The use and consequences of different aggregation rules in Experiment 1.

(a) Probability of each aggregation rule being the closest to the observed consensus estimates of the groups (filled circles). Also plotted is the probability (mean is black line, and shaded region is 95% confidence intervals) that the aggregation rule is the closest to the observed consensus estimates when a ‘noisy’ geometric mean simulation is instead used to aggregate the initial estimates (S1 Methods). (b) Range and skew in initial estimates for each group. The range of initial estimates is plotted against the relative distribution of the estimates. Skew is ρ = (x2−x1)/(x3−x1), and is close to zero if the highest estimate (x3) is a relative outlier, and close to one if lower estimate (x1) is a relative outlier. The threshold between groups of low and high range (red line) is an approximate point that separates the region with any configuration (≤40) to the region with the two lower estimates being much closer to each other than to the higher (>40). (c) As (a), but separately for groups with a low range of estimates (blue dots, line and shaded area) and high range of estimates (red dots, line and shaded area). (d) Absolute error if each of the strategies had been followed exactly by groups with low range (blue) and high range (red). The notation for strategies as in Fig 4, with a final column added in (d) for the error of the observed group consensus estimates. The threshold used to define groups with low and high ranges did not have any effect on the trends in (c) and (d) (S6 Fig). 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).