Abstract
In face-based identity matching tasks, participants are shown to achieve higher performance when working in collaborative pairs than when working alone (Dowsett & Burton, 2015). However, identity-matching performance is also better when the responses of multiple participants are averaged (i.e., fused) on each item (White et al., 2013; White et al., 2015). Here we asked whether social collaboration adds power to the benefits of simple response averaging. In a challenging unfamiliar face-matching test, we compared individuals (n=68) and social dyads (n=34). Each participant performed the task once as an individual, and once as a member of a dyad, using different subsets of stimuli for each condition. The stimuli consisted of 84 pairs of frontal view face images (42 same-identity; 42 different-identity), varying in illumination and expression. Participants or dyads judged whether each face-pair depicted the same identity or two different identities, responding on a 5-point scale (1: sure same; 5:sure different). Performance was measured as area under the ROC (aROC). Accuracy was higher for social dyads (Mean aROC=0.88) than for individuals (Mean aROC=0.84), [F(1,100) = 5.78, p= .018], replicating Dowsett and Burton (2015). Next, to compare social collaboration to simple averaging, a "blind fusion" condition was formed by averaging the responses of pairs of participants within each social dyad on each stimulus. The accuracy obtained through the blind fusion condition (Mean aROC=0.91) was marginally greater than social dyad (Mean aROC=.88), [F(1,66)=4.13, p=.05]. Consistent with the literature, performance was also greater for blind fusion (Mean aROC=0.91), than for individuals (Mean aROC=0.84), [F(1,100)=18.14, p< .0001]. These findings suggest that the wisdom-of-crowds phenomenon observed in unfamiliar face-matching performance does not depend predominantly on the social aspect of dyad collaboration.
Meeting abstract presented at VSS 2017