August 2016
Volume 16, Issue 12
Open Access
Vision Sciences Society Annual Meeting Abstract  |   September 2016
Understanding the social dimensions of facial attractivenss
Author Affiliations
  • Amanda Song
    Cognitive Science Department, University of California, San Diego
  • Linjie Li
    Electrical and Computer Science Department, University of California, San Diego
  • Vicente Malave
    Cognitive Science Department, University of California, San Diego
  • Angela Yu
    Cognitive Science Department, University of California, San Diego
  • Garrison Cottrell
    Computer Science and Engineering Department, University of California, San Diego
Journal of Vision September 2016, Vol.16, 493. doi:10.1167/16.12.493
  • Views
  • Share
  • Tools
    • Alerts
      This feature is available to authenticated users only.
      Sign In or Create an Account ×
    • Get Citation

      Amanda Song, Linjie Li, Vicente Malave, Angela Yu, Garrison Cottrell; Understanding the social dimensions of facial attractivenss. Journal of Vision 2016;16(12):493. doi: 10.1167/16.12.493.

      Download citation file:

      © ARVO (1962-2015); The Authors (2016-present)

  • Supplements

What makes a face attractive has long been an interesting topic in face perception research. Previous studies have shown that an attractive face is conferred with other positive social attributes, such as trustworthiness, and industriousness; conversely, attractiveness appears to be at least partially predictable from other perceived attributes, such as averageness and familiarity. To better understand facial attractiveness in the context of social attributes of faces, we investigate the relationship between facial attractiveness and multiple social attributes of faces based on a large face dataset with social feature ratings and attractiveness ratings (Wilma, Isola, & Oliva, 2013). It provides attractiveness ratings and social feature ratings of over 2000 faces. Correlations between every social feature and attractiveness show that 37 of them are significantly correlated with attractiveness (p< 0.001). We calculate how much variance each individual social feature explains in attractiveness ratings. The top 5 positively correlated social feature predictors are: interesting (correlation coefficient: 0.66, explained variance: 43% ), sociable (0.60, 39%), memorable (0.54, 29%), confident ( 0.53, 28%), normal (0.53, 28%), whereas the top 5 negatively correlated social feature predictors are: boring (-0.59, 35%), weird (-0.56, 31%), forgettable (-0.51, 26%), introverted ( -0.49, 24%), and unhappy (-0.47, 22%). Since social attributes are highly correlated, we use principal component analysis to get an orthogonal feature matrix, then re-calculate how many principal components are correlated with attractiveness. This time, we find that only 11 principal components significantly correlated (p< 0.001) with attractiveness. We computed the prediction error (MSE±SD) as a function of the number of reduced dimensions and find that the average prediction performance in the test set indeed reaches a plateau after 11 dimensions (MSE±SD ≈ 0.7 ± 0.05). This suggests that social attributes can serve as good predictors of attractiveness and the intrinsic dimension relevant to it is 11.

Meeting abstract presented at VSS 2016


This PDF is available to Subscribers Only

Sign in or purchase a subscription to access this content. ×

You must be signed into an individual account to use this feature.