Figure 6: The critical value curves comparison.
From: The impact of variable commitment in the Naming Game on consensus formation
All the experiments are based on networks generated by Barabasi-Albert (BA)17 and Erdos-Renyi (ER) models with parameters N = 1000, m0 = m = 5, p ≈ 0.01, where the average degree of each network is 10, the mean of commitment strengths 
of each distribution is 10. The standard deviation in each case is 0, 2, 5.196 and 16.3 for constant value, and then distributed normally, uniformly and according to power-law commitment strengths. Each critical value is averaged over 100 runs, half runs reaches consensus state A, while the other half reaches consensus state B. For simplicity, we just plot the critical values when pA > pB. Subfigures (a) and (b) show the critical value curves in the waning commitment naming game. pA, pB are the initial fractions of committed agents at state A, B. Subfigures (c) and (d) show the critical value curves in the variable commitment naming game. The variable commitment yields very similar results to increasing commitment, but runs much faster. pA, pB are the initial fractions of agents able to commit at state A, B, which are initialized as uncommitted, but are likely to become committed during interactions. Subfigures (a) and (c) are simulations in BA networks, while subfigures (b) and (d) are simulations in ER networks.
