Revealing the latent structure of multidimensional facial features that drive social trait perceptions

Hensel, Laura Birka (2022) Revealing the latent structure of multidimensional facial features that drive social trait perceptions. PhD thesis, University of Glasgow.

Full text available as:
[img] PDF
Download (30MB)


Humans readily attribute social traits to others based on their facial appearance, influencing behaviors, social interactions, and decision making. According to influential models of social perception, these judgments are based on two central dimensions – trustworthiness/warmth and dominance/competence. Because of the wide-reaching impact of such social attributions, a long-standing focus has been to identify the facial features that elicit these social judgments. However, the face is complex, comprising 3D shape, 2D complexion, and dynamic facial expressions, making such investigations empirically challenging. As a result, central questions regarding fundamental facial features of social traits, and how these relate to other social judgments such as social class and emotion, remain unanswered. In this thesis, I use data-driven psychophysical methods to mathematically model the 3D shape, 2D complexion and dynamic facial expressions that elicit judgments of four key social trait dimensions – dominance, competence, trustworthiness, and warmth. I then identify the latent face feature space underlying these judgments using a data-reduction technique. Results reveal two latent 3D shape and three latent 2D complexion feature spaces on the basis of which social traits cluster into four distinct subgroups. Moreover, I show that these social trait feature spaces correlate positively with facial expression features (shape) and age-cues (complexion). I then examine whether these features also drive perceptions of another important social judgment – social class. To do so, I model the 3D shape and 2D complexion of social class, using the same approach as for social traits. I compare these models using a data reduction technique and a supervised machine learning approach. Results show that in line with conceptual overlaps arising from social class stereotypes (e.g., poor = incompetent), social class and social trait dimensions share facial features. However, no single trait’s features fully account for social class features. Finally, I compare the social trait facial expression models to emotional expressions and social trait face shapes to reveal the features shared between each. Results showed that longstanding associations between perception of emotion and social traits (e.g., happy = trustworthy), only partially account for social trait perception. The current work informs central theories of social perception, highlights drivers of socially relevant stereotype associations, and can aid the development of psychologically grounded digital agents.

Item Type: Thesis (PhD)
Qualification Level: Doctoral
Subjects: B Philosophy. Psychology. Religion > BF Psychology
Colleges/Schools: College of Science and Engineering > School of Psychology
Supervisor's Name: Jack, Professor Rachael E. and Schyns, Professor Philippe G.
Date of Award: 2022
Depositing User: Theses Team
Unique ID: glathesis:2022-82971
Copyright: Copyright of this thesis is held by the author.
Date Deposited: 16 Jun 2022 10:39
Last Modified: 16 Jun 2022 10:44
Thesis DOI: 10.5525/gla.thesis.82971
Related URLs:

Actions (login required)

View Item View Item


Downloads per month over past year