Abstract

In 2 experiments, we used event-related brain potentials (ERPs) to examine the temporal dynamics of neural processes related to adjustments of cognitive control following errors in the counting Stroop task. The ERPs elicited by errors revealed the error-related negativity (ERN) and error positivity consistent with a large body of literature. In addition, errors were associated with a frontal slow wave between 200 and 2000 ms after the response that was consistent with the activity of neural generators in the lateral frontal cortex. The ERN and frontal slow wave were correlated with posterror slowing of response time and positive affect (i.e., happiness and calmness) during task performance. These data are consistent with the idea that interactions between anterior cingulate cortex and lateral frontal cortex support adjustments of cognitive control and that this neural network is sensitive to the influence of affect experienced during task performance.

Introduction

An important question in the area of cognitive neuroscience is related to how the cognitive architecture can maintain optimal information processing in the absence of external feedback. Classic work using choice-response tasks has revealed the existence of an error-processing system that permits the adaptive regulation of cognitive control based on the occurrence of errant responses (Rabbitt 1968; Laming 1979). More recent studies incorporating functional neuroimaging and neuropsychological methodologies have revealed that the error-processing system is supported by a distributed network that includes the anterior cingulate cortex (ACC), dorsolateral prefrontal cortex, and basal ganglia (Carter et al. 1998; Lawrence 2000; Ullsperger and von Cramon 2001; Swick and Turkin 2002; Fiehler et al. 2004). The ACC is thought to play a critical role in error, or more generally conflict, detection (Botvinick et al. 2001; Yeung et al. 2004; Brown and Braver 2005). The dorsolateral prefrontal cortex is thought to support an individual's ability to establish and maintain a representation of goals or task demands that may need to be modified following errors (Carter et al. 2000; Botvinick et al. 2001; Kerns et al. 2004). The current experiments were designed to consider the temporal dynamics of error processing and had 2 primary goals: 1) to examine the timing of the neural processes supported by lateral frontal cortex underlying adjustments of cognitive control in response to errors (Kerns et al. 2004; West and Moore 2005) and 2) to consider the influence of affect on the efficiency of these processes (Gray 2004).

Event-related brain potentials (ERPs) have been used extensively to investigate the temporal dynamics of some of the neural processes underlying error processing. The error-related negativity (ERN) has been commonly associated with error detection (Gehring et al. 1993). The ERN represents a phasic modulation of the ERPs that peaks between 50 and 100 ms after an errant response and is distributed over the midline frontal-central region of the scalp. The ERN is commonly associated with the activity of neural generators located in the ACC (Dehaene et al. 1994; van Veen and Carter 2002), indicating that error detection supported by the ACC is a relatively phasic process (Holroyd and Coles 2002). The error positivity (Pe) reflects a more sustained modulation of the ERPs that follows the ERN and typically peaks between 200 and 400 ms after the response (Gehring et al. 1993). The functional significance of the Pe has not been extensively investigated, although some evidence indicates that it may be associated with error awareness (Overbeek et al. 2005). This account is consistent with the idea that the Pe reflects a response-locked manifestation of the P3 component that is associated with the recognition or categorization of a target stimulus (Kok 2001; Overbeek et al. 2005). The neural generators of the Pe may lie within the posterior cingulate cortex (van Veen and Carter 2002).

In contrast to the large number of studies examining the neural correlates of error detection, far less effort has been devoted to considering the neural correlates of processes underlying adjustments of cognitive control following error detection. Within conflict theory, adaptation to the presence of response conflict associated with errors is thought to arise from interactions between ACC and lateral prefrontal cortex (Botvinick et al. 2001). Specifically, the detection of response conflict by the ACC is thought to result in the generation of a conflict signal that is propagated to the lateral frontal cortex that is in turn responsible for tuning the implementation of cognitive control based on the magnitude of the conflict signal (Botvinick et al. 2001). Evidence from studies employing neuropsychological and functional neuroimaging methodologies is consistent with this premise. Studies of patients with focal lesions indicate that the amplitude of the ERN can be altered by damage to the lateral frontal cortex (Gehring and Knight 2000; Ullsperger et al. 2002). Complimenting these data, Kerns et al. (2004) demonstrated that ACC activity elicited by response conflict on the current trial was positively correlated with activity in lateral prefrontal cortex on the following trial in a study using functional magnetic resonance imaging. Furthermore, ACC activity on the current trial was also predictive of response time measures of adaptive control on the next trial. The findings of Kerns et al. lead to the suggestion that the neural correlates of adjustments of cognitive control supported by lateral frontal cortex should be distributed over the frontal region of the scalp between the errant response and onset of the next stimulus.

The results of work examining the ERP correlates of sequential trial effects (i.e., trial-to-trial changes in response time associated with the presence of response conflict, Botvinick et al. 2001) are consistent with predictions derived from the findings of Kerns et al. (2004). In a recent study, West and Moore (2005) examined modulations of the ERPs that differentiated congruent trials (i.e., trials with low response conflict) and incongruent trials (i.e., trials with high response conflict) in the Stroop task during the response-to-stimulus interval (RSI—the interval between the response on one trial and onset of the stimulus for the next trial) when adjustments of cognitive control would be expected to occur based on conflict theory (Botvinick et al. 2001). These data revealed sustained modulations of the ERPs that began shortly after the response and persisted for at least 1 s thereafter and were distributed over the frontal and parietal regions of the scalp (West and Moore 2005). The finding that adjustments of cognitive control are associated with sustained neural activity between the response and the onset of the next stimulus is interesting within the broader context of work examining the ERP correlates of task and response preparation. An extensive body of literature has revealed that the presentation of an exogenous cue that signals the onset of a stimulus or the timing of a response elicits slow-wave activity (e.g., the contingent negative variation) that is associated with preparatory processing that predicts the speed or accuracy of the response (Loveless and Sanford 1974; Brunia and van Boxtel 2001). When considered within the context of the broader literature related to the ERP correlates of task preparation, the findings of West and Moore (2005) lead to the suggestion that preparatory processing can also be recruited by endogenous sources of information (i.e., response conflict). Together, this body of literature leads to the prediction that processes supporting the updating of cognitive control in response to errors should give rise to modulations of the ERPs that are distributed over the frontal region of the scalp and that persist for one or more seconds after the response.

A growing body of literature indicates that the neural architecture underlying cognitive control and error processing is modulated by mood and emotion (Gray 2004; Hajcak et al. 2004). Work examining the relationship between trait measures of positive and negative emotionality has revealed that the amplitude of the ERN is modulated by individual differences in negative emotionality (Luu et al. 2000) and anxiety (Hajcak et al. 2003). Specifically, the amplitude of the ERN can be greater for individuals high in negative emotionality than for individuals low in negative emotionality (Luu et al. 2000; Hajcak et al. 2004). Other work reveals the influence of state-level modulation of emotion on the efficiency of cognitive control supporting working memory (Gray 2004). For instance, in a study using a mood-induction paradigm, Gray (2001) observed that negative and positive affect modulated the efficiency of working memory, and that the effect of emotion on working memory was mediated by the lateral prefrontal cortex (Gray et al. 2002). Given these findings, it seems reasonable to expect that the ERP correlates of error detection and adjustments of cognitive control following errors should correlate with levels of positive and negative emotion experienced during task performance.

Experiment 1

The first experiment was designed to examine the time course of the ERP correlates of processes that are associated with updating cognitive control in response to errors. In the experiment, individuals completed the counting Stroop task where they were required to indicate the number of digits presented in the display for each trial (Salthouse and Meinz 1995; West et al. 2004). For half of the trials, the number and identity of the digits was congruent (22) and for half of the trials, the number and identity of the digits was incongruent (33). Additionally, for half of the trials the RSI was 500 ms and for the remaining trials the RSI was 2000 ms. Previous research with this task has revealed that intrusion errors (i.e., instances where individuals respond to the identity of the digits rather than the number of digits on incongruent trials) occur on 10–15% of trials (West et al. 2004) making this a good paradigm to examine the ERP correlates of error processing. Also, other work has revealed that this manipulation of the RSI modulates the neural correlates of cognitive control and conflict processing in the counting Stroop task (West and Schwarb 2006). This finding leads to the expectation that varying the RSI may also influence aspects of cognitive control that facilitate error processing.

Several predictions were derived from the findings of previous research examining the ERP correlates of error processing and conflict monitoring. First, intrusion errors were expected to elicit an ERN and Pe. The amplitude of these modulations of the ERPs was expected to be insensitive to RSI because both tend to be resolved by 400–500 ms after the response. Second, intrusion errors were expected to elicit an additional modulation of the ERPs that would be distributed over the frontal region of the scalp if interactions between ACC and lateral frontal cortex serve to instantiate adjustments of cognitive control following errors (Botvinick et al. 2001; Kerns et al. 2004). If adjustments of cognitive control in general require temporally extended processing to be achieved (West and Moore 2005), the time course of the frontal modulation may persist beyond that of the ERN. Furthermore, if adjustments of cognitive control in response to errors are achieved by temporally extended processing, the frontal modulation may be sensitive to the RSI.

Method

Participants

Twenty individuals between the ages of 19 and 23 years participated in the study for course credit. The sample included 9 males and 11 females, and 18 right-handed individuals and 2 left-handed individuals. Data for 1 individual were excluded from the analyses due to high levels of movement artifact in the electroencephalogram (EEG). All participants provided informed consent, and the study was approved by the Human Subjects Institutional Review Board of the university.

Materials and Procedure

The counting Stroop task included 3 phases. The key acquisition phase included 40 trials in which individuals pressed a key mapped to the digits 1–4 (i.e., V—1, B—2, N—3, Y—4) when a string of X's was presented on the screen (e.g., XX—press B). The practice phase included 24 trials (12 congruent and 12 incongruent). Congruent stimuli represented the presentation of a string of digits equal in length to the identity of the digit (e.g., 333) and incongruent stimuli represented the presentation of a string of digits where the number and identity did not match (e.g., 222). The test phase included 20 blocks of 48 trials (24 congruent and 24 incongruent) that were equally divided between the short RSI and the long RSI conditions. In the task, individuals alternated between short and long RSI blocks with half of the participants beginning with a short RSI block and the remaining participants beginning with a long RSI block. In all phases of the task, the stimuli remained on the screen until a response was made. In the key acquisition and practice phases, the next stimulus was presented 500 ms after the response for the previous trial. In the test phase, the next stimulus was presented 500 ms (short RSI) or 2000 ms (long RSI) after the response. The stimuli were presented in the center of a 17″ cathode ray tube computer monitor in light gray on a black background; the screen was then blank during the RSI.

Recording and Analysis of the EEG Data

The EEG (bandpass 0.01–100 Hz, digitized at 256 Hz, gain 2500, 12-bit analog-to-digital conversion) was recorded from an array of 45 tin electrodes sewn into an Electro-cap or affixed to the skin with an adhesive patch that was interfaced to an Isolated Bioelectric Amplifier (James Long Company, Carogo Lake, NY) and a Daqbook/112 (IO Tech, Inc., Cleveland, OH) digitizer. Vertical and horizontal eye movements were recorded from electrodes placed below and beside the eyes. During recording, all electrodes were referenced to electrode Cz; for data analysis, the data were rereferenced to an average reference, electrode Cz was reinstated, and a 30-Hz low-pass filter was applied. Ocular artifacts associated with blinks were corrected using a covariance technique that simultaneously modeled artifact and artifact-free EEG (Sourcesignal Imaging, San Diego, CA). Trials contaminated by other artifacts (peak-to-peak deflections over 100 μV) were rejected before averaging.

ERP analysis epochs were extracted offline and included −300 ms of preresponse activity and 2000 ms of postresponse activity. The baseline reflected mean voltages from −300 to −200 ms before the response. ERPs were averaged for correct congruent trials, correct incongruent trials, and intrusion errors in the short and long RSI conditions. Modulations of the ERPs were quantified in a series of analysis of variances (ANOVAs) performed on measures of mean amplitude using the Huynh–Feldt epsilon adjusted degrees of freedom when necessary (Jennings 1987). The effects of trial type and RSI on 3 modulations of the ERPs were quantified in a series of 2 (RSI: short or long) × 3 (condition: congruent correct, incongruent correct, and intrusion error) × 3 or 5 (electrode) ANOVAs. For the ERN and Pe electrodes, Fz, Cz, and Pz were included in the analyses; for the frontal slow-wave electrodes, Fp1, Fpz, Fp2, Af3, and Af4 were included in the analyses. The amplitude of the ERN was quantified as mean voltage between 70 and 90 ms, the amplitude of the Pe as mean voltage between 250 and 500 ms, and the amplitude of the frontal slow wave as mean voltage between 250 and 500 ms and 1500 and 2000 ms.

Results

The behavioral data for the counting Stroop task were similar to those reported in previous research (West et al. 2004). Response accuracy was quite high for congruent trials and was not sensitive to RSI (short RSI mean = 0.02, standard deviation [SD] = 0.02; long RSI mean = 0.02, SD = 0.02; t(18) = 0.000, P = 1.00); in contrast, there were more intrusion errors in the short RSI condition (mean = 0.13, SD = 0.03) than the long RSI condition (mean = 0.11, SD = 0.02; t(19) = 3.23, P < 0.01). A comparison of response time for correct congruent and correct incongruent trials revealed a main effect of condition (Fig. 1) (F1,18 = 81.96, P < 0.001, η2 = 0.82) with response time being slower for incongruent trials than for congruent trials, and a main effect of RSI (F1,18 = 19.79, P < 0.001, η2 = 0.52) with response time being slower in the long RSI condition than the short RSI condition. The condition × RSI interaction was also significant (F1,18 = 4.63, P < 0.05, η2 = 0.20) with the interference effect being greater in the short RSI interval condition than the long RSI condition. A comparison of response time for correct incongruent trials and intrusion errors revealed a main effect of condition (F1,18 = 49.32, P < 0.001, η2 = 0.73) with response time being faster for intrusion errors than for correct incongruent trials. In this analysis, the condition × RSI interaction was not significant, F < 1.00.

Figure 1.

Mean response time as a function of RSI for correct congruent and incongruent trials and intrusion errors in Experiment 1. The error bars represent the standard error of the mean.

Figure 1.

Mean response time as a function of RSI for correct congruent and incongruent trials and intrusion errors in Experiment 1. The error bars represent the standard error of the mean.

The grand-averaged ERPs for 4 of the midline electrodes are presented in Figure 2. These data reveal 3 modulations of the ERPs that distinguish neural activity elicited by intrusion errors from neural activity elicited by correct responses (ERN, Pe, and frontal slow wave). The time course and topography (Fig. 3) of the ERN and Pe were similar to what is typically observed in the published literature. The frontal slow wave emerged at approximately 200 ms after the response and was broadly distributed over the frontal region of the scalp. In the short RSI condition, the frontal slow wave was maintained well after the onset of the next stimulus, whereas in the long RSI condition the frontal slow wave appeared to be resolved by roughly 500 ms after the response.

Figure 2.

Grand-averaged ERPs at 4 midline electrodes for intrusion errors, incongruent correct trials, and congruent correct trials demonstrating the time course of the ERN (open arrow), Pe (filled arrow), and frontal slow wave (gray arrow). The tall bar reflects the response and the short bars reflect 500-ms increments. The SS marks the onset of the next stimulus in the short RSI condition, and the SL marks the onset of the next stimulus in the long RSI condition.

Figure 2.

Grand-averaged ERPs at 4 midline electrodes for intrusion errors, incongruent correct trials, and congruent correct trials demonstrating the time course of the ERN (open arrow), Pe (filled arrow), and frontal slow wave (gray arrow). The tall bar reflects the response and the short bars reflect 500-ms increments. The SS marks the onset of the next stimulus in the short RSI condition, and the SL marks the onset of the next stimulus in the long RSI condition.

Figure 3.

Difference waves (intrusion error—incongruent correct) for 42 scalp electrodes in the short and long RSI conditions demonstrating the time course and distribution of the ERN, Pe, and frontal slow wave. The ERN is marked by the open arrow, the Pe by the filled arrow, and the frontal slow wave by the gray arrow. The tall bar for the difference waves represents the response.

Figure 3.

Difference waves (intrusion error—incongruent correct) for 42 scalp electrodes in the short and long RSI conditions demonstrating the time course and distribution of the ERN, Pe, and frontal slow wave. The ERN is marked by the open arrow, the Pe by the filled arrow, and the frontal slow wave by the gray arrow. The tall bar for the difference waves represents the response.

The analyses of the ERN and Pe revealed a commonly observed condition × electrode interaction (ERN: F4,72 = 5.19, P < 0.01, η2 = 0.22, ϵ = 0.76; Pe: F4,72 = 15.68, P < 0.001, η2 = 0.47, ϵ = 0.57). A comparison of the ERPs elicited by intrusion errors and correct incongruent trials revealed that the ERN was present at electrodes Fz (F1,18 = 13.51, P < 0.01, η2 = 0.43) and Cz (F1,18 = 9.41, P < 0.01, η2 = 0.34) but not at electrode Pz (F < 1.00). A similar comparison revealed that the Pe was present at electrodes Pz (F1,18 = 19.29, P < 0.001, η2 = 0.52) and Cz (F1,18 = 9.16, P < 0.01, η2 = 0.34); also, at electrode Fz the ERPs elicited by intrusion errors were more negative than the ERPs elicited by correct incongruent trials (F1,18 = 5.01, P < 0.04, η2 = 0.22). This finding may reflect the expression of the frontal slow wave at electrode Fz. Consistent with our predictions, in these analyses the ERN and Pe were insensitive to the RSI, as none of the interactions involving this variable were significant (F < 1.99, P > 0.16).

The effects of condition and RSI on the frontal slow wave were examined in 2 epochs (i.e., 250–500 and 1500–2000 ms after the response). In the first epoch, the stimulus processing demands for the short and long RSI conditions were similar as the next stimulus had not been presented in the short RSI condition. The second epoch allowed us to consider the influence of additional processing associated with the onset of the next stimulus in the short RSI condition on the frontal slow wave. In the first epoch, the main effect of condition was significant (F2,36 = 11.43, P < 0.01, η2 = 0.39, ϵ = 0.60) reflecting greater negativity for intrusion errors than for correct congruent trials (F1,18 = 5.96, P < 0.03, η2 = 0.25) and greater positivity for correct incongruent trials than for correct congruent trials (F1,18 = 22.91, P < 0.001, η2 = 0.56). In this epoch, the condition × RSI interaction was not significant (F < 1.00). This finding indicates that the frontal slow wave was similar in amplitude for the short and long RSI conditions in the first epoch. In the second epoch, the condition × RSI interaction was significant (F2,36 = 4.65, P < 0.03, η2 = 0.21, ϵ = 0.70). In the long RSI condition, there were no significant differences between the task conditions (F < 1.00). In the short RSI condition, the effect of condition was significant (F2,36 = 7.20, P < 0.01, η2 = 0.29, ϵ = 0.64) with the ERPs for intrusion errors being more negative than those elicited by correct congruent trials (F1,18 = 4.40, P < 0.05, η2 = 0.20) and the ERPs for correct incongruent trials were more positive than those elicited by correct congruent trials (F1,18 = 9.18, P < 0.01, η2 = 0.34). The effect of RSI on the amplitude of the frontal slow wave during the second epoch may indicate that updating cognitive control in response to errors is disrupted or prolonged when the information-processing system is faced with the demands of encoding and responding to a new stimulus before the adjustment of control is fully implemented.

To examine the neural generators of the ERN, Pe, and frontal slow wave, a series of spatiotemporal dipole models was fit to the intrusion error minus the correct incongruent difference wave in the short RSI condition where the frontal slow wave was most strongly expressed. The dipole models were fit assuming a 3-shell spherical model of the head using the source module of the EMSE software (Sourcesignal Imaging). In order to isolate the ERN and Pe from slow-wave activity, a 1- to 6-Hz zero phase shift band-pass filter was applied to the difference wave; in fitting the neural generators of the frontal slow wave a . 1- to 6-Hz filter was used so as not to attenuate the amplitude of the frontal slow wave. An initial series of principal component analyses (PCAs) revealed that the first principal component accounted for 97.12%, 94.42%, and 91.38% of the covariance for the ERN, Pe, and frontal slow wave, respectively. The ERN (85–90 ms) was modeled using a single dipole placed in the central midline region (Cartesian coordinates x = 10, y = 10, z = 20; Fig. 4; Dehaene et al. 1994). This model provided a reasonably good fit to the data (residual variance = 9.60%) that was not dramatically improved by moving the dipole 1 or 2 cm in any direction. The central midline dipole revealed a phasic peak at roughly 100 ms after the response that was consistent with the time course of the ERN and a later more sustained modulation during the time window of the Pe. The Pe (340–370 ms) was also modeled using a single dipole placed in the posterior midline region (x = −20, y = −10, z = 30; van Veen and Carter 2002). The posterior midline dipole also revealed a phasic peak at roughly 100 ms after the response followed by a more sustained modulation during the temporal window of the Pe. This model provided a good fit to the data (residual variance = 7.88%) that was not dramatically improved by moving the dipole 1 or 2 cm in any direction.

Figure 4.

Results of the dipole models for the ERN, Pe, and frontal slow wave in the short RSI condition. The upper panel portrays the location of the best fitting dipoles for the 3 modulations superimposed on an average brain and the lower panel portrays the time course of activation for the dipoles.

Figure 4.

Results of the dipole models for the ERN, Pe, and frontal slow wave in the short RSI condition. The upper panel portrays the location of the best fitting dipoles for the 3 modulations superimposed on an average brain and the lower panel portrays the time course of activation for the dipoles.

Because the frontal slow wave has not been characterized in previous research, we chose to rely on the broader literature related to the functional neuroanatomy of error processing and cognitive control in order to establish the starting positions of the dipoles (Carter et al. 1998; Kerns et al. 2004). Given data indicating that the lateral prefrontal cortex is involved in implementing cognitive control and that the tuning of cognitive control may arise from interactions between lateral frontal cortex and ACC, we decided to explore a model consisting of a pair of dipoles placed in the left and right frontal cortex. The model for the frontal slow wave was fit to the mean voltage between 1200 and 1500 ms after the response where the slow wave appeared to be greatest in amplitude. A model with symmetric dipoles placed in the left (x = 30, y = 35, z = 36) and right (x = 30, y = −36, z = 36) lateral frontal regions provided a reasonably good fit to the data (residual variance = 13.09%), given the length of the analyzed epoch, that did not change dramatically when the dipoles were moved 1 or 2 cm in any direction. The coordinates of the lateral frontal dipoles are relatively close to error-related activation within the middle frontal gyrus reported by Carter et al. (1998). The lateral frontal dipoles revealed small phasic responses coinciding with the peak of the ERN followed by more sustained activity that lasted through the analyzed epoch. The time course data for the 4 dipoles used to model the activity of the ERN, Pe, and frontal slow wave lead to the suggestion that error detection is associated with synchronized phasic activity in the cingulate and lateral frontal cortex (Luu et al. 2004) that is followed by more temporally extended activation related to updating cognitive control in the lateral frontal cortex.

In addition to the frontal slow wave that distinguished intrusion errors from correct responses, there was also slow-wave activity over the frontal region of the scalp that distinguished incongruent correct trials from congruent correct trials. Specifically, the ERPs were more negative for congruent correct trials than for incongruent correct trials. Given the ordinal change in voltage between incongruent trials, congruent trials, and intrusion errors, one might wonder whether the difference between congruent correct and incongruent correct trials arises from the activity of the same set of neural generators that are reflected in the intrusion error—incongruent correct difference wave. To consider this question, the dipole model for the frontal slow wave was fit to the congruent correct minus incongruent correct difference wave between 1200 and 1500 ms after the response after a . 1- to 6-Hz filter was applied. When the location and orientation of the dipoles were fixed, the model provided a poor fit to the data (residual variance = 41.85%); when the location of the dipoles was fixed and the orientation allowed to vary, the fit improved but remained rather poor (residual variance = 29.73%). These findings may be taken to indicate that sustained differences between the 3 types of trials during the RSI reflect the activity of distinct neural generators following errors and correct responses.

Experiment 2

The second experiment was designed to provide a replication of the novel findings of Experiment 1 related to the frontal slow wave and to examine the relationship between the frontal slow wave, posterror slowing of response time, and measures of emotion during task performance. In the experiment, the trials were limited to incongruent stimuli in order to increase the number of intrusion errors available for analysis and to allow for the calculation of an index of posterror slowing that was not confounded by trial type. If the frontal slow wave is in fact a neural correlate of the processes underlying adjustments of control, one would expect to observe a significant relationship between the amplitude of the frontal slow wave and posterror slowing (Kerns et al. 2004). Based on previous research demonstrating that emotion modulates the efficiency of cognitive control implemented within ACC and lateral frontal cortex, we predicted that the amplitude of the ERN and frontal slow wave would be correlated with levels of affect experienced during task performance (Luu et al. 2000; Gray et al. 2002).

Method

Participants

Twenty individuals between the ages of 18 and 22 years participated in the study for course credit. The sample included 11 males and 9 females and 17 right-handed individuals and 3 left-handed individuals. All participants provided informed consent, and the study was approved by the Human Subjects Institutional Review Board of the university.

Materials and procedure

The key acquisition phase was identical to Experiment 1 and the practice phase included 24 incongruent trials. In this experiment, the test phase included 10 blocks of 108 incongruent trials that were equally divided between the short and long RSI conditions. Half of the participants performed the short RSI blocks then the long RSI blocks and this order were reversed for the remaining participants. Acquisition and preprocessing of the EEG data were identical to Experiment 1. ERP analysis epochs were extracted offline and included –300 ms of preresponse activity and 2000 ms of postresponse activity for incongruent correct trials and intrusion errors in the short and long RSI conditions.

Current affect was assessed following Gray (2001). For the affect measure, individuals indicated the degree to which their current mood was related to 1 of 8 adjectives (bored, sad, energetic, amused, calm, angry, happy, and anxious) by placing a tick on a scale that represented a 10-cm line bound by the descriptors “not at all” or “extremely” immediately before and after performance of the counting Stroop task. The measure for each emotion represented the distance of the tick from the left edge of the line in millimeters. Responses for the pretest and the difference between the pretest and posttest were included in the analyses.

Partial Least Squares Analysis

Partial least squares (PLS) analysis (McIntosh et al. 1996; Lobaugh et al. 2001) was used to examine brain–behavioral relations between ERP amplitude, posterror slowing, and measures of affect. PLS is a multivariate data analytic technique similar to PCA that allows one to decompose the time course and topography of the ERPs into latent variables that reflect correlations between ERP amplitude and behavioral indices (e.g., posterror slowing) in the current application. PLS differs from PCA in that it utilizes a constrained covariance matrix rather than the full covariance matrix. For the reported analyses, the number of latent variables was equal to the product of the number of task conditions and the number of behavioral measures that were included in the analysis. PLS analysis has been used successfully to examine brain–behavior relations in a number of ERP studies examining the neural correlates of cognitive control (West and Schwarb 2006) and prospective memory (West et al. 2006). The findings of these studies indicate that PLS analysis provides a statistically elegant means of testing our predictions related to the relationship between the amplitude of the ERN, Pe, and frontal slow wave and behavioral indices of error processing and affect.

PLS analysis was applied to an ERP data matrix containing subjects and conditions in the rows and the amplitudes for all time points and channels in the columns. The input matrixes for the PLS analyses were formed by calculating the within-condition correlations of the behavioral measures and ERP amplitude across time and space; for the analysis, the individual correlation matrices were then stacked into a single data matrix. Singular value decomposition (SVD) was performed on the correlation matrix to identify the structure of the latent variables. Three outputs were obtained from the SVD that were used to interpret the brain–behavior relationships. The first was a vector of singular values that represents the unweighted magnitude of each latent variable and can be used to calculate the proportion of the cross-block covariance matrix (i.e., the percentage of behavior-related variance) attributable to each latent variable. The singular values are similar to eigen values in PCA. The second and third outputs contain the structure of the latent variables and are orthogonal pairs of vectors (saliences). One vector (behavior saliences) defines the strength of the ERP–behavior correlations. The other vector (ERP saliences) identifies where, in time and space, the effects in the behavior saliences for each latent variable are expressed on the scalp (i.e., time points and spatial locations where ERP amplitude and behavior are reliably correlated). The ERP saliences are similar to component loadings in PCA. The significance of the latent variables' singular values was determined through a permutation test (500 replications) that provided an exact probability of observing the singular value by chance (e.g., P = 0.001); the stability of the ERP saliences at each time point and location on the scalp was established through bootstrap resampling (500 replications) that provides a standard error (SE) for each of the saliences. The ratio of the salience to its bootstrapped SE is approximately equivalent to a z score; therefore, bootstrap ratios greater than 2.0 can be taken to indicate stable saliences or time points that differ from zero at the P ≤ 0.05 level. In the figures presenting the results of the PLS analyses, stable saliences are indicated by a o placed above the time point in the electrode plots. Bootstrap resampling was also used to calculate the 95% confidence interval (CI) around the obtained ERP–behavior correlations. The correlation was considered significant if the CI did not include zero. Matlab code to perform PLS analysis can be obtained at http://www.rotman-baycrest.on.ca.

Results

The response accuracy data were consistent with those of Experiment 1 with intrusion errors being more frequent in the short RSI condition, mean = 0.15, SD = 0.04, than the long RSI condition, mean = 0.12, SD = 0.02, t(19) = 3.28, P < 0.01. A comparison of response time for intrusion errors and correct trials failed to reveal a main effect of trial (F1,19 = 2.41, P > 0.12, η2 = 0.11) or a trial × RSI interaction (F1,19 = 1.20, P > 0.28, η2 = 0.06). This finding was surprising given the results of Experiment 1 wherein response time for intrusion errors was faster than response time for incongruent correct trials. In light of this, we performed separate analyses for the 2 RSI conditions. For the short RSI condition, response time was faster for intrusion errors, mean = 503 ms, SD = 96, than for correct trials, mean = 546 ms, SD = 85, t(19) = 5.15, P < 0.001; in contrast, for the long RSI condition there was little difference in response time for intrusion errors, mean = 593 ms, SD = 159, and correct trials, mean = 598 ms, SD = 106, t(19) = 0.15, P > 0.88. A comparison of response time for trials following intrusion errors and for trials following correct responses revealed a main effect of trial (F1,19 = 7.40, P < 0.014, η2 = 0.28) with response time being slower for trials following intrusion errors, mean = 636 ms, SD = 153, than for the trials following correct responses, mean = 597 ms, SD = 130. This finding demonstrates significant posterror slowing. In the analysis of posterror slowing, the trial × RSI interaction was not significant, F < 1.00.

The pre- and posttest scores for the measures of affect are presented in Table 1. These data reveal significant changes across the course of the task for 4 variables (bored, energetic, amused, and happy). Levels of boredom increased during the task, whereas the levels of energy, amusement, and happiness decreased during the task. The degree of sadness and anger was relatively low across the testing session, the level of calmness relatively high across the testing session, and the level of anxiety was at an intermediate level.

Table 1

Mean ratings in millimeters for emotion adjectives at the pre- and posttests

  Pretest Posttest t P 
Bored Mean 35 63 5.21 0.001 
SD 19 26 
Sad Mean 11 −1.01 0.32 
SD 10 15 
Energetic Mean 42 29 2.80 0.02 
SD 25 24 
Amused Mean 61 37 5.09 0.001 
SD 28 27 
Calm Mean 73 80 −1.97 0.06 
SD 22 22 
Angry Mean 11 −1.76 0.10 
SD 13 17 
Happy Mean 64 54 2.24 0.04 
SD 24 23 
Anxious Mean 35 25 1.48 0.16 
SD 21 23 
  Pretest Posttest t P 
Bored Mean 35 63 5.21 0.001 
SD 19 26 
Sad Mean 11 −1.01 0.32 
SD 10 15 
Energetic Mean 42 29 2.80 0.02 
SD 25 24 
Amused Mean 61 37 5.09 0.001 
SD 28 27 
Calm Mean 73 80 −1.97 0.06 
SD 22 22 
Angry Mean 11 −1.76 0.10 
SD 13 17 
Happy Mean 64 54 2.24 0.04 
SD 24 23 
Anxious Mean 35 25 1.48 0.16 
SD 21 23 

The grand-averaged ERP data portraying the ERN, Pe, and frontal slow wave are presented in Figures 5 and 6. In these data, the frontal slow wave and possibly the ERN appear to be sensitive to RSI, whereas the Pe does not appear to be influenced by RSI. These data were quantified in a series of ANOVAs similar to those employed in Experiment 1.

Figure 5.

Grand-averaged ERPs at 4 midline electrodes for intrusion errors and incongruent trials demonstrating the time course of the ERN (open arrow), Pe (filled arrow), and frontal slow wave (gray arrow) in Experiment 2. The tall bar reflects the response and the short bars reflect 500-ms increments.

Figure 5.

Grand-averaged ERPs at 4 midline electrodes for intrusion errors and incongruent trials demonstrating the time course of the ERN (open arrow), Pe (filled arrow), and frontal slow wave (gray arrow) in Experiment 2. The tall bar reflects the response and the short bars reflect 500-ms increments.

Figure 6.

Grand-averaged ERPs at electrodes Cz and Fc1 demonstrating the ERN in Experiment 2. The tall bar reflects the response.

Figure 6.

Grand-averaged ERPs at electrodes Cz and Fc1 demonstrating the ERN in Experiment 2. The tall bar reflects the response.

The analysis of the ERN revealed 2 interesting effects. The condition × electrode interaction was significant (F2,38 = 5.79, P < 0.02, η2 = 0.23, ϵ = 0.66) with the ERN being present at electrodes Fz (F1,19 = 6.37, P < 0.03, η2 = 0.25) and Cz (F1,19 = 9.12, P < 0.01, η2 = 0.32) and not electrode Pz (F < 1.00). The condition × RSI interaction was also significant (Fig. 6) (F1,19 = 5.82, P < 0.03, η2 = 0.24) with the amplitude of the ERN being greater in the short RSI condition, mean = 2.22 μV, than the long RSI condition, mean = 0.71 μV, at electrode Cz. Additionally, for the long RSI condition, the amplitude of the ERN was not significantly different from zero, t(19) = 1.17, P = 0.26, at electrode Cz. However, further analysis of the data for electrode Fc1 revealed that the amplitude of the ERN did differ from zero, mean = 1.74 μV, t(19) = 2.40, P = .03, in the long RSI condition, and that the condition × RSI interaction was not significant at this electrode (Fig. 6) (F1,19 = 1.35, P = 0.26, η2 = 0.07). This finding indicates that the ERN was robust in both RSI conditions and that the effect of RSI on the amplitude of the ERN varied between the frontal-central and central electrode locations.

In the analysis of the Pe, the condition × electrode interaction was significant (F2,38 = 9.91, P < 0.001, η2 = 0.34, ϵ = 0.84) with the Pe being expressed at electrodes Pz (F1,19 = 11.39, P < 0.01, η2 = 0.38) and Cz (F1,19 = 6.31, P < 0.03, η2 = 0.25) but not electrode Fz (F < 1.00). The condition × RSI interaction was not significant, F < 1.00, indicating that the Pe was insensitive to RSI.

As can be seen in Figure 5, the frontal slow wave is clearly present in the short RSI condition and was markedly attenuated in the long RSI condition. The analysis of the 250- to 500-ms epoch revealed a main effect of condition (F1,19 = 5.86, P < 0.03, η2 = 0.24) and a condition × RSI interaction (F1,19 = 7.44, P < 0.02, η2 = 0.28). The interaction reflected the presence of the frontal slow wave in the short RSI condition (F1,19 = 12.28, P < 0.01, η2 = 0.39) and not the long RSI condition, F < 1.00. The analysis of the 1500- to 2000-ms epoch revealed only a condition × RSI interaction (F1,19 = 4.38, P < 0.05, η2 = 0.19) with the frontal slow wave being present in the short RSI condition (F1,19 = 4.18, P < 0.05, η2 = 0.18) but not the long RSI condition, F < 1.00. These data replicate the results of Experiment 1 for the short RSI condition; in contrast, for the long RSI condition mean differences between intrusion errors and correct trials for the frontal slow wave were not observed in Experiment 2.

A set of PLS analyses was performed to examine brain–behavior relationships between the neural correlates of error processing (ERN, Pe, frontal slow wave) and behavioral measures related to error processing and affect. The ERP data for these analyses represented the intrusion error minus incongruent correct trials difference waves for the short and long RSI conditions (0–2000 ms after the response) at a subset of the electrodes that express the ERN, Pe, and frontal slow wave (Pz, P3, P4, Cz, Fc1, Fc2, Fz, Af3, Af4, Fpz, Fp1, and Fp2). The behavioral measures represented the proportion of errors that were committed, the degree of posterror slowing (mean response time for trials following errors minus mean response time for trials following correct responses), the values for the measures of affect at the beginning of the study, and the degree of change in the measures of affect during task performance (e.g., happy posttask minus happy pretask).

The analyses of the cognitive variables revealed 2 interesting findings. First, the proportion of intrusion errors was not significantly related to ERP amplitude (P = 0.09). Second, the degree of posterror slowing was related to ERP amplitude (Fig. 7). In the analysis including posterror slowing, the first latent variable was significant, P = 0.026, and accounted for 65% of the covariance. This latent variable reflected positive correlations between ERP amplitude and posterror slowing in the short and long RSI conditions. The electrode saliences were stable over the frontal-central region between 200 and 500 ms after the response (Fig. 7, electrodes Fz and Cz) and the frontal-polar region during the period of the frontal slow wave (Fig. 7, electrode Fpz). These findings indicate that greater posterror slowing was associated with the amplitude of a late component of the ERN (van Veen and Carter 2002) and the frontal slow wave. These data are consistent with the idea that interactions between ACC and lateral frontal cortex support adjustments of cognitive control that give rise to posterror slowing of response time (Kerns et al. 2004).

Figure 7.

Results of the PLS analysis (behavior saliences and ERP saliences at limited electrodes) examining brain–behavior relationships between posterror slowing and ERP amplitude. Note the early modulation at electrode Cz during the time course of the ERN and the more sustained modulation at electrode Fpz during the time course of the frontal slow wave. The bars for the behavior saliences indicate 95% CIs and the o′s above the x axis mark stable ERP saliences where the bootstrap ratio exceeds 2.0.

Figure 7.

Results of the PLS analysis (behavior saliences and ERP saliences at limited electrodes) examining brain–behavior relationships between posterror slowing and ERP amplitude. Note the early modulation at electrode Cz during the time course of the ERN and the more sustained modulation at electrode Fpz during the time course of the frontal slow wave. The bars for the behavior saliences indicate 95% CIs and the o′s above the x axis mark stable ERP saliences where the bootstrap ratio exceeds 2.0.

A set of 8 PLS analyses revealed significant brain–behavior relationships between 3 of the measures of affect and ERP amplitude (first latent variables—happy, P = 0.001; calm, P = 0.002; and bored, P = 0.052). The first latent variable for happy accounted for 72% of the covariance and reflected positive correlations between ERP amplitude and the level of happiness before the task began and negative correlations between ERP amplitude and changes in happiness during task performance (Fig. 8). As can be seen in the left panel of Figure 8, this latent variable revealed stable electrode saliences at Cz, Fz, and Fpz electrodes during the time course of the ERN and frontal slow wave. In addition, this latent variable did not reveal any stable saliences at the electrodes Pz indicating that this effect was not related to the Pe. These findings indicate that the amplitude of the ERN and frontal slow wave was greater for individuals who were happier at the beginning of the task and who experienced the greatest reduction in happiness during task performance.

Figure 8.

Results of the PLS analyses (behavior saliences and ERP saliences at limited electrodes) examining brain–behavior correlations between levels of happiness, calmness, and boredom and ERP amplitude. For happy note, the early modulation at electrode Cz during the time course of the ERN and the more sustained modulation at electrodes Fz and Fpz during the time course of the frontal slow wave; for calm note, the modulation at electrode Fz during the time course of the ERN and at electrode Fpz during the time course of the frontal slow wave; for bored note, the early modulation at electrode Fz during the time course of the frontal slow wave, the later modulation at electrodes Cz and Pz during the time course of the Pe, and the lack of a sustained modulation at electrodes Fz or Fpz during the time course of the frontal slow wave. The bars for the behavior saliences indicate 95% CIs and the o′s above the x axis mark stable ERP saliences where the bootstrap ratio exceeds 2.0.

Figure 8.

Results of the PLS analyses (behavior saliences and ERP saliences at limited electrodes) examining brain–behavior correlations between levels of happiness, calmness, and boredom and ERP amplitude. For happy note, the early modulation at electrode Cz during the time course of the ERN and the more sustained modulation at electrodes Fz and Fpz during the time course of the frontal slow wave; for calm note, the modulation at electrode Fz during the time course of the ERN and at electrode Fpz during the time course of the frontal slow wave; for bored note, the early modulation at electrode Fz during the time course of the frontal slow wave, the later modulation at electrodes Cz and Pz during the time course of the Pe, and the lack of a sustained modulation at electrodes Fz or Fpz during the time course of the frontal slow wave. The bars for the behavior saliences indicate 95% CIs and the o′s above the x axis mark stable ERP saliences where the bootstrap ratio exceeds 2.0.

The first latent variable for calm accounted for 61% of the covariance and reflected a positive correlation between ERP amplitude and the degree of calmness before the task began for the short RSI condition and negative correlations between ERP amplitude and changes in calmness over the course of the task for both RSI conditions. As can be seen in the middle panel of Figure 8, the stable electrode saliences for this latent variable reflected an early effect at Fz during the time course of the ERN and a sustained effect from 0 to 1000 ms after the response at Fpz during the time course of the frontal slow wave. As was the case for happiness, calmness was not related to the amplitude of the Pe. These findings indicate that the amplitude of the ERN and frontal slow wave was greater for those individuals who were more calm at the outset of the task in the short RSI condition and somewhat greater for those individuals who became less calm over the course of the task.

The first latent variable for bored accounted for 48% of the covariance and reflected positive correlations between ERP amplitude and the level of boredom before the task began and negative correlations between ERP amplitude and increases in boredom over the course of the task. As can be seen in the right panel of Figure 8, the stable electrode saliences for this latent variable reflected an early effect at the electrode Fz that captured the time course of the ERN and a later effect at electrode Pz during the time course of the Pe. This latent variable does not appear to reflect the frontal slow wave given the near absence of stable electrode saliences at Fpz. These findings indicate that the amplitude of the ERN and Pe was greater for those individuals who were least bored at the beginning of the task and was also greater for those individuals whose level of boredom changed the least over the course of task performance. Together, the results of these PLS analyses may indicate that different components of the error-monitoring system are sensitive to different aspects of emotion or mood, with structures supporting error detection (ERN) and adjustments of control (frontal slow wave) being sensitive to positive affect and structures supporting error awareness (ERN, Pe) being sensitive to boredom.

Given the similarities in the pattern of stable electrode saliences for the analyses including posterror slowing, happiness, and calmness, we wondered whether there might be a common relationship between these 2 measures of affect, the ERN and frontal slow wave, and posterror slowing. To examine this question, we performed a PLS analysis that included these 3 behavioral measures. In this analysis, the first latent variable was significant, P = 0.001, and accounted for 58% of the covariance. The latent variable reflected positive correlations between ERP amplitude and levels of happiness and calmness (for the short RSI condition) at the beginning of the task and negative correlations between ERP amplitude and changes in happiness and calmness, and posterror slowing (Fig. 9). The stable electrode saliences reflected an early effect at the Cz that captured the time course of the ERN and a sustained effect at Fz and Fpz electrodes that captured the time course of the frontal slow wave. Additionally, none of the electrode saliences was stable for the Pz indicating that this latent variable did not express the Pe. These finding are consistent with a growing body of literature revealing a functional relationship between levels of positive affect, in this case level of happiness and calmness, and the efficiency of neural networks supporting the implementation of cognitive control (Gray 2004).

Figure 9.

Results of the PLS analysis (behavior saliences and ERP saliences at limited electrodes) examining brain–behavior relationships between happiness, calmness, and posterror slowing and ERP amplitude. Note, the early modulation at electrode Cz during the time course of the ERN and the more sustained modulation at electrodes Fz and Fpz during the time course of the frontal slow wave. The bars for the behavior saliences indicate 95% CIs and the o′s above the x axis mark stable ERP saliences where the bootstrap ratio exceeds 2.0.

Figure 9.

Results of the PLS analysis (behavior saliences and ERP saliences at limited electrodes) examining brain–behavior relationships between happiness, calmness, and posterror slowing and ERP amplitude. Note, the early modulation at electrode Cz during the time course of the ERN and the more sustained modulation at electrodes Fz and Fpz during the time course of the frontal slow wave. The bars for the behavior saliences indicate 95% CIs and the o′s above the x axis mark stable ERP saliences where the bootstrap ratio exceeds 2.0.

General Discussion

In 2 experiments, we have examined the ERP correlates of error processing in the counting Stroop task. Consistent with a large literature (reviewed by Ullsperger and van Cramon 2004), intrusion errors elicited an ERN and Pe in both experiments. In addition, intrusion errors elicited a slow wave that was distributed over the frontal region of the scalp. Although the frontal slow wave has not been characterized in previous research, its presence was anticipated based on the neural architecture underlying error processing within conflict theory (Botvinick et al. 2001; Kerns et al. 2004; Yeung et al. 2004) and the findings of work examining the ERP correlates of adjustments of cognitive control and task preparation (Brunia and van Boxtel 2001; West and Moore 2005).

The frontal slow wave began around 200 ms after the response and was reasonably well fit by a pair of dipoles placed in the lateral frontal region, an area that is known to activate following errant responses (Carter et al. 1998). The amplitude of the frontal slow wave was modulated by the RSI. For the short RSI condition, the frontal slow wave persisted for several hundred milliseconds after the onset of the next stimulus in both experiments; for the long RSI condition, the frontal slow wave was resolved by roughly 500 ms after the response in Experiment 1 and was not present in Experiment 2. The effect of RSI on the frontal slow wave leads to 2 suggestions. First, this effect may indicate that adjustments of cognitive control supported by lateral frontal cortex require several hundred milliseconds to be fully implemented (West and Moore 2005). Second, this effect may indicate that the processes underlying adjustments of cognitive control compete for access to a common pool of attentional resources that is taxed when the cognitive system is required to engage processes associated with stimulus encoding and response selection in parallel with processes involved in updating cognitive control. In contrast, the long RSI condition may have allowed individuals to adopt a task set where these 2 activities could be completed in a more sequential manner given the relatively slow pacing of the task. In order to test this idea, it might be possible to examine the effects of stimulus congruity for the N + 1 trial on the frontal slow wave in the short RSI condition. If the persistence of the frontal slow wave is dependent on demands placed on response selection, it might be greater for incongruent than congruent trials after the onset of the next stimulus. Unfortunately, Experiment 1 of the current study did not produce a sufficient number of trials to test this hypothesis in the present data.

A set of PLS analyses examining brain–behavior relationships between the neural correlates of error processing and measures of affect during task performance revealed several interesting findings. Levels of happiness, calmness, and boredom at the beginning of the task and changes in the levels of these emotions during task performance appeared to be differentially related to the ERN, Pe, and frontal slow wave. The amplitude of the ERN and frontal slow wave was correlated with levels of happiness and calmness, whereas the ERN and Pe was correlated with levels of boredom. These findings may indicate that processes underlying error detection, error awareness, and adjustments of cognitive control in response to errors are differentially sensitive to various components of affect. The pattern of correlation for the initial level of happiness and calmness—for the short but not the long RSI condition—was similar, revealing that the amplitude of the ERN and frontal slow wave was greater for those individuals who were happier and calmer at the beginning of the task. These findings may indicate that positive affect had a facilitative effect on the neural processes underlying error detection and adjustments of cognitive control.

The influence of positive affect on the neural correlates of error processing is interesting within the context of other work where negative rather than positive affect has been associated with the amplitude of the ERN (Luu et al. 2000; Hajcak et al. 2004). There are at least 2 factors that may have contributed to the differences observed across studies. First, levels of negative affect (sadness and anger) were relatively low at the beginning of the study, and there was very little change in these emotions over the course of task performance. This restriction of range could lead to an attenuation of the correlation between negative affect and the neural correlates of error processing in the current study. Second, the nature of the assessment of affect may have led to differences in our findings and those of earlier work. Our assessment of emotion asked individuals to report their current level of affect (i.e., state level mood), whereas in previous work (Luu et al. 2000; Hajcak et al. 2004) individuals were asked to report their level of affect over the last 2 weeks (i.e., trait level mood). Although there is generally good correspondence between reports of affect over these intervals (Watson et al. 1988), the current data may reveal differential influences of state and trait emotion on the efficiency of processes underlying error processing. One means of examining this possibility would be to examine the neural correlates of error processing in groups of individuals who vary in their trait levels of positive and negative affect while also considering state levels of affect during task performance. We have adopted this approach in a new study examining the effects of depression on the ERP correlates for error processing that should provide both longer term and transient measures of emotional liability.

Examining the relationship between changes in happiness and calmness during the course of task performance, and the amplitude of the ERN and frontal slow wave revealed that the amplitude of these 2 modulations of the ERPs was the greatest for those individuals who experienced the most extensive decrease in levels of happiness and calmness over the course of task performance. The finding that changes in affect are correlated with the expression of the neural correlates of error processing are generally consistent with a diverse body of evidence. As described in the Introduction, Gray et al. (2002) have demonstrated that the induction of positive and negative mood can influence the recruitment of lateral frontal cortex supporting the implementation of cognitive control as related to working memory. These findings are also consistent with evidence from literature examining the relationship between emotion regulation, executive attention, and social development (see Fox et al. 2001). Within this literature, investigators have consistently demonstrated that individuals who are better able to regulate emotion in response to behavioral challenges demonstrate higher levels of executive control during the performance of cognitive tasks and in the context of social interactions.

One question that may strike the reader is why the frontal slow wave has not been characterized in previous research. We believe that there are a number of reasons why this modulation of the ERPs may not have been observed in prior studies. First, most of the literature using ERPs to examine the neural correlates of error processing has focused on the ERN and Pe, with far more attention being directed toward exploring the characteristics of the ERN. Given this, investigators have only tended to focus on the ERPs elicited in the first few hundred milliseconds after the response. Because the frontal slow wave does not emerge until 200–300 ms after the response, it is not surprising that this modulation has not attracted significant attention in the published literature. A second factor that may have obscured the frontal slow wave in previous research represents the use of a relatively long intertrial interval (i.e., 1500–3000 ms; see Luu et al. 2000; Gehring and Fencsik 2001; van Veen and Carter 2002). In contrast to this practice, the frontal slow wave was the greatest in amplitude in the short RSI condition in our experiments. Related to this issue, we are not aware of a previous study where the response-locked ERPs have been examined after the onset of the next stimulus, where the amplitude of the frontal slow wave was the greatest in the short RSI condition in our experiments. Finally, it may be the case that the use of a relatively conservative high-pass filter (e.g., 1 Hz) served to obscure the presence of the frontal slow wave in some prior studies examining the ERP correlates of error processing. Regardless of the reasons for the failure to observe the frontal slow wave in previous research, the current experiments reveal that this modulation of the ERPs can be reliably evoked and meaningfully related to theoretically motivated behavioral indices of error processing and measures of affect.

The rational for the current experiments was grounded in the conflict theory of error processing wherein the detection of response conflict by the ACC leads to updating of cognitive control by lateral frontal cortex (Botvinick et al. 2001; Kerns et al. 2004; Yeung et al. 2004). The findings of the 2 experiments were generally consistent with predictions derived from conflict theory as the ERN and frontal slow wave were related to posterror slowing and measures of positive affect. However, conflict theory is not the only account of the functional role of the ACC in error processing that has been offered in recent years. The ERN has also been interpreted as an index of reinforcement learning that arises when behavioral outcomes are worse than expected (Gehring et al. 1993; Holroyd and Coles 2002; Brown and Braver 2005). Brown and Braver (2005) sought to integrate ideas from the conflict monitoring and reinforcement learning accounts of the ERN in a study demonstrating that the ACC is generally responsive to the likelihood of negative outcomes, be they in the form of response conflict or negative reinforcement. An important feature of the integrated framework for the current discussion is that this model retains the critical feature of output from the ACC being used to tune control settings outside the ACC.

The ERN has also been associated with affective aspects of error processing (Luu et al. 2000). This idea seems compatible with more recent developments within goal maintenance theory wherein the ACC and lateral frontal cortex have been conceptualized to support the integration of cognitive and affective aspects of information processing (Gray et al. 2002; Gray 2004). The current data are consistent with this general idea and may indicate that various components of the error-processing system are differentially sensitive to diverse emotions. For instance, in our data the ERN and frontal slow wave were modulated by variation in positive affect (happiness and calmness); in contrast, the Pe appeared to be sensitive to levels of boredom, but revealed little if any sensitivity to levels of positive affect. The influence of boredom on the Pe is consistent with recent evidence revealing that the Pe is attenuated by sleepiness (Murphy et al. 2006).

The current findings compliment those from a growing number of studies demonstrating the utility of ERPs in examining the temporal dynamics of the action-monitoring system. In addition to providing insight into processes associated with error detection and updating of cognitive control in response to errors as was the case in the current study, ERPs have also been used to study patterns of neural activity that appear to predict or foreshadow the occurrence of errors (West and Alain 2000). For instance, 2 studies have characterized a modulation of the ERPs labeled the error-preceding positivity that is thought to provide an index of disruptions of the action-monitoring system that lead to the occurrence of errors (Ridderinkhof et al. 2003; Hajcak et al. 2005). The current data, together with those of recent studies, indicate that ERPs can be used to track both the neural events that lead to errors in addition to the neural events that allow the information-processing system to recover from errant behavior. One possible line of future inquiry may be to explore the generality of the processes underlying action monitoring as most existing studies have utilized a relatively limited set of stimulus-response compatibility paradigms to examine the neural correlates of error processing.

Conclusions

In this study, we have characterized a previously unreported frontal slow wave that is associated with error processing. The frontal slow wave was tightly coupled to the ERN in the PLS analyses consistent with the idea that interactions between ACC and lateral frontal cortex serve to implement adjustments of cognitive control elicited by errant responses (Botvinick et al. 2001; Kerns et al. 2004; Yeung et al. 2004). The frontal slow wave persisted well after the onset of the next stimulus in the short RSI condition, indicating that implementing an adjustment of cognitive control may be a rather slow process relative to the time course of error detection reflected in the phasic activity of the ERN. The amplitude of the ERN and frontal slow wave was correlated with measures of positive affect (happiness and calmness) at the outset of the task and changes in the level of these emotions over the course of task performance. These data extend recent work in the area of working memory, revealing interactions between emotion and the efficient implementation of cognitive control (Gray 2004). Finally, dipole modeling of the frontal slow wave activity revealed that distinct regions of prefrontal cortex may be recruited to support adjustments of control in response to errors and in response to the presence of response conflict on correct trials.

Conflict of Interest: None declared.

References

Botvinick
MM
Braver
TS
Barch
DM
Carter
CS
Cohen
JD
Conflict monitoring and cognitive control
Psychol Rev
 , 
2001
, vol. 
108
 (pg. 
624
-
652
)
Brown
JW
Braver
TS
Learned predictions of error likelihood in the anterior cingulate cortex
Science
 , 
2005
, vol. 
307
 (pg. 
1118
-
1121
)
Brunia
CHM
van Boxtel
GJM
Wait and see
Int J Psychophysiol
 , 
2001
, vol. 
43
 (pg. 
59
-
75
)
Carter
CS
Braver
TS
Barch
DM
Botvinick
MM
Noll
D
Cohen
JD
Anterior cingulate cortex, error detection, and the online monitoring of performance
Science
 , 
1998
, vol. 
280
 (pg. 
747
-
749
)
Carter
CS
MacDonald
AM
III
Botvinick
M
Ross
LL
Stenger
VA
Noll
D
Cohen
JD
Parsing executive processes: strategic vs. evaluative functions of the anterior cingulate cortex
Proc Natl Acad Sci USA
 , 
2000
, vol. 
97
 (pg. 
1944
-
1948
)
Dehaene
S
Posner
MI
Tucker
DM
Localization of a neural system for error detection and compensation
Psychol Sci
 , 
1994
, vol. 
5
 (pg. 
303
-
305
)
Fiehler
K
Ullsperger
M
van Cramon
DY
Neural correlates of error detection and error correction: is there a common neuroanatomical substrate?
Eur J Neurosci
 , 
2004
, vol. 
19
 (pg. 
3081
-
3087
)
Fox
NA
Henderson
HA
Marshall
PJ
Nelson
CA
Luciana
M
The biology of temperament: an integrative approach
The handbook of developmental cognitive neuroscience
 , 
2001
Cambridge (MA)
The MIT Press
(pg. 
631
-
645
)
Gehring
WJ
Fencsik
DE
Functions of the medial frontal cortex in the processing of conflict and errors
J Neurosci
 , 
2001
, vol. 
21
 (pg. 
9430
-
9437
)
Gehring
WJ
Goss
B
Coles
MGH
Meyer
DE
Donchin
E
A neural system for error detection and compensation
Psychol Sci
 , 
1993
, vol. 
4
 (pg. 
385
-
390
)
Gehring
WJ
Knight
RT
Prefrontal-cingulate interactions in action monitoring
Nat Neurosci
 , 
2000
, vol. 
3
 (pg. 
516
-
520
)
Gray
JR
Emotional modulation of cognitive control: approach-withdrawal states double dissociate spatial from verbal two-back task performance
J Exp Psychol Gen
 , 
2001
, vol. 
130
 (pg. 
436
-
452
)
Gray
JR
Integration of emotion and cognitive control
Curr Dir Psychol Sci
 , 
2004
, vol. 
13
 (pg. 
46
-
48
)
Gray
JR
Braver
TS
Raichle
ME
Integration of emotion and cognition in the lateral prefrontal cortex
Proc Natl Acad Sci USA
 , 
2002
, vol. 
99
 (pg. 
4115
-
4120
)
Hajcak
G
McDonald
N
Simons
RF
Anxiety and error-related brain activity
Biol Psychol
 , 
2003
, vol. 
64
 (pg. 
77
-
90
)
Hajcak
G
McDonald
N
Simons
RF
Error-related psychophysiology and negative affect
Brain Cogn
 , 
2004
, vol. 
56
 (pg. 
189
-
197
)
Hajcak
G
Nieuwenhuis
S
Ridderinkhof
KR
Simons
RF
Error-preceding brain activity: robustness, temporal dynamics, and boundary conditions
Biol Psychol
 , 
2005
, vol. 
70
 (pg. 
67
-
78
)
Holroyd
CB
Coles
MGH
The neural basis of human error processing: reinforcement learning, dopamine, and the error-related negativity
Psychol Rev
 , 
2002
, vol. 
109
 (pg. 
679
-
709
)
Jennings
JR
Editorial policy on analyses of variance with repeated measures. Psychophysiology
1987
, vol. 
24
 (pg. 
474
-
475
)
Kerns
JK
Cohen
JD
MacDonald
AW
III
Cho
RY
Stenger
VA
Carter
CS
Anterior congulate conflict monitoring and adjustments in control
Science
 , 
2004
, vol. 
303
 (pg. 
1023
-
1026
)
Kok
A
On the utility of P3 amplitude as a measure of processing capacity
Psychophysiology
 , 
2001
, vol. 
38
 (pg. 
557
-
577
)
Laming
DRJ
Autocorrelation of choice-reaction times
Acta Psychol (Amst)
 , 
1979
, vol. 
43
 (pg. 
381
-
412
)
Lawrence
AD
Error correction and the basal ganglia: similar computations for action, cognition and emotion?
Trends Cogn Sci
 , 
2000
, vol. 
4
 (pg. 
365
-
367
)
Lobaugh
NJ
West
R
McIntosh
AR
Spatiotemporal analysis of experimental differences in event-related potential data with partial least squares
Psychophysiology
 , 
2001
, vol. 
38
 (pg. 
517
-
530
)
Loveless
NE
Sanford
AJ
Slow potential correlates of preparatory set
Biol Psychol
 , 
1974
, vol. 
1
 (pg. 
303
-
314
)
Luu
P
Collins
P
Tucker
DM
Mood, personality, and self-monitoring: negative affect and emotionality in relation to frontal lobe mechanisms of error monitoring
J Exp Psychol Gen
 , 
2000
, vol. 
129
 (pg. 
43
-
60
)
Luu
P
Tucker
DM
Makeig
S
Frontal midline theta and the error-related negativity: neurophysiological mechanisms of action regulation
Clin Neurophysiol
 , 
2004
, vol. 
115
 (pg. 
1821
-
1835
)
McIntosh
AR
Bookstein
FL
Haxby
JV
Grady
CL
Spatial pattern analysis of functional images using partial least squares
Neuroimage
 , 
1996
, vol. 
3
 (pg. 
143
-
157
)
Murphy
TI
Richard
M
Masaki
H
Segalowitz
SJ
The effect of sleepiness on performance monitoring: I know what I am doing, but do I care?
J Sleep Res
 , 
2006
, vol. 
15
 (pg. 
15
-
21
)
Overbeek
TJM
Nieuwenhuis
S
Ridderinkhof
KR
Dissociable components of error processing: on the functional significance of the Pe vis-à-vis the ERN/Ne
J Psychophysiol
 , 
2005
, vol. 
19
 (pg. 
319
-
329
)
Rabbitt
PMA
Three kinds of error-signalling responses in a serial choice task
Q J Exp Psychol
 , 
1968
, vol. 
20
 (pg. 
179
-
188
)
Ridderinkhof
KR
Nieuwenhuis
S
Bashore
TR
Errors are foreshadowed in brain potentials associated with action monitoring in cingulate cortex in humans
Neurosci Lett
 , 
2003
, vol. 
348
 (pg. 
1
-
4
)
Salthouse
TA
Meinz
EJ
Aging, inhibition, working memory, and speed
J Gerontol B Psychol Sci Soc Sci
 , 
1995
, vol. 
50
 (pg. 
P297
-
P306
)
Swick
D
Turkin
AU
Dissociation between conflict detection and error monitoring in the human anterior cingulate cortex
Proc Natl Acad Sci USA
 , 
2002
, vol. 
99
 (pg. 
16354
-
16359
)
Ullsperger
M
von Cramon
DY
Subprocesses of performance monitoring: a dissociation of error processing and response competition revealed by event-related fMRI and ERPs
Neuroimage
 , 
2001
, vol. 
14
 (pg. 
1387
-
1401
)
Ullsperger
M
van Cramon
DY
Neuroimaging of performance monitoring: error detection and beyond
Cortex
 , 
2004
, vol. 
40
 (pg. 
593
-
604
)
Ullsperger
M
von Cramon
DY
Müller
NG
Interactions of focal cortical lesions with error processing: evidence from event-related brain potentials
Neuropsychology
 , 
2002
, vol. 
16
 (pg. 
548
-
561
)
van Veen
V
Carter
CS
The timing of action-monitoring processes in the anterior cingulate cortex
J Cogn Neurosci
 , 
2002
, vol. 
14
 (pg. 
593
-
602
)
Watson
D
Clark
LA
Tellegen
A
Development and validation of brief measures of positive and negative affect: the PANAS scales
J Pers Soc Psychol
 , 
1988
, vol. 
54
 (pg. 
1063
-
1070
)
West
R
Alain
C
Evidence for the transient nature of a neural system supporting goal-directed action
Cereb Cortex
 , 
2000
, vol. 
10
 (pg. 
748
-
752
)
West
R
Bowry
R
Krompinger
J
The effects of working memory demands on the neural correlates of prospective memory
Neuropsychologia
 , 
2006
, vol. 
44
 (pg. 
197
-
207
)
West
R
Bowry
R
McConville
C
Sensitivity of medial frontal cortex to response and non-response conflict
Psychophysiology
 , 
2004
, vol. 
41
 (pg. 
739
-
748
)
West
R
Moore
K
Adjustments of cognitive control in younger and older adults
Cortex
 , 
2005
, vol. 
41
 (pg. 
570
-
581
)
West
R
Schwarb
H
The influence of aging and frontal status on the neural correlates of regulative and evaluative aspects of cognitive control
Neuropsychology
 , 
2006
, vol. 
20
 (pg. 
468
-
481
)
Yeung
N
Cohen
JD
Botvinick
MM
The neural basis of error detection: conflict monitoring and the error-related negativity
Psychol Rev
 , 
2004
, vol. 
111
 (pg. 
931
-
959
)