Dynamical networks of influence in small group discussions
Fig 4
Networks of influence for different group compositions, as they emerged in simulations after t = 100 successive discussions. At the end of each discussion, each group member could update the weights given to all others, based on their relative errors in the previous discussion. Over time, the weights gradually converged to the above values. Here, the composition of the group is varied (a good performer is represented in blue whereas a bad performer is represented in red), as well as the social discounting bias α. When the majority of group members are good performers (i.e., 3 or 2 blue dots), a small bias of α = 0.1 yields the best performances (see Fig 3A and 3B) and produces networks of influence similar to the normative ones (see Fig 2). When the majority of group members are bad performers (i.e., 2 or 3 red dots), larger bias of α = 1 is preferable, producing networks similar to the normative ones.