Perceived trustworthiness biases memory for emotional expressions
Poster Presentation 43.318: Monday, May 18, 2026, 8:30 am – 12:30 pm, Banyan Breezeway
Session: Face and Body Perception: Social cognition 2
Schedule of Events | Search Abstracts | Symposia | Talk Sessions | Poster Sessions
Chunliang Yang1 (), Stefan Uddenberg1; 1University of Illinois Urbana-Champaign
People rapidly and automatically form trait impressions from faces, and subtle changes in facial traits influence perceived emotional valence. Indeed, the same emotional expression is judged as more positive on trustworthy-looking faces as opposed to untrustworthy-looking faces, indicating a tight coupling between trait impressions and emotion perception (Oosterhof & Todorov, 2009). Whether facial traits also distort memory for emotional expressions, however, remains unknown. Here we tested whether invariant facial morphology biases the remembered valence of a briefly viewed expression. First, we synthesized 60 hyper-realistic, emotionally neutral identities (demographically homogenous, to limit potential other-race effects in memory). These identities were designed to be especially high or low in perceived trustworthiness (30 each), using a trait impression model trained on over one million human judgments (Peterson et al., 2022). Using open source emotion editing tools, we then created a mean luminance-matched emotional continuum (happy-neutral-sad) for each identity. In the main experiment, observers (N=100) engaged in a continuous memory reproduction task, first viewing a single face displaying a given expression from the continuum, then reproducing the remembered expression (after a brief delay) by adjusting a test face using a circular slider. Our primary measure of interest was “memory bias”, operationalized as the signed difference between the reproduced and originally viewed expression as a function of the identity’s trustworthiness level (high or low). Across both conditions, on average, observers reproduced faces as happier than they originally were. However, this “happiness” bias in memory was stronger for trustworthy-looking faces than untrustworthy-looking ones. These results suggest that enduring trait-like signals in facial morphology penetrate not only online perception of emotion, but also the encoding and/or reconstruction of facial expressions in memory, biasing what we remember about how others felt.