Siddharth Munupatrula¹ (), Daniel N. Albohn², Ignazio Ziano³, Yasin Koc⁴, Stefan Uddenberg¹; ¹University of Illinois Urbana-Champaign, ²University of Chicago Booth School of Business, ³University of Geneva, ⁴University of Groningen

Mental illness is stigmatized, leading to discrimination and social exclusion. Are there shared visual stereotypes of mental illness amongst the general population? If so, what is the visual content of such stereotypes? And are individual representations of mental illness influenced by the level of stigmatizing beliefs the perceiver might hold? Here we explored these issues using hyper-realistic generative reverse correlation (Albohn et al., 2022; 2025). Participants (N=420) read a description of a prevalent mental illness (i.e., bipolar disorder, general anxiety disorder [GAD], major depressive disorder [MDD], or post-traumatic stress disorder [PTSD]) and judged whether synthetic faces (300 faces, presented serially) looked like they "had" or "did not have" the described condition (they could also respond "not sure" if uncertain about their choice). We extracted mental representations of these conditions by taking the difference between the averaged faces judged as “having” versus “not having” the condition, yielding a vector in face space distinguishing the two categories (Albohn et al., 2022; 2025). These representations were then compared to our pre-existing models of social attribute impressions (Peterson et al., 2022) using signed cosine similarity (s) where s>0 indicates alignment and s<0 anti-alignment with the attribute vector. Across all conditions, illness representations were visually interpretable and reliably anti-aligned with perceived health, trustworthiness, and attractiveness, while being reliably aligned with perceived masculinity (all conditions). In addition, some conditions were aligned with perceived age (GAD/PTSD), dominance (PTSD only) and darker skin tone (PTSD only). Contrary to our expectations, participants' Mental Illness Stigma Scale (Day et al., 2007) scores were not systematically associated with visual differences in the extracted representations of the illnesses, suggesting that one's visual stereotypes of mental illness may not be driven by stigmatizing beliefs.