Maintenance of Subliminal Facial Expressions in Working Memory and Its Subsequent Bias on Facial Judgement

Affective states implicitly bias evaluations of unrelated stimuli, even in the absence of conscious awareness of the affective source. According to affect-as-information theory, this originates from affective misattribution: individuals erroneously attribute affect elicited by one stimulus to a distinct, unrelated stimulus, thereby causing perception of the latter to adopt the emotional valence of the former. Although robust evidence shows that subliminal affective stimuli bias emotional perception, social behavior, and decision-making through transient priming, it remains unknown whether these influences persist beyond the immediate exposure period. Across four experiments (N = 105) utilizing delayed facial discrimination tasks, we demonstrate that subliminal affective information is not only stored in nonconscious working memory (Experiment 1), but also actively maintained against suprathreshold interference (Experiment 2). This representation retains abstract affective content rather than precise identity details (Experiment 3). Crucially, these latent affective trace drive valence-congruent misattribution onto subsequently encountered neutral faces (Experiment 4), establishing affective misattribution as a consequence of nonconscious working memory. This study provides a novel mechanism for affect-as-information theory, revealing the continuous influence of subliminal affective information on supraliminal processing. It also deepens our understanding of the irrational underpinnings in behavioral economics by tracing their origins to latent biases in affective memory.