Beyond faces: The effect of context on ensemble emotion perception

Abstract

The ability to judge the overall emotion of a crowd is fundamental to navigating naturalistic social scenes. However, research has typically studied such judgments using isolated faces, without any contextual information. In two experiments, this study investigates how facial expressions, background context, and observers’ emotional states jointly influence ensemble emotion perception in naturalistic scenes. In Experiment 1, participants judged the overall emotion of crowd scenes that were constructed by superimposing human figures, whose facial emotional values were quantified on a continuous scale based on naturalistic photographs, onto emotionally positive, negative, or neutral backgrounds. Bayesian model comparisons revealed that both facial emotion and background valence influenced mean judgments, with positive (negative) backgrounds biasing ratings in a more positive (negative) direction. In Experiment 2, we examined how observer mood modulates this integration by inducing positive or negative affect through an autobiographical recall task. Model comparisons revealed that observer mood also influenced judgment, extending the model of Experiment 1. Although observer mood exhibited no reliable main effect on mean ratings, it was found to significantly affect response variability: participants in a positive mood showed reduced variability, which suggests more consistent ensemble judgments. Together, these findings demonstrate that ensemble emotion perception integrates multiple sources of information beyond facial expressions; specifically, both external and internal contexts substantially influence how collective mood is perceived in social scenes. These findings emphasize the importance of studying ensemble perception in naturalistic settings. The trial-level data and analysis codes for all the experiments are available on the Open Science Framework (OSF) repository (https://doi.org/10.17605/OSF.IO/B9GMQ). None of the experiments were preregistered.