Valence Asymmetry in Facial Impressions: Evidence from Idiosyncratic Visual Modeling

While prior research has shown the valence asymmetry in semantic spaces, less is known about whether the valence of social traits shapes how people visually represent others. The current research examined whether positive and negative trait representations differ in their degree of consistency across individuals. Applying the generative reverse-correlation method, we constructed individualized visual models of a positive trait (i.e., friendly) or a negative trait (i.e., unfriendly) while measuring how well these representations converge across individuals. We hypothesized that visual representations of positive social traits will be more similar across individuals, whereas representations of negative traits will be more idiosyncratic. With two pre-registered studies (Study 1a & 1b), we found that the visual models of friendliness were significantly more similar to each other than the visual models of unfriendliness. This suggested that individuals may represent positive traits in a more shared manner, while they perceive negative traits with greater personal differences (Study 1). Moreover, the computational simulations revealed a potential mechanism behind this asymmetry (Study 2). Using identities generated from the idiosyncratic visual model, the results replicated the pattern that friendly prototypes were more similar than unfriendly prototypes, especially when positive identities were sampled more frequently. The observed valence asymmetry could be attributed to both the asymmetric frequency with which individuals encounter positive versus negative social exemplars and the inherent distribution of positive versus negative traits in representational space. Overall, these findings suggested that valence asymmetry is not only reflected in language but also in visual perception of social traits. To obtain better generalizability, future studies could extend this paradigm to other trait pairs with opposite valence, such as competent versus incompetent.