V., & Cordan, G. (2009). Barely legal: Is attraction and estimated age of young female faces disrupted by alcohol use, make up, and the sex of the observer? British Journal of Psychology, 100 (2), 415-427 DOI: 10.1348/000712608X357858