Face Race, Face Sex, and Emotion Categorisation: A Bayesian Psychometric and Diffusion Model Investigation of Happy-Angry Judgements

Abstract

Human observers often agree when facial expressions are clearly happy or clearly angry. However, when expressions become ambiguous, emotion judgements may be influenced by characteristics of the face itself. Previous research suggests that facial emotion recognition is affected by face sex, gender stereotypes, facial structure, and social categorisation processes. The present study investigates whether the race and sex of a face influence the categorisation of emotional expressions presented along an angry-to-happy continuum. Participants will complete an emotion categorisation task in which they view morphed facial expressions derived from the Chicago Face Database. The stimulus set consists of 20 identities (5 Black Female, 5 Black Male, 5 White Female, and 5 White Male faces). Each identity is presented at nine levels of emotional intensity ranging from strongly angry, through neutral, to strongly happy. On every trial, participants categorise the face as either “Happy” or “Angry,” and both response choice and reaction time are recorded. The study employs a fully repeated-measures design with three within-subject factors: Face Race (Black vs White), Face Sex (Female vs Male), and Expression Intensity (9 levels). This produces 36 unique stimulus conditions completed by every participant. The primary goal of the study is to determine whether category-dependent biases emerge during emotion recognition and, critically, to identify the cognitive mechanisms responsible for any observed effects. To achieve this, two complementary modelling approaches will be used. First, Bayesian psychometric functions will be fitted separately for each face category. These models will be used to estimate posterior distributions for the Point of Subjective Equality (PSE), Just Noticeable Difference (JND), and Weber Ratio. PSE will be interpreted as the location of the category boundary between Happy and Angry responses, while JND and Weber Ratio will be interpreted as indices of perceptual sensitivity. Second, hierarchical Bayesian diffusion models will be fitted to both response and reaction-time data. Drift rate (v) will be used as an index of evidence accumulation, whereas starting-point bias (z) will be interpreted as an index of prior response bias or expectation. These parameters allow an explicit distinction between perceptual accounts of emotion recognition and decisional or stereotype-based accounts. The study is motivated by three competing theoretical perspectives. Gender stereotype accounts predict category-dependent shifts in decision thresholds. Perceptual-structure accounts predict changes in sensitivity and evidence accumulation due to structural similarities between facial appearance and emotional expressions. Social categorisation and threat-perception accounts predict category-dependent response biases associated with race and sex. The primary hypotheses are that face race and face sex will influence psychometric thresholds (PSE) and diffusion-model starting-point bias (z), and that perceptual accounts may additionally predict differences in psychometric steepness (JND/Weber Ratio) and drift rate (v). Participant sex and participant race/ethnicity will be retained as exploratory moderator variables. The expected outcome of the project is a more detailed understanding of whether biases in emotion recognition arise because observers extract different emotional information from particular categories of faces or because they begin the decision process with different expectations. By combining psychometric modelling with diffusion modelling, the study aims to provide a mechanistic account of how face race and face sex influence emotion categorisation.