Abstract
The wisdom-of-crowds effect is the tendency of a group to perform better than most individuals and sometimes better than the best individual within the group. This phenomenon was first demonstrated for tasks that required estimations of weights or sizes (Bruce, 1935; Galton, 1907; Gordon, 1924). However, a ubiquitous observation is that certain groups perform better than others, and one factor that appears to play a role in obtaining a good group performance is diversity. For example, simulations have shown that groups of diverse algorithms can outperform groups made of the best algorithms (e.g. Hong & Page, 2004). Here, we evaluated whether group diversity in the individual members’ use of information for face identification—measured using the Bubbles method—is also associated with group performances in face recognition—measured using the Cambridge Face Memory Test (CFMT). We randomly generated groups of sizes 2 to 11 from a sample of 102 participants. Group performance was obtained by averaging the result of the application of the majority rule on all CFMT trials. Group diversity was indexed by the inverse of the average Pearson correlation between the group members’ standardized classification images. Our main result is that, contrary to what we expected from the literature, diversity in use of information is negatively correlated with group performance (across group sizes: r = -.23; p < .05, two-tailed, Bonferroni-corrected). This seems to stem from inefficient human strategies for face identification being more diverse than efficient ones and, therefore, from diverse groups containing more unskilled than skilled participants. In any case, our results show that factors other than diversity can be important for predicting group performances.