Social Vision, Gender Atypicality, & Body Language: Applying a Gender Predictive Approach to Study Gait

• In Event: 30663 - Computational Approaches to Culture and Cognition

Abstract

Recent social science research has begun to document “infracategorical inequalities” (e.g. colorism), which fall along axes of perceived categorical cues and typicality (i.e. skin tone, gender expression) rather than categorical membership alone (i.e. race, gender identity). In particular, multiple studies have found disparities in health, bullying, and labor market outcomes by gender expression. These studies find that people who have cues that are considered atypical for their gender or who are misgendered experience worse outcomes on average. Drawing on experimental approaches, social vision science offers promising methods to study how gendered cues are perceived by others and how these perceptions in turn shape these social outcomes. However, previous approaches to studying gender expression in social vision studies largely rely on selecting or manipulating stimuli based on gender stereotypes and then measuring how these cues influence observers’ perceptions. We argue there is need to develop alternative methods to select stimuli, which capture cues related to “non-declarative cultural” conceptions of gender—subtle, implicit, and non-articulated understandings that are learned and internalized through habituation rather than explicitly articulated stereotypes. Our project begins to address this gap by leveraging computer vision. In this paper, we apply a “gender predictive approach” to analyze gait in a corpus of video data of people in public parks. A gender predictive approach uses supervised machine learning (SML) to train and fit a gender classification model using empirical data and then derives a measure of gender atypicality from the model’s predictive probabilities. The gender atypicality scores derived from this model can then be integrated into the selection of video stimuli for experimental research on social vision, body language, and gender expression. We discuss how this approach could provide insights into inequalities by gender expression across a range of outcomes, including hiring discrimination, street harassment, victimization, policing, and dating.

Authors

• Doron Raoul Shiffer-Sebba, Northwestern University

• Kris Rosentel, Northwestern University

• Yoav Goldstein, Stanford University