Approach Not Avoid: Validation of Data-Driven Computational Models of Trustworthiness Unconfounded by Attractiveness

Abstract

Trustworthiness judgments from faces influence key decisions such as approach/avoidance responses. In Experiment 1, we revalidated the Standard Trustworthiness Model, a data-driven computational model of trustworthiness, which visualizes impressions of trustworthiness (Oosterhof & Todorov, 2008). By applying the model to 25 novel facial identities, we showed that as model manipulation increases, both trustworthiness and attractiveness ratings of the faces increase. This finding exemplified that as the Halo Effect predicts, trustworthiness is confounded by attractiveness. In Experiment 2, we validated a new model: the Difference [trust - attr] Model, in which the attractiveness model was subtracted from the trustworthiness model. In Experiment 3, we introduced and validated an Orthogonal [trust ⟂ attr] Model, in which trustworthiness was uncorrelated to attractiveness. Both of these models manipulated the perceived trustworthiness of faces while controlling for attractiveness. These new trustworthiness models in which attractiveness is trivial, open the door for researchers to examine what facial cues such as emotional expressions and typicality, in addition to attractiveness, influence trustworthiness impressions.