vidal et al biol psy 03

18
Biological Psychology 64 (2003) 265–282 Error negativity on correct trials: a reexamination of available data F. Vidal a,b,, B. Burle b , M. Bonnet b , J. Grapperon c , T. Hasbroucq a,b a Institut de Médecine Navale du Service de Santé des Armées, BP 610, 83800 , Toulon Naval, France b Laboratoire de Neurobiologie de la Cognition, Centre National de la Recherche Scientifique and Université de Provence, Marseille, France c Laboratoire d’Explorations Fonctionnelles du système nerveux, Hˆ opital d’Instruction des Armées, Toulon, France Received 29 November 2001; accepted 31 January 2003 Abstract The error negativity, an EEG wave observed when subjects commit an error in a choice reaction time (RT) task, is often considered as a sign of error detection. Recently, reports of Ne-like waves on correct responses did challenge this interpretation. It has been proposed, however, that these Ne-like waves result either from an artifactual contamination of response-locked activities by stimulus-locked ones, or from an implicit monitoring of the time elapsing during the RT. Our aim was to reprocess published data: (1) to compare the shape and amplitude of EMG-locked and stimulus-locked ERPs on correct trials, and (2) to compare the size of the EMG-locked Ne-like waves obtained on fast and slow trials. The results neither support the artifact hypothesis nor the RT monitoring one. Therefore, it seems that the Ne-like waves observed on correct trials do correspond to a Ne, which suggests that the Ne has a broader significance than just error detection. © 2003 Elsevier B.V. All rights reserved. Keywords: Errors; Ne; ERN; Laplacian; Correct responses 1. Introduction When subjects produce an error in a reaction time (RT) task, a large negative EEG wave called “error negativity” (Ne; Falkenstein et al., 1991) or “error-related negativity” Corresponding author. Present address: Institut de M´ edecine Navale du Service de Sant´ e des Arm´ ees, BP 610, 83800, Toulon Naval, France. Tel.: +33-494099618; fax: +33-494099251. E-mail address: [email protected] (F. Vidal). 0301-0511/$ – see front matter © 2003 Elsevier B.V. All rights reserved. doi:10.1016/S0301-0511(03)00097-8

Upload: melagkolia

Post on 24-Apr-2015

12 views

Category:

Documents


0 download

TRANSCRIPT

Page 1: Vidal Et Al BIOL PSY 03

Biological Psychology 64 (2003) 265–282

Error negativity on correct trials:a reexamination of available data

F. Vidala,b,∗, B. Burleb, M. Bonnetb,J. Grapperonc, T. Hasbroucqa,b

a Institut de Médecine Navale du Service de Santé des Armées, BP 610, 83800 , Toulon Naval, Franceb Laboratoire de Neurobiologie de la Cognition, Centre National de la Recherche Scientifique and

Université de Provence, Marseille, Francec Laboratoire d’Explorations Fonctionnelles du système nerveux, Hˆopital d’Instruction des Armées,

Toulon, France

Received 29 November 2001; accepted 31 January 2003

Abstract

The error negativity, an EEG wave observed when subjects commit an error in a choice reactiontime (RT) task, is often considered as a sign of error detection. Recently, reports of Ne-like waves oncorrect responses did challenge this interpretation. It has been proposed, however, that these Ne-likewaves result either from an artifactual contamination of response-locked activities by stimulus-lockedones, or from an implicit monitoring of the time elapsing during the RT. Our aim was to reprocesspublished data: (1) to compare the shape and amplitude of EMG-locked and stimulus-locked ERPson correct trials, and (2) to compare the size of the EMG-locked Ne-like waves obtained on fast andslow trials. The results neither support the artifact hypothesis nor the RT monitoring one. Therefore,it seems that the Ne-like waves observed on correct trials do correspond to a Ne, which suggests thatthe Ne has a broader significance than just error detection.© 2003 Elsevier B.V. All rights reserved.

Keywords:Errors; Ne; ERN; Laplacian; Correct responses

1. Introduction

When subjects produce an error in a reaction time (RT) task, a large negative EEGwave called “error negativity” (Ne;Falkenstein et al., 1991) or “error-related negativity”

∗ Corresponding author. Present address: Institut de Medecine Navale du Service de Sante des Armees, BP610, 83800, Toulon Naval, France. Tel.:+33-494099618; fax:+33-494099251.

E-mail address:[email protected] (F. Vidal).

0301-0511/$ – see front matter © 2003 Elsevier B.V. All rights reserved.doi:10.1016/S0301-0511(03)00097-8

Page 2: Vidal Et Al BIOL PSY 03

266 F. Vidal et al. / Biological Psychology 64 (2003) 265–282

(ERN;Gehring et al., 1993) appears at the midline with a frontocentral maximum. The Nepeaks about 100 ms after the onset of the electromyogram (EMG) of the incorrect responseagonist. Several experimental arguments (seeFalkenstein et al., 2000andColes et al., 2001for a review) led to the idea that the Ne reflects an error-detection process. One of the mostimportant pieces of evidence in favor of this error-detection hypothesis was the specificity ofthe Ne to erroneous responses. Recently, however, several research groups observed Ne-likewaves on correct responses, after computing either an averaged reference (Falkenstein et al.,2000; Luu et al., 2000) or an estimation of the surface Laplacian (Vidal et al., 2000). ThisNe-like wave, however, was always smaller than the Ne evoked by errors.

A question clearly pointed out byColes et al. (2001, p. 175)is, “if the ERN/Ne is in-deed an error-related ERP component, related to error detection, why is it observed oncorrect trials?”. This issue is important because if the Ne is not specific to errors its func-tional significance must be reconsidered.Coles et al. (2001), assuming that the Ne doesrepresent an error-detection mechanism, discussed possible reasons for the presence ofNe-like waves on correct trials, as follows. In certain experimental situations, “false” errordetections may occur on correct trials.1 However, these authors admit that this explana-tion cannot apply to data where subjects had to perform a Go/Nogo task with very rare(10%) Nogo trials (Vidal et al., 2000; Experiment II): The likelihood of committing anerror was very low, the stimulus–response mapping was very simple and compatible, andthe stimuli were large and unambiguous as to the behavior they required. To account forthe Ne-like wave observed on correct (Go) trialsColes et al. (2001)proposed two possibleexplanations:

1. There is no “true” Ne on correct trials. Stimulus-evoked potentials overlap withresponse-evoked ones, which results in an illusory response-locked negativity. Condi-tions eliciting short RTs are, of course, especially prone to this kind of artifactual effect.Therefore, in Experiment II ofVidal et al. (2000), the response-locked negative wave ob-served on correct (Go) responses could result from a contamination by a stimulus-lockednegativity.

2. There is a “true” Ne on correct trials because subjects monitor more parameters of theirresponses than those explicitly specified by instructions. This causes a discrepancy be-tween an internal (subject-dependent) implicit definition of “correct” responses and theexternal (experimenter-dependent) explicit definition of correct responses. An especiallyrelevant additional parameter could be the speed of the response process. Indeed,Luuet al. (2000)have demonstrated that when subjects have to give their response before anRT deadline, the Ne on correct trials increases with RT and is particularly obvious for RTsexceeding the deadline. The same effect had been reported, previously, byJohnson et al.(1997). Therefore,Coles et al. (2001)proposed that the temporal parameter was implic-itly monitored by the subjects, i.e. the subjects would have implicitly set an RT deadlinebeyond which the response was “too late”. As a consequence, “too late” responses evokedan error signal resulting in a Ne, whereas “soon enough” responses did not. Therefore,Vidal et al.’s (2000)data are a mixture of objectively correct but subjectively incorrect

1 For instance, when the stimulus is either ambiguous or degraded, or when stimulus-response mapping isincongruent.

Page 3: Vidal Et Al BIOL PSY 03

F. Vidal et al. / Biological Psychology 64 (2003) 265–282 267

(too late) responses, and objectively and subjectively correct (soon enough) ones. Thepresence of the Ne-like wave in the average of correct trials could be due exclusively toa subgroup of late response trials.

The aim of the present work is to evaluate the two hypothesis proposed byColes et al.(2001)to account for the presence of Ne-like waves on correct responses: averaging artifactand/or monitoring of the temporal parameter (RT) of the response.

To achieve this goal, the data obtained in the Go/Nogo task of Experiment II ofVidalet al. (2000)were re-processed as follows:

1. Correct responses were averaged time-locked to the stimulus. If the EMG-locked Ne-likewave obtained on correct responses results from a contamination by stimulus evokedcomponents, these should show up better on stimulus-locked averages than on EMG-lockedones. Moreover, for a negative stimulus-evoked component(s) to be a candidate for suchan artifact, its latency should be compatible with the mean premotor time (PMT; PMTis the delay separating stimulus onset and EMG onset).

2. Correct responses were sorted according to their PMT. Correct trials associated withshort PMTs were averaged separately from correct trials associated with long PMTs. If,as suggested byColes et al. (2001), subjects have set an implicit RT deadline, then correctresponses associated with short RTs, should elicit smaller Nes than correct responsesassociated with long RTs, as in the experiments ofLuu et al. (2000) and Johnson et al.(1997).

3. The trials were sorted as a function of increasing PMT. Then, raster-like plots of stimulus-locked Laplacians amplitude were constructed, representing the amplitude of the Lapla-cian as a function of time and as a function of increasing PMTs. Two methods wereused: (1) the one published byJung et al. (2001), and (2) the one used byColes et al.(2001). This permits a dissociation of stimulus-locked from EMG-locked components.

2. Method

2.1. Subjects

There were 12 subjects (11 men, 2 women, from 24 to 50 years old). Written informedconsent was given before the experiment. They were all right handed as assessed by theEdinburgh inventory (Oldfield, 1971) and they had a normal or corrected to normal vi-sion. One subject had to be replaced by another one because his data contained too manyartefacts.

2.2. Stimuli

Stimuli consisted of the words: “droite”, “gauche”, “toutes” and “pièges” (“right”, “left”,“all” and “trap” in French, respectively). Each word written in white was presented in thecenter of a faradized video monitor (Stim System of Neuroscan©; total of visual angle1.5◦).

Page 4: Vidal Et Al BIOL PSY 03

268 F. Vidal et al. / Biological Psychology 64 (2003) 265–282

2.3. Task

Responses consisted of button presses on a response pad (Stim System of Neuroscan),and had to be made as quickly as possible after the presentation of the words on the screen.

2.4. Trial events

There were three pairs of blocks with 80 trials per block: in one pair, the response consistedof a right thumb key-press in response to the word “droite”, in another pair, the responseconsisted of a left thumb key-press in response to the word “gauche”, and in a third pairthe response consisted of joint, two thumb key-press in response to the word “toutes”. Theorder of pairs of blocks was counterbalanced across subjects. Each block contained 10%“catch” trials in which the word “pièges” was presented and the subjects had to give noresponse. In this case, the next RS was given 1300 ms after the presentation of the catchtrial. However, there was no uncertainty on the response to be given if required and theuncertainty on the requirement (response or no response) was still low: 10%.

Note that, in a counterbalanced order, subjects also performed a three-choice RT taskwhere stimuli, responses and stimulus-to-response mapping where the same as in the taskanalyzed here, except that there were no “catch” trials (Vidal et al., 2000for details).

2.5. Electrophysiological recordings

EEG was recorded continuously during the experiment from 10 scalp electrodes (seeFig. 1). The reference and ground were on the right and left mastoids, respectively.Impedances were kept below 5 k�.

Fig. 1. Position of the electrodes on the scalp. A nodal electrode (black) and its three surrounding neighbours forma “module”. FCz is comprised in a module (interelectrode mean distance, 3.7 cm). The reference and the groundare respectively on the right and left mastoid apophysis.

Page 5: Vidal Et Al BIOL PSY 03

F. Vidal et al. / Biological Psychology 64 (2003) 265–282 269

2.5.1. Electrodes placementIn order to be able to estimate the time course of surface Laplacian by the source derivation

method (Hjorth, 1975), as modified byMacKay (1983), we used an electrode configurationthat partly differs from the standard 10–20 electrode placement system. This configurationpermitted the Laplacian to be estimated at three electrodes-called “nodal” electrodes. Anodal electrode was surrounded by three other electrodes that formed the vertices of anequilateral triangle, so that the nodal electrode was at the center of that triangle. The dis-tanceD between a nodal electrode and its three surrounding electrodes was 1/20th of theinion–nasion plus tragus–tragus distance (i.e. 3.7 cm in average). A nodal electrode and itsthree surrounding neighbours formed a “module”. To ensure that the distances and anglesbetween electrodes were kept constant, we used a rigid, three branched, wire template, inwhich each branch was separated from its neighbouring branch by an angle of 120◦. Threeelectrode modules were constructed which enabled us to place two nodal electrodes on C3′and C4′ (about 1 cm medial from C3 and C4) and one nodal electrode on FCz.

EEG and EOG signals were fed in Nicolet amplifiers, amplified (30 000 times) filteredand digitized (bandwidth: .1–100 Hz, 12 dB per octave, sampling rate: 256 Hz). EOG wasrecorded bipolarly between electrodes situated above the right eye and at its outer canthus.No selective “notch” 50 Hz filter or additional digital filtering was used.

EMG was recorded from the flexor pollicis brevis of each hand, by paired surface Ag/AgClelectrodes (6 mm diameter), amplified (5000 times), filtered (high frequency cut-off: 1 kHz),full wave rectified and integrated (integration window: 5 ms), and then digitized (samplingrate: 256 Hz).

2.6. Artifacts rejection

Although the use of surface Laplacian has been shown to remove ocular contamination(Law et al., 1993), large ocular artifacts (> 50�V) were rejected on the basis of visualinspection of the monopolar recordings, i.e. the characteristic shape of these artifacts, EOGrecordings, and the gradients of activity obtained at different locations. We also carefullyrejected local artifacts (i.e. artifacts present at single electrodes: phasic artifacts as wellas slow electrical shifts) because the use of the Laplacian transformation enhances them.The remaining monopolar recordings were averaged and Laplacians were calculated on thebasis of these monopolar averages.

2.7. Averaging

Two kinds of averaging were performed on correct responses. Stimulus-related activi-ties were averaged time-locked to the stimulus and EMG-related activities were averagedtime-locked to EMG onset. The onset of EMG activities was detected by visual inspectionof each trial (Van Boxtel et al., 1993; Hasbroucq et al., 1999) for further averaging.

2.8. Data processing

2.8.1. LaplaciansUnder the assumption that the scalp is isotropic, one can demonstrate that the sur-

face Laplacian (�V(x,y) = ∂2V(x,y)/∂x2 + ∂2V(x,y)/∂y2, where�V is the Laplacian of the

Page 6: Vidal Et Al BIOL PSY 03

270 F. Vidal et al. / Biological Psychology 64 (2003) 265–282

electric potential,x andy the cartesian coordinates andV the function of these two variablesthat describes the spatial distribution of the potential) is proportional to the radial compo-nent of the gradient of the current density (also called more shortly “scalp current density”(SCD) or “current source density” (CSD)). Intuitively, the Laplacian can be viewed as thelocal degree of curvature of the scalp potential field at a given instant. It presents certain ad-vantages: it is reference free, and, acting as a high-pass spatial filter, it removes the blurringeffect of the diffusion of the currents through the highly resistive skull (Katznelson, 1981).It is often considered as giving a good approximation of what would be the corticogram(Gevins, 1989). The deblurring effect of the surface Laplacian makes that it greatly reducesthe contribution of remote sources to the local recordings. Conversely, it presents some dis-advantages: its sensitivity to a source decreases sharply with its depth (Pernier et al., 1988;Manahilov et al., 1992). The surface Laplacian is more sensitive to spatial and electricalnoise than monopolar recordings and, in the case of the source derivation method, it needsa regular symmetrical arrangement of the electrodes in order to be correctly approximated.

We used the source derivation method (Hjorth, 1975). This method is aimed at calcu-lating an approximation of the surface Laplacian at certain electrodes that we have called“nodal”. It can be demonstrated that if one approximates the gradient of potential (alongone spatial dimension) by the potential differences between two electrodes divided by theirdistance, then, if a nodal electrode is at the center of a triangle or of a square on whichapexes are placed surrounding electrodes, then the Laplacian is proportional to three timesthe potential at the nodal electrode (four times in the case of a square), minus the sumof potentials at the surrounding electrodes, divided by the square of the inter-electrodesdistance.

2.8.2. Fast and slow trailsEMG-locked averages corresponding to “fast” and “slow” responses were obtained as

follows. For each subject, the median PMT value was calculated. Trials were sorted intotwo groups according to their PMT: shorter or longer than the median PMT value. In orderto increase contrast, trials yielding PMTs equal to the median value were dropped.

2.8.3. Raster-like plots

1. A raster-like plot of stimulus-locked Laplacian was constructed as follows (Jung et al.,2001), using the EEGLAB and ICA toolbox software tools: (Delorme et al.:www.cnl.salk.edu/∼arno/eeglab.html; Makeig et al.:www.cnl.salk.edu/∼scott/ica.html). Estimation ofsurface Laplacians was calculated for each accepted trial on the basis of monopolar data.Then, the trial-by-trial Laplacian data were digitally high-pass filtered above 3 Hz in or-der to eliminate the slow components, which were of no interest for the present study.After filtering, trials containing activities larger than± 2.5 �V/cm2 were rejected. Allthe trials of all the subjects were sorted and classed from the shortest to the longestPMTs. Then, bins of 15 adjacent trials were averaged together. All the bins were repre-sented on the same graph in a raster-like plot of stimulus-locked Laplacian changes. Onthis three-dimensional graph, the Laplacian amplitude, represented by a color scale,is plotted as a function of time, in abscissa, and as a function of increasing PMT,in ordinate.

Page 7: Vidal Et Al BIOL PSY 03

F. Vidal et al. / Biological Psychology 64 (2003) 265–282 271

No error was possible in this task. Therefore, to allow for a comparison betweenthe rasters corresponding to correct responses and those corresponding to errors, weprocessed in the same way the data corresponding to error trials obtained from the samesubjects in a three-choice RT task performed before or after the simple RT task analyzedhere, in a counterbalanced order.

2. For the sake of comparison between our data and those ofColes et al. (2001)a secondraster-like plot of stimulus-locked data was also constructed by applying exactly thesame procedure as that described byColes et al. (2001)(p.183) except that: (1) theminimal number of trials averaged per 50 ms bin for each subject was 21 or more persubjects instead of 30 or more, and (2) our bins were constructed on the basis of PMTsinstead of RTs.

We constructed three plots for the correct responses (errors were to few to applythis procedure). The first plot corresponds to our monopolar data obtained on correctresponses. The second plot corresponds to the same data after they have been digitally3 Hz high-pass filtered (seeLuu and Tucker, 2001). The third plot corresponds to thesame data after Laplacian transformation.

3. Results

3.1. Behavioral data

Mean RT was 329 ms, mean PMT was 251 ms and median PMT was 242 ms.The number of trials yielded by each subjects, after artifact rejection performed on the

basis of monopolar recordings, in seven PMT bins is presented inTable 1. In the sameway, we also present, inTable 2, the total number of trials, the median PMT, the mean and

Table 1Number of trials obtained from each subject (rows 2–13) and all the subjects (last row), after artefact rejectionbased on monopolar recordings, in seven PMT bins from the fastest (<150 ms) to the slowest ones (> 400 ms)

<150 ms 150–200 ms 200–250 ms 250–300 ms 300–350 ms 350–400 ms > 400 ms

S.B. 24 17 39 77 132 103 42 9B.G. 24 22 92 185 84 17 5 5S.D. 25 12 42 63 103 105 52 43F.C. 25 22 41 127 138 61 30 10P.F. 26 28 75 177 112 22 7 9F.B. 26 38 88 126 75 59 23 11P.L. 27 36 112 125 73 37 13 9C.L. 33 25 108 145 82 24 6 6Y.D. 36 18 49 135 139 39 14 11J.R. 45 15 64 249 70 12 4 4J.G. 48 9 78 119 138 53 13 7P.G. 50 34 40 64 106 76 46 47

All subjects 276 828 1592 1252 608 255 171

The first column indicates the initials and the age of the subjects from the youngest (24) to the oldest (50).

Page 8: Vidal Et Al BIOL PSY 03

272 F. Vidal et al. / Biological Psychology 64 (2003) 265–282

Table 2This table presents for each subject (rows 2–13) and all the subjects (last row) the total number of trials obtainedafter artefact rejection based on monopolar recordings (second column), the median PMT (third column), the mean(fourth column) and the standard deviation (fifth column) of the PMTs, and the percentage of trials occurring in a200–300 ms premotor bin (last column)

Number of trials Median PMT Mean PMT Standard deviation 200–300 (%)

S.B. 24 419 277 274 66 50B.G. 24 410 230 229 51 66S.D. 25 420 293 293 78 40F.C. 25 429 254 260 64 62P.F. 26 430 227 234 60 67F.B. 26 420 234 243 79 48P.L. 27 405 223 230 70 49C.L. 33 396 223 228 61 57Y.D. 36 405 250 252 66 68J.R. 45 416 227 227 40 76J.G. 48 417 250 250 57 62P.G. 50 413 285 286 97 41

All subjects 4982 242 250.5 65.75 57

the standard deviation of the PMTs, and the percentage of trials occurring in a 200–300 msPMT bin.

FromTable 1, it appears that, (1) except for the slowest subject (S.D.), the largest numberof responses corresponds to PMTs included in the 200–250 ms or in the 250–300 ms bin,and (2) these two bins do not correspond to a fast or a slow tail of the distribution in anysubject. FromTable 2it appears that, depending on the subjects, the 200–300 ms PMT bincontains from 40% of the trials (slowest subject S.D.) to 76% of the trials (fastest subjectJ.R.). Moreover, all the means and medians of the subjects are comprised in this 200–300ms PMT bin. Therefore, this 200–300 ms bin does not contain the slowest or the fastest tailof the distribution for any subject but, rather, corresponds to median values of PMTs in allthe subjects.

3.2. Electrophysiological data

The surface Laplacian data presented here were obtained by the source derivation method(Hjorth, 1975) modified byMacKay (1983).

Fig. 2 represents stimulus-locked (a) and EMG-locked (b) grand averages for correctresponses (same scale). In this figure, the size of the stimulus-locked N1 is smaller than thesize of the EMG-locked Ne; moreover, the stimulus-locked N1 occurs too early to accountfor the presence of any EMG-locked negativity (see alsoFigs. 3a and 4c). As the observedmean differences already negate the contamination view, no statistical analyses are required.

Fig. 3a represents raster-like plots of stimulus-locked Laplacian changes (color scale)as a function of time (abscissa) and as a function of increasing PMTs (ordinate) for thecorrect responses of the task analyzed here (simple RT task in Experiment II ofVidal et al.,2000). The procedure used here is that described byJung et al. (2001). The vertical blackline indicates the moment of stimulus onset, the S-shaped black line represents the time of

Page 9: Vidal Et Al BIOL PSY 03

F. Vidal et al. / Biological Psychology 64 (2003) 265–282 273

Fig. 2. Amplitude of surface Laplacian estimated at FCz (ordinate, in�V/cm2) as a function of time (abscissa, inms) obtained on correct responses. (a) time-locked to stimulus onset (0 of time: vertical line); (b) time-locked toEMG onset (0 of time: vertical line).

EMG onsets. This figure confirms the interpretations drawn on the basis ofFigs. 2a and b: anegative wave, time-locked to the stimulus, occurs in the time window of the N1. Anotherwave, clearly time-locked to EMG onset follows the usual S-shape distribution of PMTs.In Fig. 3b one can compare the same raster-like plot for errors committed by the samesubjects in a three-choice RT task (choice RT task in Experiment II ofVidal et al., 2000).Although there are much less trials in the case of errors, a Ne with a large amplitude appearstime-locked to EMG onset. EMG onset latency is shorter for correct responses because theyare emitted in a simple RT situation, which was not the case for the errors.

Fig. 4 represents raster-like plots of stimulus-locked EEG changes (color scale) as afunction of time (abscissa) and as a function of increasing PMTs (ordinate) for the correctresponses of the task analyzed here (simple RT task in Experiment II ofVidal et al., 2000).The procedure used is the one described byColes et al. (2001). Fig. 4a corresponds to ourmonopolar data;Fig. 4b represents the same monopolar data, after they have been digitally 3Hz high-pass filtered;Fig. 4c corresponds to the same data, after they have been Laplacian

Page 10: Vidal Et Al BIOL PSY 03

274 F. Vidal et al. / Biological Psychology 64 (2003) 265–282

Fig. 3. Raster-like plot of stimulus-locked Laplacian changes (procedure ofJung et al., 2001). On thisthree-dimensional graph, the Laplacian amplitude, represented by a color scale (scale: tenth of�V/cm2), is plottedas a function of time (in abscissa: ms), and as a function of increasing PMTs (in ordinate); (a): correct responsesin the simple RT task; (b): errors committed by the same subjects in a similar three-choice RT task. The verticalblack lines indicate the moment of stimulus onset, the S-shaped black lines indicate the time of EMG onsets.

Page 11: Vidal Et Al BIOL PSY 03

F. Vidal et al. / Biological Psychology 64 (2003) 265–282 275

Fig. 4. Raster-like plots of stimulus-locked EEG changes (procedure ofColes et al., 2001) on correct responses.On this three-dimensional graph, EEG amplitude, represented by a color scale (a and b:�V; c: tenth of�V/cm2), isplotted as a function of time (in abscissa: ms), and as a function of PMTs (in ordinate: ms). (a) original monopolardata; (b) monopolar high-passed filtered (3 Hz) data; (c) Laplacian transformed data. The vertical white linesindicate the moment of stimulus onset, the tilted dashed white lines indicate the time of EMG onsets.

transformed. The vertical white lines indicates the moment of stimulus onset, the tilteddashed white lines indicate the time of EMG onsets.2

In Fig. 4a no clear negative wave time-locked to EMG onset can be evidenced. InFig. 4b,the 3 Hz high-pass filtering has partially unmasked an EMG-locked negative-going wave.Fig. 4c confirms the interpretations drawn on the basis ofFigs. 2a and b and 3a: a negativewave, time-locked to the stimulus, occurs in the time window of the N1. Another negativewave, clearly time-locked to EMG onset follows the linear evolution of PMTs.

2 The lack of continuity between the elements of the tilted line result from technical limitation and do not revealdiscontinuity in the data.

Page 12: Vidal Et Al BIOL PSY 03

276 F. Vidal et al. / Biological Psychology 64 (2003) 265–282

Fig. 5. Amplitude of surface Laplacian estimated at FCz (ordinate, in�V/cm2,) as a function of time (abscissa,in ms) obtained on correct responses, time-locked to EMG onset (0 of time: vertical line). Thin lines correspondto PMTs shorter than the median value of PMTs, bold lines correspond to PMTs longer than the median value ofRTs.

Fig. 5 represents EMG-locked averages corresponding to correct responses. Fast andslow trials are averaged separately. This figure shows that, in this experiment, the amplitudeof the Ne is not related to the PMT duration of the corresponding response.

4. Discussion

4.1. Reciprocal contamination of stimulus-locked and EMG-locked averages

The comparison between stimulus-locked and EMG-locked averages (Figs. 2a and b)indicates that the Ne-like wave occurring on correct trials cannot be accounted for by anaveraging artifact, that is a contamination of EMG-locked averages by stimulus-lockedERPs.Figs. 3a and 4calso indicate that the Ne-like wave obtained on correct responsesdoes not result from an overlap between stimulus-locked and EMG-locked EEG activities.On the contrary,Figs. 3a and 4csuggest that, when comparing stimulus-locked data leadingto different RTs, one must be cautious in interpreting stimulus-locked amplitude differences.For instance, many responses occur in the time window of the P300; therefore, EMG-lockednegativities might partly cancel the P300, reducing its apparent amplitude. When comparingthe amplitude of the P300 between conditions leading to different RTs (or conditions withand without a response), certain P300 differences might be simply due to partial cancellationby EMG-locked negativities. This is in line with the results ofSalisbury et al. (2001). Theseauthors studied the amplitude and topography of the P300 in button pressing and silentcounting. They suggested that the reduced amplitude of the P300 in button pressing wasdue to a contamination by EMG-locked components.

4.2. Implicit monitoring of the temporal parameter

More important is the fact that the amplitude of the Ne obtained on correct trials is notlarger as PMT get longer (Figs. 3a, 4c and 5). Luu et al. (2000) and Johnson et al. (1997)

Page 13: Vidal Et Al BIOL PSY 03

F. Vidal et al. / Biological Psychology 64 (2003) 265–282 277

obtained a Ne on correct trials which showed maximal amplitude when the RTs exceededa deadline. The present analysis indicates that, contrary toColes et al.’s (2001)hypothesis,the modulation of the size of the Ne as a function of RT on correct trials (Johnson, 1997;Luuet al., 2000) is the result of explicit (not implicit) setting of such a deadline. The presenceof a Ne on correct trials in the present experiment cannot be accounted for by a mismatchbetween subject’s evaluation of the ongoing RT and an implicit deadline.Pailing et al.(2000, p. S76)did report Ne-like waves on correct responses for long RTs in a task whereno explicit RT deadline was given. However, after they analyzed the correlation between theamplitude of this Ne-like wave and the size of the preceding P300, these authors concludedthat “. . . the greater ERN-like component for slow responses may be an artifact of largeRT variability for slow responses and the consequent P300 amplitude reduction.”3

To reconcile the present data withColes et al. (2001)opinion, one should add a sup-plementary hypothesis: the implicit deadline varies from trial to trial (remember that thesubjects of Luu et al. received a feedback on each trial). However, this explanation would bead hoc and, to our knowledge, there is no available data supporting (or excluding) this possi-bility. Another possibility would be that, in absence of explicit requirement, responses thatare too fast might also be considered as errors, for example, if subjects make fast guesses.If one considers the 200–300 ms PMT bin ofFig. 3a, it is clear fromTables 1 and 2that thisPMT bin does not correspond to the fastest or the slowest tails of the distribution of PMTs inany subject but, rather, to median PMT values. In this bin, however (extending roughly fromtrials 1000–3800), a negative wave appears inFig. 3a. InFig. 4c, all the subjects contributeequally to each bin of the distribution. In this figure, there is no evidence that the negativewave observed on correct responses disappears for the middle latencies of the distribution.Therefore, even in “middle latency” responses does a negative wave occur, time-locked toEMG onset.

4.3. Ne on correct responses: a mandatory component?

Now, althoughColes et al. (2001)did not discuss this possibility, a possible way toreconcile the existence of a Ne on correct trials with the error-detection hypothesis wouldbe to consider that the Ne-like wave observed on correct trials is not a “true” Ne butcorresponds to another EMG-locked “mandatory” component, independent from the Ne.This component would be present on all trials (correct or incorrect). On errors, another “true”Ne would add to this mandatory component. The EMG-locked negativity observed on errorswould result from the summation of two waves: a mandatory EMG-related component plusan error-related one. This possibility, however, does not seem to be supported by data.

Vidal et al. (2000), on the basis of CSDs maps showed comparable distributions for thenegative wave elicited on correct trials, the Ne evoked on errors and the Ne evoked afterincorrect EMGs. With more sophisticated methods,Luu and Tucker (2001)compared ingreat details the temporal and spatial properties of the Ne obtained on correct responses andon errors. These authors, observed that most of the RTs produced by the subjects occurredin the time window of the P300. This led them to consider that very large activities like

3 Note that such an artefact cannot account for the data reported byLuu et al. (2000)because these authorsfiltered their data, in the time domain, in order to eliminate the P300 component.

Page 14: Vidal Et Al BIOL PSY 03

278 F. Vidal et al. / Biological Psychology 64 (2003) 265–282

those evoked by erroneous trials showed up very clearly because of their size. However,they suggested that smaller EMG-locked activities such as the Ne corresponding to correctresponses could be dampened by the superimposition of a stimulus-locked P300 (this ar-gument is symmetrical to one made by Coles et al.). The time constant of the P300 beinglarger than that of the Ne, these authors filtered their data (4–12 Hz band pass) for both cor-rect and erroneous responses. After filtering, the waveforms corresponding to Ne obtainedon correct and incorrect responses (interpreted by the authors as a part of midline frontal� rhythm) were compared. The Ne was especially prominent on error trials but was alsopresent, although smaller, on correct responses. However, the Nes corresponding to correctresponses and to errors were not different regarding their latency, their scalp distribution(averaged reference: 128 electrodes), the position of their estimated sources (with brainelectrical source analysis; BESA), and their source solutions derived from the weightedminimum norm method. In spite of the large number of electrodes used to map the Ne, andof the different independent signal processing methods used byLuu and Tucker (2001)toanalyze the Ne on correct and incorrect responses, it was not possible to find any evidencethat the Ne elicited by correct responses and by errors were different components.

Therefore, the most parsimonious position to account for the available data seems to bethat: (1) There is an EMG-locked Ne-like wave on correct responses, and (2) This Ne-likewave can be distinguished from the Ne evoked by erroneous responses neither on its temporalnor on its spatial properties, which suggest that the Ne-like and the true Ne are of samenature. Moreover, the data presented here indicate that this Ne is not due to an implicitmonitoring of the temporal (RT) parameter. This makes problematic the interpretation ofthe Ne in terms of error detection.

Luu et al. suggested that the Ne is part of an oscillatory potential such as a midline frontaltheta rhythm. The present data support this view. InFig. 2, the Ne and the negative wavewhich follows it (peak around: 300 ms) can be viewed as the negative side of an oscillationin the� band. The same interpretation can be drawn fromFig. 3a, after EMG onset.

4.4. Observations on patients

Observations on patients do not fit very well with the error-detection hypothesis, either.Gehring and Knight (2000)recorded EEG activity of patients with unilateral prefrontallesions (PFC) performing an RT task. In these patients, the Ne evoked by correct responseswas as large as that evoked by errors. The same observations hold for schizophrenic patients(Ford, 1999), whose symptoms are often related to abnormal prefrontal functions.Coleset al. (2001)argued that this effect could be explained if one assumes that, in these patients,the representation of the appropriate response is unavailable to the comparator. In such acase, a Ne would always occur whether the response is correct or not.4 This suggests thatthe nervous system has no access to the status of the ongoing response. This viewpoint,however, is challenged by two kinds of observations.

4 Note that if frontal dysfunction unmasks a large Ne on correct responses instead of diminishing the Neon errors, then the large Ne on errors corresponds to the process by default. The small Ne observed on correctresponses would result from an inhibition of the structure generating the Ne by the prefrontal cortex.

Page 15: Vidal Et Al BIOL PSY 03

F. Vidal et al. / Biological Psychology 64 (2003) 265–282 279

1. WhileFord (1999)did not report the performance of her patients,Gehring and Knight(2000) did. The reduction of response force on error trials (as compared to correctresponses) was smaller in patients than in aged matched and young subjects. Now,considering the complexity of the task, it is remarkable that the error rate did not differbetween the patients and the two control groups (aged matched and young subjects).The post-error slowing (longer RTs on correct responses following an error:Rabbitt,1966) did not differ either between patients and aged matched control subjects. Moreover,as pointed out byGehring and Knight (2000, p. 518)“although the PFC group showedan unusually large ERN on correct trials, they did not ‘correct’ their correct responsesmore often than did the controls”. How is it possible that patients slow down theirRT after an error as well as normal subjects do, and do not “correct” their correctresponses more often than the controls do, if no representation of the correct response isavailable?

2. AlthoughFord (1999)did not discuss this point, her data show a large error positivity(Pe;Falkenstein et al. 1991) on errors in control subjects and in schizophrenic patients(Fig. 10, p. 679). The Pe is completely absent on correct responses in control subjectsand in schizophrenic patients. Although the traces shown byGehring and Knight (2000)stopped shortly after the resolution of the Ne, a close inspection of the data correspondingto their patients (Fig. 2, p. 518) shows that, as soon as 100 ms, after the erroneousresponse, a steep positive wave begins to develop whereas the potentials are at thebaseline level on correct responses. Therefore, the patients ofFord (1999)and probablythose of Gehring and Knight (2001) present a Pe on errors only. How is it possible thata Pe is elicited, on errors only, if no representation of the correct response is available?Now, if a correct representation of the correct response is actually available, and if theNe represents the output of a comparison process, how is it possible that the Ne is aslarge on correct responses as it is on errors?

In light of the present analysis and of recent available data from the literature, it seemsdifficult to maintain that the Ne reveals error detection, that is the outcome of a comparisonprocess, rather than the activity of this comparison process itself.

4.5. A conflict monitoring account of the Ne?

Recently, a new account of the Ne has been proposed byBotvinick et al. (2001), theconflict monitoring model. Basically, their model is a three-layered connectionist model,one layer for stimulus representations, one layer for response representations, and the lastone, the attentional module, represents the role of attention on perceptual processes. To thesethree layers model,Botvinick et al. (2001)added a “conflict monitoring” module whichcontinuously evaluate the degree of conflict between responses. The conflict is the productof the activations of the two responses, scaled by the strength of the inhibitory connectionbetween them. By mean of simulations, the authors showed that the degree of conflict invarious tasks and experimental conditions correlates with the Ne amplitude recorded inthese situations.

The presence of a Ne-like on correct trials in simple RT is a challenge for this version ofthe conflict model, even if we consider that the very few Nogo trials could induce a conflict

Page 16: Vidal Et Al BIOL PSY 03

280 F. Vidal et al. / Biological Psychology 64 (2003) 265–282

between “respond” and “not respond” (Nieuwenhuis et al., 2003). Indeed,Nieuwenhuiset al. (2003)recently applied the conflict hypothesis to account for the EEG activities inthe Go–Nogo task. Their idea was that “not respond” can be considered as a response, andhence that a conflict can occur between the two response representations. However, theyinterpreted the N200 as an index of the conflict in correct responses, that is an activityoccurring before the response (see alsoBotvinick et al., 2001, p.635), whereas the Ne inthe present study clearly occurs after the response. Therefore, if we follow Nieuwenhuiset al.’s conclusions, the conflict in the Go–Nogo task occurs before the response, and hencethe “Ne-like” wave cannot be interpreted as a conflict. One may argue, in line withCohenet al. (2001), that a conflict occurs after the correct response because the subjects had a poorrepresentation of the two responses (Go and Nogo), inducing a post-response activationof the “Nogo” representation even after the response in the Go trials. Although we cannotexclude such a possibility, it seems very unlikely as we see no reasons why it did not occuralso in Nieuwenhuis et al.’s data. Therefore, the presence of a “Ne-like” wave on correcttrials in the present task is not easily explained by the conflict model.

Finally, one cannot exclude a possibility put forward by several authors (Tucker et al.,1999; Gehring and Knight, 2000; Luu et al., 2000; Vidal et al., 2000; Gehring and Fencsik,2001) that the Ne represents an emotional or affective reaction.

4.6. Methodological aspects

The fact that the Ne on correct responses often goes unnoticed raises questions about itsreliability. One explanation may be found inLuu and Tucker (2001). As mentioned earlier,these authors, noticed that most of their responses occurred in the time window of the P300.They supposed that the small Nes (on correct trials) might be masked by a superimpositionof the P300. On errors, the large size of the Ne would be sufficient to show up despite theP300 masking effect. Therefore, they high-pass filtered their data in the temporal domain,to remove, at least in part, the P300 effect. Indeed, the Ne showed up more clearly aftersuch a processing. Thus, removing a part of the “raw” data unmasked or at least enhancedthe Ne on correct trials.

In a certain sense, the Laplacian transformation produces the same effect: it acts as ahigh-pass filter in the spatial domain. As a spatial filter, the Laplacian transformation doesnot create new data but removes part of them. By removing some components it unmasksthe remainder. In the same line of reasoning as that ofLuu and Tucker (2001), one cansuppose that, if the P300 and/or the resolution of the CNV, cancel the Ne on correct trialsand if they are not generated beneath FCz, high-pass spatial filtering removes or stronglyattenuates them.

These two filtering methods (temporal and spatial) are aimed at the same goal, unmask-ing small components overlapped by larger ones. This is achieved by removing largercomponents generated in another time domain (temporal filtering) or by removing othercomponents generated in another spatial domain or area (spatial filtering). While the Lapla-cian transformation is one of the available spatial filtering methods, it is not the only one.For instance, the averaged reference is another method but it is less efficient (Mac Farlandet al., 1997). Nonetheless, in the case of the Ne on correct trials, this method has proved tobe useful (seeFalkenstein et al., 2000).

Page 17: Vidal Et Al BIOL PSY 03

F. Vidal et al. / Biological Psychology 64 (2003) 265–282 281

The comparison ofFigs. 4a, 4b and 4cis consistent with our interpretation.Fig. 4a showsa negative wave time-locked to the stimulus occurring in the time window of the N1. Afterthe response, a large slow positive wave develops for at least 300 ms. A small decrease inthis positive activity can be suspected especially for the shortest and the slowest RTs. Aclose examination of Coles et al. data shows the same kind of effects, especially for theshortest and the slowest RTs, although the data are collected in a very different paradigm.Therefore, our monopolar data and Coles et al.’s ones show very similar patterns.

Now, after 3 Hz high-pass filtering (temporal filtering) the same monopolar data, a pro-cedure comparable to that used byLuu and Tucker (2001), we did unmask, at least in part,a negative component time-locked to EMG onset (Fig. 4b). Finally, after Laplacian trans-formation of the monopolar data (spatial filtering), we unmasked a negative wave clearlytime-locked to EMG onset (Fig. 4c). These results justify our processing and that ofLuuand Tucker (2001)in order to observe properly the negative wave elicited after correctresponses.

In conclusion, the presence of Ne-like waves on correct responses results neither froma contamination of EMG-locked activities by stimulus-locked ones, nor from an implicitmonitoring of the time elapsing during the RT. Therefore, the Ne-like and the true Ne arelikely of same nature. As a consequence, the current models accounting for the generationof the Ne should take into account this empirical fact.

Acknowledgements

We are very grateful to Dominique Reybaud for her technical contribution and to SoniaAllain for her help in data processing. This research was supported by grant “cognitique”ACT 54b from the Ministère de la Recherche and by grant 99-CO-02 from the DélégationGénérale à l’Armement. The authors are indebted to W.A. MacKay for language editingand comments.

References

Botvinick, M.M., Braver, T.S., Barch, D.M., Carter, C.S., Cohen, J.D., 2001. Conflict monitoring and cognitivecontrol. Psychol. Rev. 108, 624–652.

Cohen, J.D., Botvinick, M., Carter, C., 2001. Anterior cingulate and prefrontal cortex: who’s in control?. Nat.Neurosci. 3, 421–423.

Coles, M.G.H., Scheffers, M.K., Holroyd, C.B., 2001. Why is there an ERN/Ne on correct trials? Responserepresentations, stimulus-related components, and theory of error-processing. Biol. Psychol. 56, 173–189.

Falkenstein, M., Hohnsbein, J., Hoormann, J., 1991. Effects of crossmodal divided attention on late ERP compo-nents. II. Error processing in choice reaction time tasks. Electroencephalogr. Clin. Neurophysiol. 78, 447–455.

Falkenstein, M., Hoormann, J., Christ, S., Hohnsbein, J., 2000. ERP components on reaction errors and theirfunctional significance: a tutorial. Biol. Psychol. 51, 87–107.

Ford, J.M., 1999. Schizophrenia: the broken P300 and beyond. Psychophysiology 36, 667–682.Gehring, W.J., Fencsik, D.E., 2001. Functions of the medial frontal cortex in the processing of conflict and errors.

J. Neurosci. 21, 9430–9437.Gehring, W.J., Knight, R.T., 2000. Prefrontal–Cingulate interactions in action monitoring. Nat. Neurosci. 3,

516–520.

Page 18: Vidal Et Al BIOL PSY 03

282 F. Vidal et al. / Biological Psychology 64 (2003) 265–282

Gehring, W.J., Goss, B., Coles, M.G.H., Meyer, D.E., Donchin, E., 1993. A neural system for error detection andcompensation. Psychol. Sci. 4, 385–390.

Gevins, A.S., 1989. Dynamic functional topography of cognitive tasks. Brain Topogr. 2, 37–56.Jung, T.-P., Makeig, S., Westerfield, M., Townsend, J., Courchesne, E., Sejnowski, T.J., 2001. Analysis and

visualization of single-trial event-related potentials. Hum. Brain Mapping 14, 166–185.Hasbroucq, T., Possamaı, C.A., Bonnet, M., Vidal, F., 1999. The effect of irrelevant location of the response signal

on choice reaction time: an electromyographic study in man. Psychophysiology 36, 522–526.Hjorth, B., 1975. An on-line transformation of EEG scalp potentials into orthogonal source derivations. Electroen-

cephalogr. Clin. Neurophysiol. 39, 526–530.Johnson, T.M., Otten, L.J., Boeck, K., Coles, M.G.H., 1997. Am I too late? The neural consequences of missing

a deadline. Psychophysiology 34, S48.Katznelson, R.D., 1981. EEG recording, electrode placement, and aspects of generator localization, in: Nuñez,

P.L. (Ed.), Electric Fields of the Brain. Oxford University Press, New York.Law, S.K., Nuñez, P.L., Wijesinghe, R.S., 1993. High-resolution EEG using spline generated surface Laplacians

on spherical and ellipsoidal surfaces. IEEE Trans. Biomed. Eng. 40, 145–153.Luu, P., Tucker, D.M., 2001. Regulating action: alternating activation of midline frontal and motor cortical net-

works. Clin. Neurophysiol. 112, 1295–1306.Luu, P., Flaisch, T., Tucker, D.M., 2000. Medial frontal cortex in action monitoring. J. Neurosci. 20, 464–469.MacKay, D.M., 1983. On-line source density computation with a minimum of electrodes. Electroencephalogr.

Clin. Neurophysiol. 56, 696–698.Mac Farland, D.J., McCane, L.M., David, S.V., Wolpaw, J.R., 1997. Spatial filter selection for EEG-based com-

munication. Electroencephalogr. Clin. Neurophysiol. 103, 386–394.Manahilov, V., Riemslag, F.C., Spekreijse, H., 1992. The Laplacian analysis of the pattern onset response in man.

Electroencephalogr. Clin. Neurophysiol. 82, 220–224.Nieuwenhuis, S., Yeung, N., van den Wildenberg, W., Ridderinkhof, K.R., 2003. Electrophysiological correlates

of anterior cingulate function in a go/no-go task: effects of response conflict and trial-type frequency. Cogn.Affect. Behav. Neurosci. 3, 17–26.

Oldfield, R.C., 1971. The assessment and analysis of handedness: the Edinburgh inventory. Neuropsychologia 9,97–113.

Pailing, P.E., Segalowitz, S.J., Davies, P.L., 2000. Speed of responding and the likelihood of error-like activity incorrect trials. Psychophysiology 37(1), S76.

Pernier, J., Perrin, F., Bertrand, O., 1988. Scalp current density fields: concepts and properties. Electroencephalogr.Clin. Neurophysiol. 69, 385–389.

Rabbitt, P.M.A., 1966. Errors and error correction in choice-response tasks. J. Exp. Psychol. 71, 264–272.Salisbury, D.F., Rutherford, B., Shenton, M.E., McCarley, R.W., 2001. Button-pressing affects P300 amplitude

and scalp topography. Clin. Neurophysiol. 112, 1676–1684.Tucker, D.M., Hartry-Speiser, A., McDougal, L., Lure, P., deGrandpre, D., 1999. Mood and spatial memory:

emotion and the right hemisphere contribution to spatial cognition. Biol. Psychol. 50, 103–125.Van Boxtel, G.J.M., Geraats, L.H.D., Van den Berg-Lessen, M.M.C., Brunia, C.H.M., 1993. Detection of EMG

onset in ERP research. Psychophysiology 30, 405–412.Vidal, F., Hasbroucq, T., Grapperon, J., Bonnet, M., 2000. Is the “error negativity” specific to errors? Biol. Psychol.

51, 109–128.