Confidence in crowd emotion judgments is shaped by disconfirmatory emotional evidence

Individuals routinely make categorical judgments about crowd emotion by integrating evidence from multiple faces. However, how confidence in these judgments is computed remains largely unknown, despite its importance in guiding adaptive behaviours. A key question is whether all faces contribute equally to confidence in crowd emotion judgments. To address this, we asked participants to judge whether the mean emotion of a facial ensemble was happier or sadder than a reference face and then to report their confidence. Ensemble elements were ranked from happiest to saddest (ranks 1–8) and used to predict trial-by-trial confidence. Results showed that confidence in mean emotion judgments (e.g. “happier”) was better predicted by variations in emotional intensity of more response-incongruent (ranks 5–8) than more response-congruent elements (ranks 1–4). Given that more response-incongruent elements lie closer to the category boundary, our results can be explained by the robust averaging principle, which assigns greater weight to evidence near the category boundary. Overall, these findings suggest that confidence in crowd emotion perception is computed through a disproportionate weighting of disconfirmatory emotional evidence.