Too sexy for my smile

CBC News recently reported on a study published in the journal Emotion that attempted to determine how body language influences perceived sexual attractiveness. I take issue with some of the methods and interpretations. You can read the actual paper here, and the CBC article here, but the gist of it is as follows:

• A large sample of men and women were shown photographs of members of the opposite sex and asked to rate their sexual attractiveness.
  
• Each photo depicted a ‘model’ displaying one of four emotions – happiness, pride, shame, or neutral.
  
• In the first study, all participants were asked to rate a single photograph. All male participants rated the same female model in one of the four possible poses. Likewise, all female participants rated the same male model in one of the four poses.

• In a second study, three large groups of participants rated a bunch of photographs that were viewed online. Again the photographs displayed a member of the opposite sex expressing one of the four emotions. This time, however, the photographs (over 400 of them) were obtained online (e.g. from Google Images) and sorted into their respective categories (2 genders • 4 emotions = 8 categories) by trained assistants according to published guidelines. So, unlike the first study, each category here contained pictures of many models, and different models were used to depict each emotion.
  

• The general result that held across both studies was that males expressing happiness were rated the least attractive and males expressing pride were the most attractive. The trend was essentially reversed for female models, such that happy females were rated the most attractive whereas females expressing pride were among the least attractive.
  
• There were other interesting results and many details I have left out for the sake of brevity. Check out the original paper for more information.
  

So what are the shortcomings of this study? My problem with the first study (which in fairness the authors do acknowledge) is that the sample size for each gender is one. All female participants rated the same male model, and all male participants rated the same female model. This study provides great evidence that this particular woman and this particular man are respectively more and less attractive when smiling, but we have no evidence that this trend exists in the population at large. It is entirely plausible that for different subjects the trend would be reversed.

To really hammer this point home, consider the question – are songs in the key of C minor more enjoyable than those played in the key of D minor? What the authors have essentially done is asked the London Philharmonic to record two versions of Beethoven’s Symphony No. 5 – one version in the original C minor, and the other transposed into D minor. They then asked 184 participants to rate the enjoyability of one of the versions, and concluded that songs in C minor are more enjoyable than songs in D minor because participants on average gave the C minor version of Beethoven’s Symphony No. 5 a higher enjoyability score. Crazy, right!? Maybe Beethoven’s other symphonies actually sound better in D minor, or maybe his symphonies sound better in C minor but his sonatas are more enjoyable in D minor, or maybe Beethoven’s compositions are generally more enjoyable in C but Bach’s are consistently more enjoyable in D, etc. Point is, you can’t make generalizations with a sample size of one. Again, the authors do actually acknowledge this point, and claim that the second study addresses this shortcoming.

Problem number deux. In the second study, where photographs were obtained from the internet and many models were used in each category, I believe there were systematic differences between categories apart from just emotional expression. Admirably, the authors have posted all of the photos used in their study here. There are a few trends that really stuck out for me. One is that photographs in the pride samples were mostly comprised of athletes in their race or match apparel, whereas few or no athletes appeared in the other three categories. Another trend is that most neutral photographs tended to show only the face and sometimes shoulders, whereas hands, upper bodies, and even full bodies appeared in the other categories. A third issue is that neutral faces were almost always facing directly toward the camera with no angle or tilt, whereas faces and bodies in other categories were much more likely to be angled. There also seem to be differing proportions of professional-looking photographs between the different categories (the authors did partially control for the number of models that appeared to be professional models, but only in two of the three samples). In sample A, all of the shame photographs appear to be professionally taken, whereas most of the neutral photographs appear to have been taken by a kid at the DMV.

Going back to the music analogy, the authors have essentially downloaded a bunch of songs from iTunes, half in C minor and half in D minor, but for whatever reason most of their C minor songs happen to fall into the Hip-Hop & Reggae genre, and most of the songs in D minor happen to belong to the Country & Western genre. Even if we have a large and random sample of the population rating the enjoyability of these different songs, any average differences observed between songs in C and D minor are not necessarily due to the different key signatures, but could just as easily be due to any of the myriad differences that (on average) distinguish Hip-Hop music from Country music. Of course the same is true for the different sets of photographs in the study described above, except the confounding variables in this case were photograph quality, angle of head from camera, proportion of body appearing in the photograph, clothing and location of the model, etc.

To conclude (finally!), I don’t really doubt the claims made in this study, I just don’t think they necessarily follow from the obtained results. There are logistical limitations to any study, and we can rarely design studies that will definitively test a hypothesis of interest while controlling for every possible confounding factor. I do however think that it is reasonable and possible to more conclusively and meticulously test the hypothesis that emotional expression influences perceived attractiveness by members of the opposite sex.

____________________________________

Tracy, J. L., & Beall, A. T. (2011). Happy guys finish last: the impact of emotion expressions on sexual attraction. Emotion 11:1379-1387. DOI: 10.1037/a0022902