Abstract
Information from multiple similar objects in a set may be coded in the form of a summary statistic such as the set average (Ariely, 2001). In a series of 6 experiments we tested identity averaging for famous face sets. Participants were shown sets of four faces from different celebrities, followed by a probe face, and indicated, whether or not the a) probe image or b) identity had been presented in the set. Strikingly, participants in Experiment 1 highly frequently responded ‘present’ not only (correctly) when the probe had been a set item, but also (incorrectly) when the probe was an average (morph) of set identities. Previous work with faces had suggested an abstractive mechanism across unfamiliar identities (de Fockert & Wolfenstein, 2009; Haberman & Whitney, 2007), but the present results suggest abstraction of stimulus-set characteristics even in circumstances that support individuation of familiar identities. Three control studies addressed alternative explanations, including the ratio of required present responses, and potential attendance to only few exemplars during set presentation. In each case robust set averaging was observed. However, limits of the abstractive representation system were revealed in two further experiments. First, reversing the presentation order of set and probe led to better rejection of set averages during image matching. If perceptual similarity between set and probe had caused the set averaging effect in the previous experiments, no such reduction would have been expected, since similarity was exactly the same. In a final experiment, we created sets from each two male and two female faces. Set averaging occurred for within-gender averages, but was significantly reduced for across-gender averages. In conclusion, summary representations for faces seem to be formed mandatorily and in parallel to exemplar representations, given that set members belong to a common subcategory (e.g., same gender).
Meeting abstract presented at VSS 2012