Why is there an ERN/Ne on correct trials? Response representations, stimulus-related

Transcription

Why is there an ERN/Ne on correct trials? Response representations, stimulus-related
Biological Psychology 56 (2001) 173– 189
www.elsevier.com/locate/biopsycho
Why is there an ERN/Ne on correct trials?
Response representations, stimulus-related
components, and the theory of error-processing
Michael G.H. Coles a,*, Marten K. Scheffers b,
Clay B. Holroyd a
a
Department of Psychology, Uni6ersity of Illinois at Urbana-Champaign,
Cogniti6e Psychophysiology. Laboratory, 603 East Daniel Street, Champaign, IL 61820, USA
b
Department of Psychology, Florida State Uni6ersity, Tallahassee, FL 32306 -1270, USA
Received 11 January 2001; received in revised form 5 February 2001; accepted 1 March 2001
Abstract
The ERN or Ne is a component of the event-related brain potential that occurs when
human subjects make errors in reaction time tasks. It is observed in response-locked
averages, time-locked to the execution of the incorrect response. Recent research has
reported that this component is present on correct response trials, thereby challenging the
idea that the component is specifically related to error-processing. In this paper, we argue
that the ERN or Ne observed on correct trials can be attributed to one or both of two
factors: either there is error-processing on correct trials, and/or the response-locked averages
used to derive the ERN/Ne are contaminated by negative components evoked by the
stimulus. For this reason, there is no reason to abandon theories that relate the ERN/Ne to
error-processing. © 2001 Elsevier Science B.V. All rights reserved.
Keywords: Error-related negativity; Ne; Error-processing; Response representations; Artifact
Recent years have seen an increasing interest in event-related brain potentials
(ERPs) that are related to the commission of errors in choice reaction time tasks.
This interest was stimulated by the pioneering work of Falkenstein and his
colleagues in Dortmund (Falkenstein et al., 1990, 1995) that was followed by
* Corresponding author. Tel.: +1-217-3332122; fax: +1-217-2445876.
E-mail address: [email protected] (M.G.H. Coles).
0301-0511/01/$ - see front matter © 2001 Elsevier Science B.V. All rights reserved.
PII: S 0 3 0 1 - 0 5 1 1 ( 0 1 ) 0 0 0 7 6 - X
174
M.G.H. Coles et al. / Biological Psychology 56 (2001) 173–189
similar studies at Illinois (Gehring et al., 1993, 1995). Although these laboratories
referred to their error-related event-related potentials by different names, the Ne in
Dortmund, and the Error-Related Negativity or ERN in Illinois, it was clear that
they had identified the same phenomenon, namely, when subjects commit errors in
choice reaction time tasks, a negative component is evident in the ERP waveform
derived by averaging EEG epochs that are time-locked to the erroneous response.
The potential starts at around the time of the onset of the response, reaching its
peak some 100 ms later. It is largest over fronto-central recording sites.
Research in these two laboratories gave rise to a model of error-related processing within which the specific ERN or Ne process is imbedded. A graphical version
of the Illinois view is shown in Fig. 1 (e.g. see Bernstein et al., 1995; Coles et al.,
1998). This view, which shares many features with the original Dortmund view (cf.
Falkenstein et al., 1991), is that the error-processing system is comprised of two
main components: a monitoring system that detects errors and a remedial action
system. The heart of the monitoring system is a comparator, which compares
representations of the correct or appropriate response1 with representations of the
actual response. In almost all cases which we and the Dortmund group have
investigated, errors are most likely to be due to fast guessing or other forms of
impulsive responding. Responses are executed before all the information necessary
to guide the correct response has been extracted from the stimulus (cf. Gratton et
al., 1988). Thus, a representation of the correct or appropriate response can be
derived from further, continued, processing of the stimulus, beyond that which is
Fig. 1. Schematic representation of the theory of the ERN/Ne. See text for explanation.
1
We use the term ‘correct or appropriate response’ to avoid confusion over the meaning of the phrase
‘correct response’. In the context of choice reaction time tasks, it is used to refer to a response with the
hand or finger designated by the experimenter. As noted below, we propose that other response
parameters can also be incorporated into what counts as a ‘correct’ response. Thus, we used the term
‘correct or appropriate’ to include both the designated hand or finger along with other response
parameters.
M.G.H. Coles et al. / Biological Psychology 56 (2001) 173–189
175
used to guide the actual response. Representations of the actual response are likely
to be derived from a central feedback system, since the latency of the ERN/Ne is
too short to allow for peripheral feedback. When the motor command is issued to
initiate the response, an ‘efference copy’ (cf. Angel, 1976) is sent to the monitoring
system. Since the latency of the ERN/Ne is not only short but relatively invariant
with respect to the response, the comparator appears to be activated by the
response itself. That is, the comparison between representations of correct/appropriate and actual responses is triggered by the arrival of the efference copy
indicating what response is being made. The comparison system does not wait until
all possible information about the appropriate response is available. Rather it uses
whatever information is available at the time of the response.
Errors are detected when the comparator reveals a mismatch between correct/appropriate and actual responses. The comparator sends an error signal to the second
error-processing component, the remedial action system. As its name implies, this
system is responsible for initiating remedial actions, actions that are designed to
inhibit the error or to correct it (Gehring et al., 1993). Another kind of remedial
action involves strategic adjustments that reduce the likelihood of errors re-occurring in the future. In reaction time tasks, these compensatory effects may involve a
slowing of responses (cf. Gehring et al., 1993; Rabbitt and Rodgers, 1977).
Our current view is that the ERN/Ne is generated by the arrival of the error
signal at the remedial action system. This view is based on both psychophysiological and neurophysiological data (Holroyd and Coles, (submitted)) and derives, in
part, from the proposal that the ERN/Ne reflects neuronal activity in the ventral
bank of the anterior cingulate sulcus, where dopaminergic activity initiated by the
basal ganglia results in a disinhibition of pyramidal neurons. A complete elaboration of the logic behind this argument is beyond the scope of this paper. It is
sufficient for present purposes to merely claim that the ERN/Ne is related to
error-detection (cf. Falkenstein et al., 1991).
In our original studies (Gehring et al., 1993) and those of Falkenstein et al.
(1991), analysis of the ERP waveforms on correct trials revealed little if any
evidence of a negative component. In fact, this specificity of the ERN/Ne to error
trials was the main reason that both Dortmund and Illinois groups used the term
‘error’ in their label for the component. However, several recent papers have drawn
attention to the presence of ERN/Ne-like activity on correct trials (e.g. Falkenstein
et al., 2000; Luu et al., 2000; Vidal et al., 2000). Furthermore, in our own research,
we have also observed an ERN/Ne on correct trials2 (e.g. Scheffers et al., 1999;
Scheffers and Coles, 2000). If the ERN/Ne is indeed an error-related ERP component, related to error-detection, why is it observed on correct trials?
In this paper, we consider two explanations for the ERN/Ne on correct trials: (a)
that there is error-processing on correct trials; (b) measurement artifacts.
2
We prefer the term ‘ERN/Ne on correct trials’ to ‘correct ERN/Ne’, because the latter term is
logically inconsistent (‘correct error-related negativity’).
176
M.G.H. Coles et al. / Biological Psychology 56 (2001) 173–189
1. Error-processing on correct trials
As the model depicted in Fig. 1 shows, an ERN/Ne should occur when
there is a mismatch between representations of the correct/appropriate
response and those of the actual response. The question is whether there might
be conditions in which a mismatch occurs even though the correct response is
actually executed. In what follows, we argue that a critical factor involves the
representation of the correct or appropriate response and that of the actual
response.
1.1. Temporal parameters and the representation of the correct or appropriate
response
The representation of the correct or appropriate response is derived from
processing of stimulus information and application of a stimulus-response mapping rule. If the stimulus is the letter ‘H’, and the subject has been instructed to
respond to the letter ‘H’ using the right hand, then the representation of the
correct or appropriate response will consist of the movement parameters associated with executing a right-hand response. Since reaction time tasks also involve
the requirement that subjects respond reasonably quickly, the representation of
the correct response will also include temporal response parameters. Thus,
slow responses made with the correct hand may well mismatch with the representation of the correct or appropriate response, leading to an ERN/Ne on
correct trials.
Support for these ideas is provided by the results of choice reaction time
studies in which an explicit deadline has been used (Johnson et al., 1997; Luu et
al., 2000). In these studies, an ERN/Ne was observed on trials when responses
were executed with the correct hand, but the reaction times of these ‘correct’
responses exceeded the deadline. It is important to note that, in both these
studies, the relevant data come from a condition in which there was no external
signal that marked the time of the deadline. Thus, subjects presumably generated
an internal representation of a correct and ‘in-time’ response. When the actual
response was correct but slow, this representation was used by the comparison
process (Fig. 1) to generate a mismatch.
Although most ERN/Ne studies have not involved the use, or suggestion, of
an explicit deadline, subjects clearly incorporate a response speed parameter into
their representation of what constitutes an appropriate response. After being
given instructions for a typical reaction time task, subjects do not take more
than a few 100 ms to respond. This leads to the prediction that slow, but
otherwise correct responses are associated with ERNs/Nes. This is precisely what
was found in a recent study by Pailing et al. (2000), who showed that the
amplitude of the ERN/Ne on correct trials increased with reaction time. Thus
one explanation for the observation of an ERN/Ne on correct trials is that
correct responses are sometimes slower than some internal standard for response
speed.
M.G.H. Coles et al. / Biological Psychology 56 (2001) 173–189
177
Fig. 2. The amplitude of the ERN on correct trials as a function of subjective judgements of response
(in)accuracy. (From Scheffers and Coles, 2000, Fig. 2, Copyright © 2000 by the American Psychological
Association. Adapted with permission.)
1.2. Problems with the representation of the correct or appropriate response
As noted above, the representation of the correct or appropriate response
depends critically on the processing of stimulus information and the application of
the stimulus-response mapping rule. If the stimulus is misperceived, or if the
mapping rule is forgotten or incorrectly applied, then the representation of the
appropriate or correct response will be compromised.
Scheffers and Coles (2000) examined the fate of the ERN/Ne under conditions in
which the stimuli were deliberately degraded. Under these circumstances, we
expected that there would be occasions when the subjects were unable to extract
any information about the stimulus or when they perceived the stimulus incorrectly.
When the average waveforms for correct and incorrect trials were evaluated, two
things were salient. First, the ERN/Ne on incorrect trials was smaller than we had
observed previously, and second, there was an ERN/Ne on correct trials (Scheffers
and Coles, 2000, Fig. 1).
To examine these effects in more detail, we relied on subjective ratings of
response (in)accuracy. Following each trial, subjects rated the accuracy of their
response on a five point scale ranging from Sure Correct, through Unsure Correct,
Don’t Know, Unsure Incorrect, to Sure Incorrect. In general, the amplitude of the
ERN/Ne covaried with perceived response inaccuracy. There was a monotonic
increase in ERN/Ne amplitude as a function of rated inaccuracy. Importantly, for
present purposes, when we looked only at ‘pure’ correct trials, that is at trials where
only the correct response was activated (there was no sign of EMG activation
associated with the incorrect response), the ERN/Ne was large when these correct
trials were judged as ‘Sure Incorrect’ (Fig. 2). In other words, when subjects
178
M.G.H. Coles et al. / Biological Psychology 56 (2001) 173–189
responded correctly, but judged that they had responded incorrectly, there was a
very large ERN/Ne. Given the degraded stimulus conditions we used in this
experiment, we reasoned that these correct trials were misclassified as incorrect
because the subject misperceived the stimulus. After responding prematurely,
further stimulus processing led to an incorrect representation of the stimulus. In
turn, this led to an inaccurate representation of the correct or appropriate response,
and a resulting mismatch with the representation of the actual response.
In almost all ERN/Ne experiments, easily identifiable stimuli have been used.
Thus, misperception of the stimulus is likely to be a rare event. Nevertheless, such
misperceptions can occur in these conditions if subjects are distracted or fatigued
(Scheffers et al., 1999).
Inappropriate application of the stimulus-response mapping rule could also result
in problems with the representation of the correct or appropriate response. In this
case, however, the stimulus would be correctly identified, but the incorrect response
representation would be generated because the wrong mapping rule is applied. As
with the problem of misperception described earlier, this would lead the comparison
process to detect a mismatch on correct trials. Note that, in this case also, the
actual response would be generated by another set of processes such as those that
occur when subjects guess.
1.3. Problems with the representation of the actual response
The comparison process could also detect a mismatch on correct trials if the
representation of the actual response was somehow compromised. Such a situation
would occur if both correct and incorrect responses were activated on the same
trial.
In the Scheffers and Coles (2000) experiment described earlier, we distinguished
between ‘pure’ correct trials, where there was no sign of incorrect response activity,
and ‘impure’ correct trials where incorrect activity was present. This incorrect
activity could be detected either in electromyographic recordings or in measures of
sub-threshold response force. Most of the analyses reported in the Scheffers and
Coles paper were confined to the pure trials. In a further analysis (Scheffers, 1999),
attention was directed to the ‘impure’ trials3. We identified two kinds of impure
trials: those where the correct response was executed, but incorrect response activity
was present; and those where the incorrect response was executed, but correct
response activity was present. We reasoned that if both incorrect and correct
responses are activated on the same trial, the process which compares the representation of the correct or appropriate response with that of the actual response would
always yield a partial mismatch, since the latter representation will include elements
of the incorrect response. Thus, as we observed (Fig. 3), there should be ERNs/Nes
on both incorrect and correct trials. However, the mismatch will be directly related
to the degree of activation of the incorrect response, and inversely related to the
3
The original purpose of this analysis was to evaluate the conflict monitoring hypothesis (e.g. Carter
et al., 1998). However, consideration of this issue is beyond the scope of the present paper.
M.G.H. Coles et al. / Biological Psychology 56 (2001) 173–189
179
degree of activation of the correct response. Thus, the ERN/Ne on incorrect trials
should be larger than that on correct trials. Again, this is what was observed.
Together, these data suggest that a compromised representation of the actual
response is associated with the presence of an ERN/Ne on correct trials.
It might be argued that co-activation of correct and incorrect responses occurs
infrequently and is confined to situations where stimuli are degraded or where
stimulus processing is somehow compromised. However, since the psychophysiological approach was first applied in mental chronometry, we have known that correct
responses can differ as a function of the presence of incorrect response activity. For
Fig. 3. The ERN/Ne (panel A) on correct and incorrect trials. Panels B and C show the pattern of EMG
activation on the same correct and incorrect trials. The trials were matched in terms of the relative EMG
activation of correct and incorrect response channels (Scheffers, 1999).
180
M.G.H. Coles et al. / Biological Psychology 56 (2001) 173–189
example, Coles et al. (1985) and Gratton et al. (1988) showed that a substantial
number of correct trials in the Eriksen flankers task (Eriksen and Eriksen, 1974) are
accompanied by electromyographic or cortical activity associated with the incorrect
response. Such activity is observed not only on confusing (incompatible) trials,
where aspects of the stimulus array are associated with the incorrect response, but
also on straightforward (compatible) trials where all elements of the stimulus array
are mapped to the correct response. Thus, it is reasonable to propose that
sub-threshold incorrect activity is present at least some of the time when the correct
response is executed.
2. Measurement artifacts
2.1. Single trial measures
Although most studies of the ERN/Ne have derived measures of its amplitude
and latency from averaged waveforms, there have been some attempts to derive
measures on individual trials (e.g. Gehring et al., 1993; Scheffers et al., 1996;
Falkenstein et al., 2000). For example, Scheffers et al. (1996) defined the amplitude
of the ERN/Ne on individual trials as the most negative value in a 250 ms window
whose left boundary was placed on the most positive point in a 160 ms window
centered on EMG onset. Since this measure was taken in the context of the EEG
signal, which is fluctuating in the frequency range of 0–40 Hz, the expected value
of the measure, given no ERN/Ne, will not be zero. Thus, ERN/Ne amplitude
values, based on single trial measures, will not be zero even when there is no
ERN/Ne! For this reason, the interpretation of the absolute value of ERN/Ne
amplitude measures derived in this way should be undertaken with caution.
2.2. Measures based on the response-locked a6erage
The majority of studies of the ERN/Ne have derived amplitude and latency
measures from average-waveforms, obtained by aligning trials with respect to the
response, and then averaging. Since the background EEG is assumed to be random
with respect to the timing of the response, it is tempting to also assume that, given
no response-locked activity, the expected value of the response-locked ERP signal
should be zero. Unfortunately, this is not the case. The problem is that stimulus-related ERP components are not necessarily eliminated when response-locked averages are computed. Thus, one can observe deflections in response-locked averages,
which resemble an ERN/Ne, but which have nothing to do with the response.
Instead, these deflections reflect residual stimulus-related activity that is not removed in the averaging process.
To illustrate this problem, we took an average stimulus-locked waveform from a
condition where no response was made by the subject. The stimulus was, in fact, a
visually presented symbol that gave subjects feedback about the speed and accuracy
of their reaction time response. It was presented 600 ms after the response and
M.G.H. Coles et al. / Biological Psychology 56 (2001) 173–189
181
Fig. 4. Simulated response-locked ERP waveforms generated by taking the parent stimulus-locked
waveform in the left panel and assuming that a reaction time response was executed with a particular
mean and standard deviation. Panels arranged horizontally indicate the effect of increasing reaction
time; panels arranged vertically indicate the effect of increasing the variability of reaction time. Numbers
in each panel refer to the mean and standard deviation of reaction time associated with each particular
waveform. So 150 (30) indicates that the waveform was derived by assuming a mean reaction time of 150
ms with a standard deviation of 30 ms. Vertical dashed arrows indicate the timing of the response.
indicated that the response was correct. The average stimulus locked waveform
(Fig. 4) is characterized by a negative peak of 4 mV with a latency of 200 ms, and
a later larger positive peak of 12 mV with a latency of 400 ms.
We computed pseudo response-locked averages by assuming that a response had
in fact been made to the stimulus. In different simulations, each involving 1000
trials, we assumed that the response was made with a mean reaction time of
between 150 and 300 ms, and that the reaction time had a standard deviation of
30, 50 or 80 ms. The results of the simulations are shown in Fig. 4.
Two points can be made on the basis of the simulations. First, when the
variability of reaction time is small, both negative and positive peaks are evident in
the response locked waveforms. As the variability of reaction time increases, so the
negative peak becomes attenuated and disappears, while the positive peak remains,
even when the standard deviation of reaction time is 80 ms. Second, as mean
reaction time changes, so the timing of both the negative and positive peaks change
with respect to the pseudo-response.
In general, it is evident that one can obtain an ERN/Ne (and also a Pe) because
stimulus-related components are not eliminated by response-locked averaging and
contaminate the response-locked average used to extract measures of error-related
ERP components. As far as the ERN/Ne is concerned, the problem is likely to be
182
M.G.H. Coles et al. / Biological Psychology 56 (2001) 173–189
most acute when reaction times are relatively homogeneous and when mean
reaction time is the same as or shorter than the latency of the negative peak in the
stimulus-locked average. When reaction times are longer than the stimulus-locked
negative peak, the temporal relationship between the artifactual ERN/Ne and the
response will become implausible. This situation might lead the investigator to infer
that there is a negative relationship between ERN/Ne amplitude and reaction time.
However, it would clearly be risky to make inferences about the relationship
between ERN amplitude and reaction time, until the artifact problem has been
solved.
2.3. Solutions to the artifact problem
Since our earliest publications on the ERN/Ne (Gehring et al., 1993), we have
advocated a matching procedure by which two groups of trials are created: a group
of all error trials, and a group of correct trials whose reaction times match the
reaction time of each of the error trials. Then, response-locked averages are created
for the two groups of trials, and error-related activity is identified by comparing or
subtracting the two groups of trials. The idea is that any stimulus-related activity,
present in the response-locked average, would be common to both correct and
incorrect trials and therefore eliminated by the subtraction.
There are two problems with this matching procedure. First, the distributions of
reaction times for correct and incorrect trials may not overlap. In this case, it would
not be possible to find any correct trials that matched the incorrect trials in reaction
time. Second, the procedure assumes that stimulus-locked activity is equivalent for
the two groups of matched trials. In this way, the comparison should isolate only
error-related components. However, as Falkenstein et al. (2000) have argued, this
assumption may not be valid. It may be the case that stimulus-related processing
differs between correct and incorrect trials, causing the difference in response
accuracy. Thus, these matched groups of trials might differ with respect to both
stimulus- and reponse-related activity.
Another way to distinguish between stimulus- and response-locked components is
to to sort the trials as a function of reaction time and then examine stimulus-locked
waveforms for different non-overlapping reaction time bins (cf. Lauber et al., 1994).
This is done separately for correct and incorrect trials. This procedure does not rely
on matching correct and incorrect trials for reaction time, and makes no assumptions about the equivalence of stimulus-locked activity for correct and incorrect
trials. Rather, for correct and incorrect trials separately, stimulus- and responselocked components are distinguished in terms of the sensitivity of their latency to
changes in reaction time. The latency of stimulus-locked components should be
relatively constant regardless of reaction time, while the latency of response-locked
components should covary with reaction time. In this way, it is possible to
distinguish between stimulus- and response-locked components.
Creation of non-overlapping reaction time bins, with sufficient trials in each bin
to create stable averages, requires a large number of correct and incorrect trials.
When it is not possible to create non-overlapping bins because of insufficient trials,
M.G.H. Coles et al. / Biological Psychology 56 (2001) 173–189
183
a moving average procedure can be used. The procedure begins by creating
overlapping bins of a particular width, which are staggered by a particular amount.
In the example described below, the width of the bins was 50 ms and the stagger
was 5 ms. This is done for all trials with reaction times ranging from 0 ms to the
longest reaction time. Then, average ERPs are computed for each subject for each
bin containing 30 or more trials. Then, grand averages are computed for those bins
that are common to all subjects. The result of this procedure is the creation of a
series of stimulus-locked ERP waveforms for trials with increasing reaction times,
where the difference in reaction time between adjacent averages is 5 ms, and where
adjacent averages are based on partially overlapping sets of trials. The ERP
waveforms are represented in a raster-like form, with one raster line corresponding
to each waveform. Variation in color is used to code variation in the voltage for
each of the lines, with negative values represented by green/blue and positive values
by red.
Fig. 5 shows the results of applying this procedure to a dataset obtained by
Holroyd and Coles (submitted). The data come from a modified version of the
Eriksen flankers task (Eriksen and Eriksen, 1974), where the probabilities of the
two responses were 0.8 and 0.2. This manipulation created a strong response bias
that was evident in the high error rate for those stimuli requiring the infrequent
response. Fig. 5 shows the data for correct responses and error trials in a condition
where the flankers were incompatible with the target and the target was highly
probable. For the correct trials (upper panel), data were available for all 15 subjects
for reaction time bins from 75 to 125 ms through 375 to 425 ms (represented in the
figure by the midpoints of these bins, 100–400 ms). There were many fewer error
trials, and only a narrow range of reaction time bins satisfied our criterion of 30 or
more trials. Furthermore, not all subjects satisfied this criterion. As a result, we
only show data (Fig. 5, lower panel) for reaction time bins ranging from 215 to 265
ms through 285 to 335 ms (again represented in the figure by the midpoints of these
bins, 240– 310 ms). Between five and seven subjects (less than 50% of the total
subjects) contributed data to these bins, and we show the data for illustrative
purposes only. In both upper and lower panels, the dotted diagonal line indicates
the value of the midpoint of the reaction time bin corresponding to each of the
horizontal raster lines. Note that the presence of response-locked activity is
indicated by a diagonally oriented, isocolored column, which is parallel to the
reaction time diagonal.
In the upper panel, for correct trials, there is a negative peak at timepoint ‘a’
indicated by the blue/green color. The latency of this peak does not vary as a
function of reaction time as indicated by the vertically oriented blue/green column
at this timepoint. Since the latency of this column is about 100 ms, we can infer that
it is an N1, time-locked to the stimulus. At timepoint ‘b’, there is another negative
component, indicated by the green/blue column that is also time-locked to the
stimulus. The latency of the midpoint of this column is about 300 ms, and it can be
classified as an N2. These two stimulus-locked negativities are also evident in the
data for incorrect trials shown in the lower panel, although these negativities are
not as pronounced as they are for the correct trial data. The lower panel also
184
M.G.H. Coles et al. / Biological Psychology 56 (2001) 173–189
Fig. 5. Raster-like plot of stimulus-locked voltage changes as a function of reaction time. Upper panel:
correct trials. Lower panel: error trials. Abscissa: time in ms relative to stimulus onset. Ordinate:
reaction time in ms. The diagonal dotted line in each panel indicates the value of the midpoint of the
reaction time bin associated with each horizontal raster plot. See text for further details. (Based on data
reported by Holroyd and Coles, (submitted).)
M.G.H. Coles et al. / Biological Psychology 56 (2001) 173–189
185
contains a third region of negativity, labeled ‘c’. Since the peak of this negativity
follows the reaction time by about 100 ms, we infer that it is an ERN/Ne.
The possibility for confusion in the interpretation of the negative peaks is evident
in the correct trials data (upper panel) for reaction time responses with a latency of
about 200 ms. Here, there is a negative peak that follows reaction time by about
100 ms. In other words, the peak has the latency characteristics of an ERN/Ne, and
it would be tempting to infer that there is, in fact, an ERN on correct trials.
However, since this negative peak has approximately the same latency (in relation
to the stimulus) for responses with both shorter and longer reaction times, it is
more reasonable to classify it as a stimulus locked component, such as an N2,
rather than a response-related ERN/Ne. The region marked by an asterisk in the
upper panel (correct trials) corresponds to the region ‘c’ in the lower panel (error
trials) where an ERN/Ne is present. So if there was an ERN/Ne for correct
responses with a latency of between 215 and 335 ms, one would expect to see a
region of negativity where the asterisk is located. Since there is no negativity there,
we can infer that there is no ERN/Ne for correct responses with a latency of
between 215 and 335 ms. Taken together, the data depicted in the upper panel
suggest that there is no ERN/Ne on correct trials, even though there is a negative
peak at the appropriate latency for a subset of the trials.4
3. Reconsideration of claims for an ERN/Ne on correct trials
The preceding arguments claim that there are two reasons why an ERN/Ne is
observed on correct trials: either there is error-processing on correct trials, or
stimulus-related ERP components are present in the response-locked average.
Our analysis of our own previous studies (e.g. Scheffers and Coles, 2000; see also
Scheffers et al., 1999), suggests that it is plausible to infer that error-related
processing occurs on correct trials when stimuli are degraded or when perceptual
processing deteriorates because of fatigue. However, some of the more striking and
perplexing examples of ERNs/Nes on correct trials have been found in situations
where degraded stimuli have not been used and where fatigue cannot be invoked as
an explanation. For example, the studies reported by Vidal et al. (2000) involve
tasks where very simple, easily discernible, stimuli are mapped to simple responses.
Their Go/Nogo study (Experiment II — simple RT task condition) provides an
excellent example. Here subjects were required to respond when one visual stimulus
was presented (e.g. the word ‘droite’) and to withhold their response when another
stimulus (the word ‘pie`ge’) occurred. The probabilities of these stimuli were 0.9 and
0.1, respectively. An ERN/Ne was observed for correct responses to the word
‘droite’. We suggest two possible reasons for this result. First, the simple go/nogo
4
It may also be possible to distinguish between artifactual and real ERNs/Nes in terms of scalp
distribution. In this case, the distribution of the candidate negative peak would be compared to that of
the ‘classic ERN/Ne’. However, it may be difficult to use the scalp distribution criterion if the
distribution of stimulus-locked negative peak is similar to that of an ERN/Ne.
186
M.G.H. Coles et al. / Biological Psychology 56 (2001) 173–189
task is a situation in which reaction times are likely to be fast and homogeneous,
a situation where the chances of contamination by stimulus-locked activity are
enhanced (see above). However, since Vidal et al. do not provide stimulus-locked
waveforms in their paper, it is impossible to determine whether their responselocked averages were contaminated by a stimulus-locked negative component with
the appropriate latency. Second, the simple task setting is likely to be one where the
temporal parameters of the task, even if they are not specified explicitly, are salient.
There are, after all, not many other response parameters for the subject to consider.
If the subjects have indeed incorporated a temporal parameter into a representation
of the appropriate response, then slow, but correct, responses should be associated
with an ERN/Ne. Again, the results described by Vidal et al. do not include data
to permit us to evaluate this possibility5.
A second set of seemingly perplexing findings comes from studies of neuropsychological and psychiatric patients. Gehring and Knight (2000) reported that
patients with dorsolateral prefrontal lesions show ERNs/Nes of equivalent amplitude on both correct and incorrect trials. Similar results have been obtained in
people with schizophrenia (see Ford, 1999). Interestingly, some theories of
schizophrenia implicate dysfunction in prefrontal cortex and of the dopaminergic
pathways that project to prefrontal cortex and basal ganglia (e.g. Cohen and
Servan-Schreiber, 1992). This commonality between prefrontal and schizophrenic
patients suggests a possible mechanism for the presence of an ERN/Ne on correct
trials.
The fact that these patients respond with an accuracy that is greater than chance
level indicates that they: (a) have a representation of the appropriate response or
goal; and (b) are able to set up the response system such that the actual response
matches the appropriate response. The issue, then, is why the monitoring system
does not use these two sources of information (the actual and appropriate response)
to detect a match (which would lead to no ERN/Ne — see Fig. 1). We have
proposed that the comparison process is implemented in the basal ganglia (Holroyd
and Coles, (submitted)). Furthermore, the prefrontal cortex has long been associated with the storage of goals and intentions (Stuss and Benson, 1984; GoldmanRakic, 1987). If the appropriate response is a goal, then damage to prefrontal
cortex, or to the pathway from prefrontal cortex to the basal ganglia, would mean
that an intact representation of the correct response would be unavailable to the
comparison process and an ERN/Ne would occur on correct trials.
4. Conclusions
In this paper, we have argued that ERNs/Nes are sometimes observed on correct
trials either because error-processing occurs on these trials or because stimulus-related activity is present in the response-locked averages used to derive the ERN/Ne
5
We do not intend this as a criticism of Vidal et al. Very few studies of the ERN/Ne have ever
included information about the relationship between the amplitude of the component and reaction time.
M.G.H. Coles et al. / Biological Psychology 56 (2001) 173–189
187
(or both). As a matter of experimental strategy, we suggest that the artifact problem
be addressed first. While we have focused here on the ERN/Ne on correct trials, the
artifact problem should also be addressed for incorrect trial data. If the deflection
in the waveform on correct trials that appears to be an ERN/Ne can be attributed
to stimulus-related artifact, then there is no phenomenon left to explain. On the
other hand, if the ERN/Ne-like deflection is not an artifact, then it requires an
explanation. In this paper, we have provided an expanded theory within which
existing reports of ERNs/Nes on correct trials may be accommodated. We have
focused on the nature of the response representations that are used to detect
whether there is a match between intended and actual responses. We have emphasized the importance of a number of factors, including temporal response parameters and co-activation of correct and incorrect responses. We have reviewed existing
data that appear to support this expanded theory. Further research could be
motivated to test predictions that can be derived from the theory. For example,
there should be a systematic relationship between the size of the ERN/Ne and the
relative activation of correct and incorrect responses; and there should be no
ERN/Ne if a ‘pure’, fast, correct response is executed under conditions of no
stimulus ambiguity.
To summarize, we see no reason, at present, to abandon the proposal that the
ERN/Ne is specifically related to error-detection. In particular, we see no need to
replace the error-detection view, as Falkenstein et al. (2000) (p. 104) and Vidal et
al. (2000) (p. 126) have recently suggested, with the idea that the ERN/Ne reflects
the activity of the comparison process, rather than an outcome of this process.
Observations of ERNs/Nes on correct trials can be accommodated within existing
theories that propose that the ERN/Ne occurs only when the comparison process
detects a mismatch between response representations, and is therefore specifically
linked to the detection of errors (cf. Bernstein et al., 1995; Coles et al., 1998;
Falkenstein et al., 1991; Gehring et al., 1993).
Acknowledgements
This research was supported by grant (MH41445) to the first author, and by a
predoctoral fellowship (MH11530) to the third author. The paper was written while
the first author was on leave at the Max Planck Institute for Psycholinguistics,
Nijmegen, The Netherlands. Although the ideas expressed in this paper are our
own, they were refined as a result of interactions with Michael Falkenstein, Tara
Johnson, Peter Hagoort, and Colin Brown, and informed by prior work by Bill
Gehring.
References
Angel, R.W., 1976. Efference copy in the control of movement. Neurology 26, 1164 – 1168.
188
M.G.H. Coles et al. / Biological Psychology 56 (2001) 173–189
Bernstein, P., Scheffers, M.K., Coles, M.G.H., 1995. ‘‘Where did I go wrong?’’: a psychophysiological
analysis of error detection. J. Exp. Psychol. Hum. Percept. Perform. 6, 1312 – 1322.
Carter, C.S., Braver, T.S., Barch, D.M., Botvinick, M.M., Noll, D., Cohen, J.D., 1998. Anterior
cingulate cortex, error detection and the online monitoring of performance. Science 280, 747 – 749.
Cohen, J.D., Servan-Schreiber, D., 1992. Context, cortex and dopamine: a connectionist approach to
behavior and biology in schizophrenia. Psychol. Rev. 99, 45 – 77.
Coles, M.G.H., Gratton, G., Bashore, T.R., Eriksen, C.W., Donchin, E., 1985. A psychophysiological
investigation of the continuous flow model of human information processing. J. Exp. Psychol. Hum.
Percept. Perform. 11, 529 –553.
Coles, M.G.H., Scheffers, M.K., Holroyd, C., 1998. Berger’s dream? The error-related negativity and
modern cognitive psychophysiology. In: Witte, H., Zwiener, U., Schack, B., Do¨ ring, A. (Eds.),
Quantitative and Topological EEG and MEG Analysis. Druckhaus Mayer Verlag, Jena-Erlangen,
pp. 96–102.
Eriksen, B.A., Eriksen, C.W., 1974. Effects of noise letters upon the identification of target letters in
visual search. Percept. Psychophys. 16, 143 – 149.
Falkenstein, M., Hohnsbein, J., Hoormann, J., Blanke, L., 1990. Effects of errors in choice reaction
tasks on the ERP under focused and divided attention. In: Brunia, C.H.M., Gaillard, A.W.K., Kok,
A. (Eds.), Psychophysiological Brain Research. Tilburg University Press, Tilburg, The Netherlands,
pp. 192–195.
Falkenstein, M., Hohnsbein, J., Hoormann, J., Blanke, L., 1991. Effects of cross-modal divided
attention on late ERP components: II. Error processing in choice reaction tasks. Electroencephalogr.
Clin. Neurophysiol. 78, 447 –455.
Falkenstein, M., Hohnsbein, J., Hoormann, J., 1995. Event-related potential correlates of errors in
reaction tasks. In: Karmos, G., Molnar, M., Csepe, V., Czigler, I., Desmedt, J.E. (Eds.), Perspectives
of Event-Related Potentials Research (EEG Journal Supplement 44). Elsevier, Amsterdam, pp.
280–286.
Falkenstein, M., Hoorman, J., Christ, S., Hohnsbein, J., 2000. ERP components on reaction errors and
their functional significance: a tutorial. Biol. Psychol. 52, 87 – 107.
Ford, J.M., 1999. Schizophrenia: the broken P300 and beyond. Psychophysiology 36, 667 – 682.
Gehring, W.J., Knight, R.T., 2000. Prefrontal – cingulate interactions in action monitoring. Nature
Neurosci. 3, 516 –520.
Gehring, W.J., Goss, B., Coles, M.G.H., Meyer, D.E., Donchin, E., 1993. A neural system for error
detection and compensation. Psychol. Sci. 4, 385 – 390.
Gehring, W.J., Coles, M.G.H., Meyer, D.E., Donchin, E., 1995. A brain potential manifestation of
error-related processing. In: Karmos, G., Molnar, M., Csepe, V., Czigler, I., Desmedt, J.E. (Eds.),
Perspectives of Event-Related Potentials Research (EEG Suppl. 44), pp. 261 – 272.
Gratton, G., Coles, M.G.H., Sirevaag, E.J., Eriksen, C.W., Donchin, E., 1988. Pre- and post-stimulus
activation of response channels: a psychophysiological analysis. J. Exp. Psychol. Hum. Percept.
Perform. 14, 331 –344.
Goldman-Rakic, P.S., 1987. Circuitry of primate prefrontal cortex and regulation of behavior by
representational memory. In: Plum, F. (Ed.), Handbook of Physiology — The Nervous System, V.
American Physiological Society, Bethesda, MD, pp. 373 – 417.
Holroyd, C., Coles, M.G.H., (submitted). The neural basis of human error processing: reinforcement
learning, dopamine, and the error-related negativity.
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.
Lauber, E., Kornblum, S., Whipple, A., De Jong, R., 1994. Evoked potentials, reaction times, and the
P300: a new way of visualizing their interrelationships. Poster presented at the meeting of the
Cognitive Neuroscience Society, San Francisco, CA, March 26 – 28.
Luu, P., Flaisch, T., Tucker, D.M., 2000. Medial frontal cortex in action monitoring. J. Neurosci. 20,
464–469.
Pailing, P.E., Segalowitz, S.J., Davies, P.L., 2000. Speed of responding and the likelihood of error-like
activity in correct trial ERPs. Psychophysiology 37, S76.
M.G.H. Coles et al. / Biological Psychology 56 (2001) 173–189
189
Rabbitt, P., Rodgers, B., 1977. What does a man do after he makes an error? An analysis of response
programming. Q. J. Exp. Psychol. 29, 727 – 743.
Scheffers, M.K., 1999. Performance Monitoring: Error Detection and the Error-Related Negativity in
Choice-Reaction Time Tasks. Doctoral dissertation, University of Illinois at Urbana-Champaign.
Scheffers, M.K., Coles, M.G.H., Bernstein, P., Gehring, W.J., Donchin, E., 1996. Event-related brain
potentials and error-related processing: an analysis of incorrect responses to go and no-go stimuli.
Psychophysiology 33, 42 –53.
Scheffers, M.K., Coles, M.G.H., 2000. Performance monitoring in a confusing world: error-related brain
activity, judgements of response accuracy, and types of errors. J. Exp. Psychol. Hum. Percept.
Perform. 26, 141 –151.
Scheffers, M.K., Humphrey, D.G., Stanney, R.R., Kramer, A.F., Coles, M.G.H., 1999. Error-related
processing during a period of extended wakefulness. Psychophysiology 36, 149 – 157.
Stuss, D.T., Benson, D.F., 1984. Neuropsychological studies of the frontal lobes. Psychol. Bull. 95,
3–28.
Vidal, F., Hasbroucq, T., Grapperon, J., Bonnet, M., 2000. Is the ‘error negativity’ specific to errors?
Biol. Psychol. 51, 109 – 128.
.