“…As mentioned above, in the BD ensemble condition, even though probe face C (the average representation of the previously presented ensemble) was not presented in the ensemble stimuli, it was perceived as equally familiar as the actually presented probe faces (B faces and D faces), and the familiarities of these three probe faces were significantly larger than those of other unpresented probe faces (A faces and E faces). This average bias accords with previous studies regarding ensemble coding of facial expressions in a member judgement task [17,45,46], where observers tend to recognize the unseen average facial expression as a member of previous ensemble stimuli, implying that observers form an average representation when they perceive a crowd of faces automatically. However, these studies either only showed within-category crowds or did not discriminate the contributions of cross-category crowds from within-category crowds.…”