Reading trustworthiness in faces without recognizing faces

Abstract

We show that developmental prosopagnosics with severe impairments in both memory for faces and perception of facial identity can make normal trustworthiness judgements of novel faces. We tested four prosopagnosics on three different face sets. The first set consisted of faces that varied on multiple dimensions and that have been used to demonstrate impairments in trustworthiness judgements of patients with bilateral amygdala damage. The second and third sets consisted of standardized faces with direct gaze, neutral expression, and similar age. On all tests, two prosopagnosics made judgements that closely agreed with control judgements while the other two showed weak agreement but within the normal range. The performance of the tests was correlated suggesting that the tests mapped the same underlying judgement irrespective of the specific face stimuli. The normal performance of two of the prosopagnosics suggests that forming person impressions from faces involves mechanisms functionally independent of mechanisms for encoding the identity of faces.

Keywords:

• Face perception
• Social cognition
• Prosopagnosia
• Trustworthiness
• Amygdala

We thank Manish Pakrashi, James Chu, Chris Said, and Valerie Loehr for their help in running the experiments, Galit Yovel for providing us with the face part/spacing discrimination task, and Ralph Adolphs for providing the stimuli used to test patients with bilateral amygdala damage and the raw data for our reanalysis. This research was supported by National Science Foundation Grant BCS-0446846.

Notes

¹ We also conducted correlational analyses using the nonparametric Spearman's rho coefficient. The results were practically identical. For example, the Pearson's correlations for the first set of faces were .64, .66, .50, and .50 for J.K., T.U., J.P., and J.L., respectively. The corresponding nonparametric correlations were .63, .66, .50, and .44 (p < .0001 for all correlations).

² Note that although within-subjects correlations are not very high, they are generally higher than between-subjects (interrater agreement) correlations. For the faces and the sample used in Engell et al. (2007), the average between-subjects correlation was .28 (SD = .17).

³ However, it should be noted that the third face set had the smallest number of faces, and, correspondingly, the significance threshold was more stringent because of the fewer degrees of freedom.

⁴ Participants were told that although the faces were emotionally neutral, they could show subtle emotional information and were asked to rate how angry or happy the person is on a 9-point scale, ranging from 1 (not at all) to 9 (moderately). Both judgements were reliable, Cronbach's α = .92 for judgements of happiness and α = .84 for judgements of anger.