Validation of data-driven computational models of social perception of faces.

Abstract

People rapidly form impressions from facial appearance, and these impressions affect social decisions. We argue that data-driven, computational models are the best available tools for identifying the source of such impressions. Here we validate seven computational models of social judgments of faces: attractiveness, competence, dominance, extroversion, likability, threat, and trustworthiness. The models manipulate both face shape and reflectance (i.e., cues such as pigmentation and skin smoothness). We show that human judgments track the models' predictions (Experiment 1) and that the models differentiate between different judgments, though this differentiation is constrained by the similarity of the models (Experiment 2). We also make the validated stimuli available for academic research: seven databases containing 25 identities manipulated in the respective model to take on seven different dimension values, ranging from -3 SD to +3 SD (175 stimuli in each database). Finally, we show how the computational models can be used to control for shared variance of the models. For example, even for highly correlated dimensions (e.g., dominance and threat), we can identify cues specific to each dimension and, consequently, generate faces that vary only on these cues.

Age-related similarities and differences in first impressions of trustworthiness

“…The images used to depict the trustee faces comprised of Caucasian male faces with neutral expressions that were all unique identities randomly generated and manipulated with FaceGen® 3.2 software (see Todorov et al, 2013). We selected 10 trustworthy and 10 untrustworthy faces that were +3 SDs and -3 SDs, respectively, from the neutral version of each face on a computer-modelled dimension of trustworthiness (see exemplar of face in Figure 1).…”

“…The trustee faces in the study by Rezlescu et al (2012) were computer-generated faces based on mathematical models that enable a tightly controlled assessment of subtle variations in facial features that indicate trustworthiness (Todorov, Dotsch, Porter, Oosterhof, & Falvello, 2013). One such feature includes increased distance between the eyebrows and the eyes, which enhances perceived trustworthiness.…”

Intact implicit processing of facial threat cues in schizophrenia

“…For example, emotionless craniofacial features such as larger facial width-to-height ratio correlate strongly with perceptions of greater propensity for aggression (Carré et al, 2009; Stillman et al, 2010), reliably elicit inferences of threat from healthy individuals (Todorov et al, 2013), and account for variance in the amount of aggressive behavior demonstrated by men (Carré & McCormick, 2008). Detecting such latent and static facial signals of threat likely has noteworthy consequences, as appraisal of social threat is believed to facilitate behavioral responses when interacting with others (Bar et al, 2006, Green & Phillips, 2004).…”

The Geneva Faces and Voices (GEFAV) database

“…Importantly, most databases do not provide stimuli that are distributed over a useful range of attractiveness. Among the very few databases of faces with controlled levels of attractiveness, one created by Todorov and colleagues consists of 25 computer-generated faces (not real human faces) varying on seven levels of attractiveness, and is freely available after completion of an agreement form (Todorov, Dotsch, Porter, Oosterhof, & Falvello, 2013). Another database (the Beautycheck project; Braun, Gruendl, Marberger, & Scherber, 2001) provides composite pictures derived from three or more original faces among neutral standardized pictures of 96 young adults (no original face available).…”