ABSTRACT
Research on facial attractiveness and face recognition has produced contradictory results that we believe are rooted in methodological limitations. Three experiments evaluated the hypothesis that facial attractiveness and face recognition are positively and linearly related. We also expected that social status would moderate the attractiveness effect. Attractive faces were recognized with very high accuracy compared to less attractive faces. We specified two estimates of facial distinctiveness (generalized and idiosyncratic) and demonstrated that the attractiveness effect on face recognition was not due to distinctiveness. This solves the long-standing problem that because facial attractiveness and distinctiveness are naturally confounded, construct validity is compromised. There was no support for the prediction, based on meta-analysis, that females would outperform males in face recognition. The attractiveness effect was so strong that gender effects were precluded. Methodological prescriptions to enhance internal, construct, and statistical conclusion validity in face recognition paradigms are presented.
Acknowledgements
The contents are solely the responsibility of the authors, and do not represent the official views of NIGMS, NIH, or BSF. Fredric Agatstein, Benjamin Jee, Avraham Kluger, and Meghan Sumeracki provided comments. Lorin Kinney and Keri Silva provided laboratory assistance. Raw data and code for statistical analyses are available from Thomas E. Malloy at [email protected].
Disclosure statement
No potential conflict of interest was reported by the author(s).
Notes
1 While beyond the scope of the present paper, binary recognition accuracy scores for signal and noise faces also have a nested structure. Mixed model ANOVA cannot be used; rather, generalized estimating equations (Hardin & Hilbe, Citation2013) can model correlated binary outcomes. This requires restructuring the data from Table 1 for analysis as a multi-level linear model.
2 For female raters, the mean difference in attractiveness ratings for attractive and average faces was 2.50 (95% confidence interval for the difference is 1.84: 3.16), for attractive and unattractive faces the difference was 3.85 (95% confidence interval for the difference is 3.19: 4.51), and for average and unattractive faces was 1.35 (95% confidence interval for the difference is .69: 2.01). Partial eta squared (ή2) is .90 (SSbetween/SStotal = 76.32/85.24 with degrees of freedom of 2 and 29, respectively). The standard error of the difference is .26 for all comparisons and Bonferonni protected comparisons of attractiveness means showed that all differed reliably with p < .001. For male raters, the mean difference in attractiveness ratings for attractive and average faces was 3.40 (95% confidence interval for the difference is 3.00: 3.80), the difference for attractive and unattractive faces was 4.00 (95% confidence interval for the difference is 3.60: 4.40), and the difference for average and unattractive faces was .60 (95% confidence interval for the difference is .20: 1.00). Partial eta squared (ή2) is .97 (SSbetween/SStotal = 93.07/96.37 with degrees of freedom of 2 and 29, respectively). The standard error of the difference is .16 for all comparisons. Bonferonni protected comparisons of attractiveness means showed that attractive compared to, average, and unattractive faces differed reliably with p < .001. Average and unattractive differed reliably with p = .002.