Dissociable correlates of response conflict and error awareness in error-related brain activity

Abstract

Errors in speeded decision tasks are associated with characteristic patterns of brain activity. In the scalp-recorded EEG, error processing is reflected in two components, the error-related negativity (ERN) and the error positivity (Pe). These components have been widely studied, but debate remains regarding the precise aspects of error processing they reflect. The present study investigated the relation between the ERN and the Pe using a novel version of the flanker task to allow a comparison between errors reflecting different causes-response conflict versus stimulus masking. The conflict and mask conditions were matched for overall behavioural performance but differed in underlying response dynamics, as indexed by response time distributions and measures of lateralised motor activity. ERN amplitude varied in relation to these differing response dynamics, being significantly larger in the conflict condition compared to the mask condition. Furthermore, differences in response dynamics between participants were predictive of modulations in ERN amplitude. In contrast, Pe activity varied little between conditions, but varied across trials in relation to participants' awareness of their errors. Taken together, these findings suggest a dissociation between the ERN and the Pe, with the former reflecting the dynamics of response selection and conflict, and the latter reflecting conscious recognition of an error.

Figure 1

Experimental procedure. Each trial began with a fixation cross followed by a blank screen. Next, a conflict (top), congruent (middle) or mask (bottom) stimulus was presented (note that only one stimulus was actually presented on any given trial). The screen remained blank during the response and error signalling response windows, and then for a varied inter-trial-interval.

Figure 2

Simulation results from the conflict monitoring model. (a) Mean RT and (b) error rates for the congruent, conflict, and mask stimuli. (c) Response activation profiles for conflict and mask conditions, separately for correct and error trials, calculated as the difference in activation levels between correct and incorrect response units (plotted on the y-axis). Positive values indicate greater activation of the correct than incorrect response. (d) Simulated response conflict differences between correct and error trials in the mask and conflict conditions. The x-axis indicates simulated time in panels (c) and (d), with R‘ indicating the time of the response.

Figure 3

Response dynamics. (a) Conditional accuracy as a function of RT, with data divided into twelve bins sorted from fastest (1) to slowest (12), separately for the congruent, mask and conflict conditions. (b) Stimulus-locked ( S‘) LRP for correct trials in the congruent, mask and conflict conditions. (c) Response-Locked ( R‘) LRP for error trials in the mask and conflict conditions.

Figure 4

Response-locked ERP data. ERP waveforms are time-locked to the response ( R‘) for (a) frontocentral and (b) parietal electrode sites, for correct and error trials in the mask and conflict conditions. Topographic distributions of the (c) ERN and (d) Pe components for the conflict condition were calculated as the voltage difference between error and correct trials.

Figure 5

Correlation between the ERN amplitude difference between mask and conflict conditions and the corresponding between-condition difference in (a) error-correcting activity evident in the LRP, (b) objective error rate, and (c) error signalling rate.

Figure 6

Logistic regression classifier results. (a) Cross-validated Az classifier scores for the ERN and Pe component time intervals, compared against classification on randomised data (to establish bootstrap significance level) for each participant (labeled A to T). Error bars indicate standard deviations obtained from 100 permutations of the logistic regression classifier. (b & c) Distribution of single trial classifier classifier output values for the ERN and Pe respectively. These histograms are formed for each participant by calculating the proportion of trials in each bin of width of 0.1 for the four conditions, before averaging together the individual participants. The vertical lines represent the mean of single trial classifier output for each condition. Topographic plots show the scalp projections obtained from the logistic regression classifier for each time window.

Figure 7

Single-trial classifier classifier output values for signalled and unsignalled errors. Classifier output values for signalled (grey circles) and unsignalled (black asterisks) errors are shown for those participants (labelled by letter) with at least 6 unsignalled errors separately for (a) the ERN and (b) the Pe. The horizontal lines represent the mean of the distribution of classifier output values for signalled (grey) and unsignalled (black) errors.