Collective intelligence in adolescents.

In experiment 1 (a,b,c), the errors in the pre-discussion initial estimates are significantly greater than both the group consensus (neg. bin. GLMM: z = 6.74, p < 0.001) and post-discussion estimates (z = 5.27, p < 0.001). Group estimates tended to have less error on average than individual estimates given after discussion although this effect was not statistically significant (z = -1.77, p = 0.077). The frequency distributions of the error in the initial estimates minus the error in the group consensus estimates (b) or post-discussion individual estimates (c) per participant show that while most participants only gain a small improvement in reduced errors (b) or show little change in errors (c), a minority of individuals vastly reduce the error in their estimates after group discussion. In Experiment 2, the absolute error in the consensus group estimate was lower than the initial (pre) estimates in all three treatments (proportion of black sweets: d: 48/200, e: 94/190, f: 121/160). The box plots show the median (thick black lines), interquartile range (enclosed by the boxes), 1.5 × the interquartile range beyond the boxes (whiskers) and outliers beyond the whiskers (open circles).