3.6 Individual Classification Efficiency
To determine the individual classification efficiency of each ERP component, we conducted a Receiver-Operating Characteristic (ROC) analysis that was based on the probe-minus-irrelevant amplitude for each ERP component (N200 at FCz, recognition P300 at Pz, FRN at Fz, and feedback P300 at Pz). We used guilty or innocent group status as the dependent variable. Results (see Table 4) show that all these ERP components, except for the FRN, can effectively distinguish guilty from innocent participants above chance (95% confidence intervals were given in parentheses): AUCs = 0.77 - 0.96, ps < 0.001. The ROC based on the FRN was not significantly different from chance level (AUC = 0.54, p = 0.58). Finally, we examined whether combining all these ERP components might further improve individual classification efficiency. Specifically, probe-minus-irrelevant amplitude for each ERP activity were transformed into standardized z-scores across all guilty and innocent individuals. We then averaged these four z-scores into a single measure for each participant (Hu et al., 2013; Sai et al., 2016). This ROC analysis yielded an AUC of 0.91. And the highest AUC (0.96) was obtained when combining N200, recognition P300 and the feedback P300 (see Table 4). The individual classification efficiency of P200 (AUC = 0.88) and that of P200 in combination with other ERP components (AUC = 0.99) can be found in the supplementary materials (see Table S3).