Opinion
Understanding evaluation of faces on social dimensions

https://doi.org/10.1016/j.tics.2008.10.001Get rights and content

People reliably and automatically make personality inferences from facial appearance despite little evidence for their accuracy. Although such inferences are highly inter-correlated, research has traditionally focused on studying specific traits such as trustworthiness. We advocate an alternative, data-driven approach to identify and model the structure of face evaluation. Initial findings indicate that specific trait inferences can be represented within a 2D space defined by valence/trustworthiness and power/dominance evaluation of faces. Inferences along these dimensions are based on similarity to expressions signaling approach or avoidance behavior and features signaling physical strength, respectively, indicating that trait inferences from faces originate in functionally adaptive mechanisms. We conclude with a discussion of the potential role of the amygdala in face evaluation.

Introduction

The human face is a perennial source of fascination as a window to one’s character [1]. Indeed, people routinely make trait inferences (e.g. aggressiveness) from faces, despite mixed evidence for the accuracy of these inferences (Box 1). As little as 100 ms exposure to unfamiliar faces provides sufficient information for trait inferences 2, 3 and evolutionarily important inferences such as threat can be made even after shorter exposures [4]. These inferences predict important social outcomes ranging from electoral success to criminal sentencing decisions 5, 6, 7, 8, 9. For example, inferences of competence based solely on facial appearance predict U.S. Senatorial and Gubernatorial elections 3, 7 and inferences of dominance predict military rank attainment [10].

The traditional approach to studying trait inferences from facial appearance has been to focus on specific trait dimensions (Box 2). For example, among trait inferences, inferences of trustworthiness [11] have received extensive research attention in both behavioral 12, 13, 14, 15 and functional magnetic resonance imaging (fMRI) studies 16, 17, 18, 19. However, focusing on a single trait dimension is problematic because trait judgments from faces are highly correlated with each other [20]. That is, for any set of faces, there are multiple social dimensions that co-vary with each other and, thus, several alternative explanations of observed empirical relationships between a trait judgment and a behavior or brain activation. For example, two trait judgments – how caring and how attractive a person is – accounted for 84% of the variance of trustworthiness judgments that predicted the amygdala activation to faces in an fMRI study of implicit face evaluation [17]. Without independent evidence for the primacy of one trait inference over another, it is equally plausible to argue that ‘caring’ inferences and attractiveness, rather than trustworthiness, drive the response of the amygdala to faces. Statistically controlling for such variables is almost impossible given the large proportion of shared variance. Although it is possible to experimentally unconfound variations of faces on social dimensions, it is not at all clear what dimensions one should choose given their exceedingly large number. (In English, there are at least 4000 adjectives that describe interpersonal relationships [21].)

Instead of focusing on single trait dimensions, we advocate an alternative, data-driven approach with the objectives of finding the structure and perceptual basis of judgments from emotionally neutral faces. This approach is better suited than traditional approaches to address two of the fundamental questions of the study of social judgments from faces: what do these judgments really measure and what is their functional basis?

Section snippets

The structure of face evaluation

To identify dimensions used to spontaneously characterize faces, we selected the most frequently used trait dimensions from unconstrained person descriptions of emotionally neutral faces [20]. The faces were then rated along these dimensions by separate groups of participants. For each face, the mean ratings on the dimensions were submitted to a principal components analysis (PCA), a technique that reduces data dimensionality (Box 3). The first two principal components (PCs) accounted for more

The perceptual basis of face evaluation

The 2D model described earlier served as a guiding framework for computer modeling of face variation on social dimensions. Given that trustworthiness and dominance judgments were closest in space to the two PCs, we modeled how faces vary on these dimensions using a data-driven statistical model of face representation (see Ref. [25], and Singular Inversions, 2005: http://www.facegen.com). First, we collected trustworthiness and dominance judgments of computer-generated faces. Second, based on

The functional basis of face evaluation

In one of the first systematic attempts to understand trait judgments from faces, Secord [28] suggested that such judgments are based on misattribution of momentary states to enduring attributes. Accessible facial cues (e.g. smile) can be generalized to stable dispositions (e.g. friendly). Subsequent theories have emphasized that evaluation of faces is constructed from cues that have adaptive significance [29]. These cues can be either dynamic, expressing emotional states [30], or invariant

The role of the amygdala in face evaluation

To date, research of the neural mechanisms underlying trait impressions from faces has largely focused on trustworthiness evaluation 11, 12, 16, 17, 18, 19. These studies have all reported involvement of the amygdala, a subcortical brain region crucial for coding the motivational value of stimuli. Patients with bilateral amygdala lesions show impairments in discriminating trustworthy- from untrustworthy-looking faces [12]. Consistent with these findings, subsequent fMRI studies on healthy

Conclusions

We have outlined a general approach to study face evaluation and described a simple 2D model based on this approach. According to this model, when specific decision context is not provided (Box 5), faces are automatically evaluated along the dimensions of valence/trustworthiness and power/dominance. These dimensions define a 2D space within which specific social judgments can be represented [20]. The facial cues used for face evaluation along these dimensions indicate that evaluation of

Acknowledgements

This research was supported by National Science Foundation Grant BCS-0446846.

References (57)

  • J. Willis et al.

    First impressions: making up your mind after 100 ms exposure to a face

    Psychol. Sci.

    (2006)
  • C.C. Ballew et al.

    Predicting political elections from rapid and unreflective face judgments

    Proc. Natl. Acad. Sci. U. S. A.

    (2007)
  • M. Bar

    Very first impressions

    Emotion

    (2006)
  • R. Hassin et al.

    Facing faces: studies on the cognitive aspects of physiognomy

    J. Pers. Soc. Psychol.

    (2000)
  • A. Todorov

    Inferences of competence from faces predict election outcomes

    Science

    (2005)
  • I.V. Blair

    The influence of Afrocentric facial features in criminal sentencing

    Psychol. Sci.

    (2004)
  • A. Mazur

    Military rank attainment of a West Point class: effects of cadets’ physical features

    Am. J. Sociol.

    (1984)
  • A. Todorov

    Evaluating faces on trustworthiness: an extension of systems for recognition of emotions signaling approach/ avoidance behaviors

    Ann. N. Y. Acad. Sci.

    (2008)
  • R. Adolphs

    The human amygdala in social judgment

    Nature

    (1998)
  • R. Adolphs

    Abnormal processing of social information from faces in autism

    J. Cogn. Neurosci.

    (2001)
  • A. Todorov et al.

    Reading trustworthiness in faces without recognizing faces

    Cogn. Neuropsychol.

    (2008)
  • U. Bellugi

    Towards the neural basis for hypersociability in a genetic syndrome

    Neuroreport

    (1999)
  • J.S. Winston

    Automatic and intentional brain responses during evaluation of trustworthiness of face

    Nat. Neurosci.

    (2002)
  • A.D. Engell

    Implicit trustworthiness decisions: Automatic coding of face properties in human amygdala

    J. Cogn. Neurosci.

    (2007)
  • C.P. Said

    Nonlinear amygdala response to face trustworthiness: contributions of high and low spatial frequency information

    J. Cogn. Neurosci.

    (2008)
  • A. Todorov

    Evaluating face trustworthiness: a model based approach

    Soc. Cogn. Affect. Neurosci.

    (2008)
  • N.N. Oosterhof et al.

    The functional basis of face evaluation

    Proc. Natl. Acad. Sci. U. S. A.

    (2008)
  • J.S. Wiggins

    A psychological taxonomy of trait descriptive terms: the interpersonal domain

    J. Pers. Soc. Psychol.

    (1979)
  • Cited by (496)

    View all citing articles on Scopus
    View full text