Table 1.
Results of three case studies testing social network measures as predictors of verdict outcomes.
Fig 1.
Methodological Setup.
Fig 2.
The Working Web of Watergate (n = 61).
The coding was used to map these relationships, generating what the authors [27, 68] call a “Working Web of Watergate”. If two agents (co-conspirators) worked together on illegal activities (i.e., illegal espionage, money laundering, or sabotage) a 1 was entered in the corresponding cell of an adjacency matrix, otherwise a 0 was entered. Thus, the link values in the completed matrix were binary.
Fig 3.
CN communication flow (n = 110).
The data set maps a communication flow between agents indicating who was communicating with whom. Each cell of the adjacency matrix created for the case contains the number of times the corresponding pair of agents communicated with each other. So as not to reveal the identity of the monitored individuals, an identification number was assigned to each. Thus, the link values of the completed matrix were non–negative integers.
Table 2.
Definition of independent variables.
Fig 4.
Confusion matrix for a two–class problem.
TP is the number of correct predictions that an instance is positive (true positive), FN is the number of incorrect predictions that an instance is negative (false negative), FP is the number of incorrect predictions that an instance is positive (false positive) and TN is the number of correct predictions that an instance is negative (true negative).
Table 3.
Performance measures for binary classification problems.
Table 4.
Classifier performance in the WC case.
Fig 5.
ROC curve for Watergate (using LOOCV).
Table 5.
Classifier performance in the CN case.
Fig 6.
ROC curve for Caviar Network (using LOOCV).
Fig 7.
Variable importance by Gini index (means, lower and upper limits of 95% confidence interval) for modelling verdict outcome in WC case.
Fig 8.
Variable importance by Gini index (means, lower and upper limits of 95% confidence interval) for modelling verdict outcome in CN case.