An official website of the United States government
Mental health stigma prevents many individuals from receiving the appropriate care, and social psychology studies have shown that mental health tends to be overlooked in men. In this work, we investigate gendered mental health stigma in masked language models. In doing so, we operationalize mental health stigma by developing a framework grounded in psychology research: we use clinical psychology literature to curate prompts, then evaluate the models’ propensity to generate gendered words. We find that masked language models capture societal stigma about gender in mental health: models are consistently more likely to predict female subjects than male in sentences about having a mental health condition (32% vs. 19%), and this disparity is exacerbated for sentences that indicate treatment-seeking behavior. Furthermore, we find that different models capture dimensions of stigma differently for men and women, associating stereotypes like anger, blame, and pity more with women with mental health conditions than with men. In showing the complex nuances of models’ gendered mental health stigma, we demonstrate that context and overlapping dimensions of identity are important considerations when assessing computational models’ social biases.
more »
« less
Gendered Mental Health Stigma in Masked Language Models
Mental health stigma prevents many individuals from receiving the appropriate care, and social psychology studies have shown that mental health tends to be overlooked in men. In this work, we investigate gendered mental health stigma in masked language models. In doing so, we operationalize mental health stigma by developing a framework grounded in psychology research: we use clinical psychology literature to curate prompts, then evaluate the models’ propensity to generate gendered words. We find that masked language models capture societal stigma about gender in mental health: models are consistently more likely to predict female subjects than male in sentences about having a mental health condition (32% vs. 19%), and this disparity is exacerbated for sentences that indicate treatment-seeking behavior. Furthermore, we find that different models capture dimensions of stigma differently for men and women, associating stereotypes like anger, blame, and pity more with women with mental health conditions than with men. In showing the complex nuances of models’ gendered mental health stigma, we demonstrate that context and overlapping dimensions of identity are important considerations when assessing computational models’ social biases.
more »
« less
- PAR ID:
- 10433110
- Date Published:
- Journal Name:
- Empirical Methods in Natural Language
- Page Range / eLocation ID:
- 2152–2170
- Format(s):
- Medium: X
- Sponsoring Org:
- National Science Foundation
More Like this
-
-
Mental health stigma manifests differently for different genders, often being more associated with women and overlooked with men. Prior work in NLP has shown that gendered mental health stigmas are captured in large language models (LLMs). However, in the last year, LLMs have changed drastically: newer, generative models not only require different methods for measuring bias, but they also have become widely popular in society, interacting with millions of users and increasing the stakes of perpetuating gendered mental health stereotypes. In this paper, we examine gendered mental health stigma in GPT3.5-Turbo, the model that powers OpenAI’s popular ChatGPT. Building off of prior work, we conduct both quantitative and qualitative analyses to measure GPT3.5-Turbo’s bias between binary genders, as well as to explore its behavior around non-binary genders, in conversations about mental health. We find that, though GPT3.5-Turbo refrains from explicitly assuming gender, it still contains implicit gender biases when asked to complete sentences about mental health, consistently preferring female names over male names. Additionally, though GPT3.5-Turbo shows awareness of the nuances of non-binary people’s experiences, it often over-fixates on non-binary gender identities in free-response prompts. Our preliminary results demonstrate that while modern generative LLMs contain safeguards against blatant gender biases and have progressed in their inclusiveness of non-binary identities, they still implicitly encode gendered mental health stigma, and thus risk perpetuating harmful stereotypes in mental health contexts.more » « less
-
Fontanesi, Lilybeth (Ed.)Concealable stigmatized identities (CSIs) are hidden identities that carry negative stereotypes and can result in a loss of social status if revealed. Instructors often choose to conceal these CSIs due to anticipated negative student reactions, yet revealing CSIs can have a positive impact on undergraduates. Women are generally more likely to reveal personal aspects about themselves in social situations, but may face greater consequences for revealing a stigmatized identity to students given their already marginalized position in academic science and engineering. Therefore, in this study, we were interested in understanding to what extent there are differences between men and women science and engineering instructors in (i) the representation of CSIs, (ii) their decisions to reveal CSIs to undergraduates in their classes, and (iii) their perceived stigma of CSIs. Based on a national survey of over 2,000 instructors in science and engineering from very high research activity doctoral institutions, we found that women were more likely than men to report having depression, anxiety, or a disability. Of instructors who held CSIs, women had 1.5x higher odds than men of revealing their CSIs to some undergraduates compared to no undergraduates and perceived greater stigma associated with all CSIs. Despite perceiving greater stigma associated with concealable stigmatized identities, women are more likely to reveal their CSIs to college science and engineering students, leading the way to a more diverse and inclusive scientific community by demonstrating themselves as role models for these identities.more » « less
-
null (Ed.)Previous research has identified a link between mental health and cyberbullying, primarily in studies of youth. Fewer studies have examined cyberbullying in adults or how the relation between mental health and cyberbullying might vary based on an individual's social media use. The present research examined how three indicators of mental health—depression, anxiety, and substance use—interact with social media use and gender to predict cyberbullying in adults. In Study 1, U.S. adults recruited through Amazon Mechanical Turk ( N = 525) completed an online survey that included measures of mental health and cyberbullying. Multiple regression analyses revealed significant three-way interactions between mental health, degree of social media use, and gender in models predicting cyberbullying victimization and perpetration. Specifically, for men, depression and anxiety predicted greater cyberbullying victimization and perpetration, particularly among men with relatively higher levels of social media use. In contrast, depression and anxiety were uncorrelated with cyberbullying for women, regardless of level of social media use. Study 2 largely replicated these findings using well-validated measures of mental health (e.g., Center for Epidemiological Studies-Depression scale, Beck Anxiety Inventory, Global Appraisal of Individual Needs Substance Use scale) in U.S. adults recruited through Prolific.co ( N = 482). Together, these results underscore the importance of examining mental health correlates of cyberbullying within the context of social media use and gender and shed light on conditions in which indicators of mental health may be especially beneficial for predicting cyberbullying in adults.more » « less
-
Gendered differences in academic confidence and self-efficacy between men and women are well-documented. In STEM fields and specifically in engineering, such differences have important consequences in that students low on these constructs are often more prone to leave their degree programs. While this evidence base is fairly established, less is known about the extent to which men and women show differences in confidence of team success, or collective efficacy, which may also be consequential in decisions to join and persist in design team experiences, or even to stay in or leave an engineering major, especially for first-year students. In this analysis, we quantitatively investigated gendered differences in confidence of team success and collective efficacy among first-year engineering students working on semester-long design projects in stable teams. Using a software tool built to support equitable teamwork, survey data on team confidence and collective efficacy was collected for these engineering students as well as for students in other courses for the sake of comparison. Three hierarchical linear models were fit to the data from 1,806 students across 31 unique course/term combinations. The results were mixed. In two of these analyses, we identified significant interactions between gender and team confidence. Specifically, men generally reported higher team confidence scores than women throughout the term with women eventually catching up, and team confidence ratings increased for men but not women following a lesson on imposter syndrome. No gendered differences were observed with respect to a collective efficacy scale administered near the middle and end of the term, however. In all cases, the results were consistent across course type (engineering, business, and others).more » « less
- Free Publicly Accessible Full Text
-
Accepted Manuscript1.0
- Conference Paper:
- The DOI is not currently available.
