Figure 2
From: Predicting future conflict between team-members with parameter-free models of social networks
Performance of parameter-free network methods for conflict prediction.
(a) We show the performance of the LR (blue) and the SB (cyan) methods, for conflict appearance and resolution. For conflict appearance we consider the ratio between the number nYY of times that the score of a YY link (positive in surveys I and II) is higher than the score of a YN link (positive in survey I and negative in survey II) in the same team and the number nYN of times the reverse is true. Analogously, for conflict resolution we consider the ratio between the number nNY of times that the score of a NY link is higher than the score of a NN link and the number nNN of times the the reverse is true. We denote these ratios as the normalized prediction performance for the appearance of conflict (nYY/nYN) and for the resolution of conflict (nNY/nNN). To establish the significance of these results, we compare the values of the normalized prediction performance obtained for the SB and LR methods to those of the null model obtained by resampling the scores of all links within each team. We find that the LR method is significantly more accurate than the null model (p = 0.030 for conflict appearance and p = 0.032 for conflict resolution), whereas the SB method is not (p = 0.704 for conflict appearance and p = 0.232 for conflict resolution). (b) We show the overlap of LR and SB methods, for conflict appearance and resolution. The numbers in the figure indicate the number of correctly ranked link pairs nYY and nNY (for conflict appearance and resolution, respectively) for each of the methods LR (blue) and SB (cyan) and for their overlap.
