Fig. 2: Predictions of triadic (top) and dyadic (bottom) models for the level of balance (f) and proportions of different triangles after four updating rounds.

Shown are predictions for different number of issues (G = 3, light red or G = 9, light blue), and team sizes (K = 4 or K = 8), equivalent to the conditions in the group experiment. Top: Predictions of the triadic model assuming that each individual updates only one of their triangletime step, like in the group experiment s in each (Q = 1 or more generally \(q=\lceil Q/{N}^{T}\rceil\), where \({N}^{T}\) is the total number of triangles in a group; see Figure S4 for more details). Bottom: Predictions of the dyadic model, assuming high levels of attention to friendly vs. unfriendly team members \(\alpha\), as observed in Szell et al.²⁹. All predictions are shown for different levels of attention to dissonance \(\beta\). Green shaded areas denote moderate-to-high levels of attention that were observed in our experiment. Error bars denote the 95% confidence intervals.