Brain and Cognition 56 (2004) 267–276
www.elsevier.com/locate/b&c
Errors without conflict: Implications for performance
monitoring theories of anterior cingulate cortex
Vincent van Veena, Clay B. Holroydb, Jonathan D. Cohenb,c, V. Andrew Stengerd,
Cameron S. Cartere,*
a
e
Department of Psychology, University of Pittsburgh, PA, USA
b
Department of Psychology, Princeton University, NJ, USA
c
Department of Psychiatry, University of Pittsburgh, PA, USA
d
Department of Radiology, University of Pittsburgh, PA, USA
Departments of Psychiatry and Psychology, University of California at Davis, 4701 X Street, Sacramento, CA 95817, USA
Accepted 16 June 2004
Available online 3 September 2004
Abstract
Recent theories of the neural basis of performance monitoring have emphasized a central role for the anterior cingulate cortex
(ACC). Replicating an earlier event-related potential (ERP) study, which showed an error feedback negativity that was modeled as
having an ACC generator, we used event-related fMRI to investigate whether the ACC would differentiate between correct and
incorrect feedback stimuli in a time estimation task. The design controlled for response conflict and frequency and expectancy
effects. Although participants in the current study adjusted their performance following error feedback, we did not observe error
feedback-evoked ACC activity. In contrast, we did observe ACC activity while the same subjects performed the Stroop task, in
which an area of the ACC activated during both conflict and error trials. These findings are inconsistent with previous dipole models
of the error feedback negativity, and suggest the ACC may not be involved in the generation of this ERP component. These results
question involvement of the ACC in the detection of errors per se when controlling for conflict.
Ó 2004 Elsevier Inc. All rights reserved.
Keywords: Functional MRI; Anterior cingulate cortex; Frontal lobes; Errors; Feedback; Attention; Performance monitoring; Executive control;
Stroop interference
1. Introduction
The anterior cingulate cortex (ACC) is usually
thought of as playing an important role in implementing
the processes underlying adjustments of performance
control. In particular, it is thought to contribute to the
processing of errors.
Several lines of research have linked ACC activity to
the presence of ‘‘slips’’ (fast and impulsive but erroneous
responses) in speeded response tasks. Event-related potential (ERP) research has shown the presence of a
*
Corresponding author.
E-mail address: [email protected] (C.S. Carter).
0278-2626/$ - see front matter Ó 2004 Elsevier Inc. All rights reserved.
doi:10.1016/j.bandc.2004.06.007
frontocentral negativity immediately following an action
slip (Coles, Scheffers, & Fournier, 1995; Coles, Scheffers,
& Holroyd, 2001; Falkenstein, Hoormann, Christ, &
Hohnsbein, 2000; Gehring, Goss, Coles, Meyer, & Donchin, 1993; Scheffers, Coles, Bernstein, Gehring, & Donchin, 1996). This component, referred to as error
negativity or error-related negativity (ERN), has consistently been modeled as being generated by the ACC
(Alain, McNeely, He, Christensen, & West, 2002; Dehaene, Posner, & Tucker, 1994; Holroyd, Dien, & Coles,
1998; Luu, Tucker, Derryberry, Reed, & Poulsen, 2003;
Mathalon, Whitfield, & Ford, 2003; Van Veen & Carter,
2002b). In support of these models, neuroimaging
studies using fMRI have shown increased ACC activity
268
V. van Veen et al. / Brain and Cognition 56 (2004) 267–276
during error trials (Carter et al., 1998; Carter, MacDonald, Ross, & Stenger, 2001; Kerns et al., 2004; Kiehl,
Liddle, & Hopfinger, 2000; Menon, Adleman, White,
Glover, & Reiss, 2001). Theories differ, however, on
the cognitive processes that it implements––that is, theories differ on the nature of the algorithms that are carried out by the ACC, and also on the nature of the
representations that these algorithms work on. Two recent theories can be distinguished.
According to one theory, by Holroyd and Coles
(2002), the ERN is generated as part of a reinforcement
learning process. According to this ‘‘reinforcement
learning’’ (RL) theory, behavior is monitored by an
‘‘adaptive critic’’ in the basal ganglia, that detects
whether events are either better or worse than expected,
and signals this with a phasic increase or a decrease,
respectively, in dopaminergic activity. According to this
proposal, the function of the ACC is to select between
mental processes competing for access to the motor system, and the ERN is assumed to be generated because
the inhibitory influence of the dopaminergic innervation
in the ACC is briefly disrupted, fine-tuning the ACC to
do a better job on future trials.
Another view holds that the ACC monitors for the
presence of conflict between simultaneously active but
incompatible processing streams (for reviews, see Botvinick, Braver, Barch, Carter, & Cohen, 2001; Carter,
Botvinick, & Cohen, 1999; Cohen, Botvinick, & Carter,
2000; Gruber & Goschke, 2004; Van Veen & Carter,
2002a). Van Veen and Carter (2002b) have suggested
that during errors, conflict is considered to arise between
the executed incorrect response and activation of the
correct response due to ongoing stimulus evaluation.
This theory, which can account for many observations
of ACC activation during both correct trials during
tasks that elicit response conflict and during errors in
functional MRI and ERP research, has also received
extensive support from the results of computational
modeling studies (Botvinick et al., 2001; Jones, Cho, Nystrom, Cohen, & Braver, 2003; Yeung, Botvinick, & Cohen, in press). In these modeling studies, conflict is
operationalized as Hopfields energy (Hopfield, 1982),
which is a measure of activity taken over all units in
the set of interest and how strongly these units are compatible with one another; a high-energy state might be
the result of crosstalk between different streams of information processing. This theory sees the ACC as detecting such high-energy states, and engaging areas involved
in top-down control which act to lower the energy state.
It is important to note that, while the ACC appears to
be engaged mostly by conflict between response representations (Milham et al., 2001; Van Veen, Cohen, Botvinick, Stenger, & Carter, 2001), and in the modeling
studies so far energy is only computed over the response
layer (Botvinick et al., 2001; Jones et al., 2003), it has always been considered that other forms of representa-
tional conflict may also be monitored by the ACC
(Botvinick et al., 2001; Carter et al., 1998; Carter
et al., 1999). Recently, reports have suggested that the
ACC may indeed be engaged by conflict at other levels
of processing (Badre & Wagner, 2004; Milham, Banich,
& Barad, 2003; Van Veen & Carter, 2002a; Weissman,
Giesbrecht, Song, Mangun, & Woldorff, 2003). At the
computational level novel or unexpected stimuli, too,
may induce a high-energy state (conflict between an
incoming and expected stimulus), and such stimuli have
also been observed to evoke ACC activity (Clark, Fannon, Lai, Benson, & Bauer, 2000; Dien, Spencer, &
Donchin, 2003; Downar, Crawley, Mikulis, & Davis,
2000; Kiehl, Laurens, Duty, Forster, & Liddle, 2001)
although in some cases this conflict can also be understood in terms of conflict between frequent, prepared responses and rare ones.
According to both the RL and conflict theories, ACC
activity may be related to competing processes, and both
theories appear to make similar predictions regarding
slips during performance. However, some differences
are readily apparent, in particular regarding its engagement following error feedback. According to conflict
theory, ACC activity during error trials is generated
immediately during and following the slip and reflected
in the ERN, as a result of conflict between the simultaneous activation of the fast erroneous response and the
slower buildup of activation of the correct response, as
outlined above (Van Veen & Carter, 2002a, 2002b).
Conversely, during correct conflict trials, conflict is largest preceding the response and reflected in the frontocentral N2 of the ERP (Van Veen & Carter, 2002a, 2002b).
But an error feedback stimulus to which no overt response is required might also evoke conflicts of sorts,
to the extent that conflict occurs between an incoming
and an expected––or desired––stimulus, or between the
active task set and the task set imposed by the feedback
stimulus. In situations in which the error feedback stimulus does not elicit such conflict, conflict theory would
not predict any ACC activity.
In contrast, according to RL theory, the ACC can
also be activated by stimuli that might not evoke conflict. Holroyd and Coles (2002) based their theory partially on evidence from ERPs to feedback stimuli
informing subjects about their performance. These
ERP waveforms show a stimulus-locked, ERN-like
component, which is increased when the stimulus informs the subject that an error has been made (for a review, see Nieuwenhuis, Holroyd, Mol, & Coles, 2004).
Source localization has shown that this component can
be modeled as being generated by the ACC (Badgaiyan
& Posner, 1998; Gehring & Willoughby, 2002; Luu
et al., 2003; Miltner, Braun, & Coles, 1997; Ruchsow,
Grothe, Spitzer, & Kiefer, 2002). According to RL theory, this phenomenon reflects the same basic mechanism
as the ERN, namely, a phasic dopamine reduction in
V. van Veen et al. / Brain and Cognition 56 (2004) 267–276
ACC following the detection of ‘‘negative outcomes,’’
which is then used to optimize performance.
Concurrently, several recent fMRI studies have
shown error feedback-related ACC activation (Holroyd
et al., 2004; Monchi, Petrides, Petre, Worsley, & Dagher, 2001; Ullsperger & von Cramon, 2003). Although
such data might support RL theory, in those studies, a
conflict interpretation of these findings cannot be completely ruled out (see discussion section). Moreover,
other fMRI studies have failed to show ACC activity
following error feedback (e.g., Cools, Clark, Owen, &
Robbins, 2002). The fact that both the negativity and
the ACC activity evoked by error feedback appear sensitive to the degree of expectancy violation is also consistent with both RL theory and conflict theory (Butterfield
& Mangels, 2003; Holroyd & Coles, 2002; Holroyd
et al., 2004; Nieuwenhuis et al., 2002). The interpretation of such error feedback-evoked ACC activity is
therefore far from established.
As stated earlier, although the ACC appears to be
activated most robustly to conflict between response
representations, other results suggest that the ACC can
be engaged by conflict between other types of representations. ACC activation has been observed in response
to conflict between representations at intermediate (conceptual) levels of processing (Beauregard, Lévesque, &
Bourgouin, 2001; Milham et al., 2003; Robertson
et al., 2000) or by conflict caused by expectancy violations or novelty (Downar et al., 2000). Thus, a conflict
view of ACC functioning would not necessarily predict
its activation following error feedback, unless this feedback elicits conflict between conceptual representations,
requires a task set reconfiguration, or violates a strong
expectancy. The latter might well be the case for ACC
activation in the Wisconsin card sorting task (WCST)
(Monchi et al., 2001) or other studies in which error
feedback is infrequent. Expectancy violations and novelty will most likely involve conflict at intermediate/semantic levels of representation, or at the level of task
set implementation. The reinforcement learning theory
predicts ACC engagement to error feedback even when
such stimuli do not involve novelty or expectancy violation. As such, one way to distinguish between the competing theories is to obtain evidence that the ACC does
indeed activate to error feedback, in situations in which
a conflict interpretation can be ruled out.
In the study by Miltner et al. (1997), ERPs were
measured during a time estimation task in which the
participant received feedback which told him or her
whether the response was executed correctly or incorrectly. During each trial of this task, participants received a cue, and had to make a response as close as
possible to 1 s after the onset of this cue. A variable
time window, centered around the moment 1000 ms
after cue onset, was used to evaluate this response: if
the response fell within this time interval, it was judged
269
to be correct; if the response was made either too fast
or too slow, it was judged to be incorrect. The duration
of the time window depended on the accuracy of the
previous trial: it decreased when the previous response
was correct, making it more difficult for the participant, and it increased when the response was incorrect,
making it easier. Then, briefly after the response was
made, a feedback stimulus informed the participant
whether the time estimation was accurate or not. The
use of a time window that varied with performance ensured the following two interesting aspects of this task.
First, each participant received about as many correct
as error feedback stimuli, thus ruling out an interpretation in terms of expectancy violation or novelty. Second, participants were most likely unable to
determine or even guess whether a response was correct
before they received feedback. Therefore, expectancyrelated factors probably did not influence the processing of the feedback stimulus in that study. Thus, in this
task, participants engaged in error processing in the absence of conflict processing.
Miltner et al. (1997) analyzed the ERPs locked to the
onset of the feedback stimuli, and found that error feedback elicited a negative deflection around 230–270 ms
following feedback stimulus onset. Miltner et al. tested
participants using visual, auditory and somatosensory
feedback stimuli, and for each modality, source localization suggested a generator in the ACC. Similar dipole
models of comparable error feedback negativities have
also been described by other researchers (Badgaiyan &
Posner, 1998; Gehring & Willoughby, 2002; Luu et al.,
2003; Ruchsow et al., 2002), although some of these
models have implicated more posterior areas of the cingulate cortex in the generation of this component (Badgaiyan & Posner, 1998; Luu et al., 2003).
Some caveats are in order before strong inferences
can be made from these ERP studies. Most importantly,
source localization is an inherently imprecise method of
localizing brain activity. Source localization can only
provide simplified models of the brain activity that
might have generated a given scalp topography but there
is always the chance that a model is not correct, regardless of how well it can explain the data. For any given
scalp topography, an infinite possibility of dipole solutions exists. It is therefore impossible to make any definitive claims about brain activity solely on the basis of
dipole source models (Peters & de Munck, 1990; Scherg,
1990). Second, the scalp distribution of the feedback
negativity in the Miltner et al. (1997) study appears to
differ somewhat from that of the ERN, as actually noted
by those authors (p. 795). Thus, it is not clear whether
the error feedback-related negativity reflects the same
cognitive process as the ERN. In fact, as mentioned, a
few studies have found that the error feedback-related
negativity might be generated by the posterior cingulate
rather than the ACC (Badgaiyan & Posner, 1998; Luu
270
V. van Veen et al. / Brain and Cognition 56 (2004) 267–276
et al., 2003). Third, Miltner et al. used only 36 electrodes; whereas it has been argued that many more electrodes are needed to obtain reasonably accurate source
models (Srinivasan, Tucker, & Murias, 1998). In fact,
both reports that modeled the error feedback-related
negativity as being generated by the posterior cingulate
cortex (Badgaiyan & Posner, 1998; Luu et al., 2003)
used high-density EEG (128 channels plus the average
reference channel). In addition, both of those studies
compared the error feedback negativity to another
ACC-generated component, and both found differences.
Luu et al. (2003) compared the error feedback negativity
to the ERN, and found that despite the similarity in
scalp topography, the ERN was explained by a model
that included both anterior and posterior cingulate
sources, whereas the error feedback negativity was explained by a posterior cingulate sources alone. Badgaiyan and Posner (1998) compared the error feedback
negativity to (what appears to be) the frontocentral
N2 (thought to be the same component as the ERN,
see Van Veen & Carter, 2002b), and found differing
scalp topographies and dipole models; the error feedback negativity was explained by a posterior cingulate
source, the frontocentral N2 by an anterior cingulate
source. Conclusive evidence that the error feedback negativity and the ERN are the same component is therefore lacking; no high density EEG study to date has
performed a within-study comparison of the scalp
topographies of these two components and found that
they are the same. (It should be noted, however, that
Luu et al. did not use an exploratory dipole fitting method, leaving open the possibility that an ACC dipole
might have provided a better fit to their model.)
For these reasons, an fMRI investigation of brain
activity in response to error feedback may shed better
light on the underlying generators of this component
and the neural substrate of processing error feedback.
Therefore, we adapted the paradigm described by
Miltner et al. (1997) for a study using fMRI in order
to examine the neural basis of error feedback processing. Because of the slowness of the BOLD response,
we increased the stimulus onset asynchronies (SOAs)
in the experiment to 12 s, allowing enough time for
the BOLD response to resolve. Because of this manipulation, we avoided the possibility of cue-related activity confounding the feedback-related activity. In sum,
a strong prediction of the reinforcement learning theory is increased activation following error feedback
compared to correct feedback, in the absence of conflict related to action planning or expectancy violation.
Conflict theory does not predict such activity. In order
to compare any possible error feedback-related activity to activity elicited by slips in a speeded response
task, participants also performed a version of the
Stroop color-word task (MacLeod, 1991; Stroop,
1935).
2. Materials and methods
2.1. Research participants
Informed consent was obtained from 14 healthy
young participants (six women, eight men; all righthanded; age range 19–28, M = 21.4 years, SD = 2.2) in
agreement with the Institutional Review Board of the
University of Pittsburgh. They received $50 per hour
for their participation.
2.2. Procedure
Each participant performed four blocks of 16 trials
each of the time estimation task. At the beginning of
each block, a fixation point (Æ) was shown on screen
for 12 s. Each trial consisted of a cue stimulus that participants had to respond to, followed by a feedback
stimulus indicating whether performance was accurate.
During each trial, the cue ‘‘X’’ was shown for 500 ms,
followed by a fixation point for 11,500 ms. Depending
on subjects performance the feedback stimulus was then
shown for 500 ms, followed by yet another fixation point
for 11,500 ms. The feedback stimulus was a plus sign
‘‘+’’ to indicate performance was correct, and it was a
minus sign ‘‘ ’’ to indicate a performance error.
Participants were instructed to make a 1-s duration
estimate by making a right index finger button press following onset of the cue as soon as they thought that 1 s
had passed. As in the report by Miltner et al. (1997), responses were judged to be correct when the button press
fell within a varying time window centered around
1000 ms following cue onset. This time window can be
described as the interval [1000 ms t; 1000 ms + t],
whereby t varied as a function of performance. When responses were too fast or too slow, t was increased with
20 ms making correct feedback more likely; when responses were correct, t was decreased with 20 ms making
error feedback more likely. The initial value of t was
150 ms. As a result, the window length adjusted itself
such that each participant received approximately as
many correct as error feedback stimuli independent of
their performance.
Next, participants performed three blocks of a button-press version of the Stroop task. Before entering
the magnet, subjects were trained on the response mapping. During this training they performed one block of
trials, having been instructed to read the word and respond using a manual response. This would increase
the mapping of the word to the response finger, thus
increasing the amount of errors and response conflict
(e.g., Ruff, Woodward, Laurens, & Liddle, 2001). Participants were instructed to make a left index finger button
press when the word was RED or YELLOW, and a
right index finger button press when the word was
GREEN or BLUE. They were instructed that inside
V. van Veen et al. / Brain and Cognition 56 (2004) 267–276
the scanner, they would have to respond to the color
rather than the word, and were reminded again prior
to the onset of the Stroop task.
During congruent trials, the color and the word were
mapped onto the same response finger (e.g., RED in the
color red or YELLOW in the color red, both requiring a
left-hand response). During incongruent trials, color
and word were mapped onto opposite response hands.
Participants performed three blocks of 124 trials each,
each block started with a fixation point lasting for
12 s. The first four trials in each block were congruent,
for the rest stimuli were presented in random order with
75% being congruent and 25% incongruent. Each stimulus lasted 300 ms, and was followed by a 2700 ms fixation
point. Following the last trial in each block, another 12 s
fixation was presented. Participants were instructed to
respond fast but accurately. Stimuli were presented
using E-Prime (Psychological Software Tools, Pittsburgh, PA).
2.3. MR image acquisition and analysis
MR images were acquired using a 3.0 Tesla scanner
(General Electric Company) with a standard head coil,
using a spiral pulse sequence (TR = 1.5 s; TE = 181 ms;
FOV = 20 cm; flip angle = 70°). We acquired 28 functional images parallel to the AC-PC line, with the
23rd slice from the top through the middle of the
AC. Data analysis was performed using BrainVoyager
software (Brain Innovation, Maastricht, the Netherlands). Functional data was preprocessed using 3D motion correction, interscan slice time correction,
Gaussian spatial filtering (FWHM = 6 mm), and highpass filtering (low cutoff: 3 cycles/block, or
.0078125 Hz). In addition, for the time estimation task,
data were smoothed using a 2.8 s FWHM Gaussian filter; this preprocessing step was omitted in the analysis
of the data from the Stroop task, because this was a
fast event-related design, and temporal smoothing is
thought to be suboptimal for such designs (Della-Maggiore, Chau, Peres-Neto, & McIntosh, 2002). For each
subject, 3-dimensional anatomical images (SPGRs)
were acquired; functional data were aligned to these
and then transformed into Talairach space. For the
time estimation task, statistical analysis was performed
using a general linear approach using subject as a random factor with three predictors: One predictor accounted for the cue, one for correct feedback, and
one for error feedback. For the Stroop task, statistical
analysis was performed using a general linear approach, also using subject as random factor, with three
predictors: one predictor accounted for congruent trials
(correct trials only), one for incongruent trials (correct
trials only), and one for errors. These predictors were
obtained using a model of the hemodynamic response
to the onset of each trial type. A statistical threshold
271
of p < .005 was used, along with a contiguity threshold
of 100 mm3.
3. Results
3.1. Time estimation task performance data
RTs were about as fast on correct judgments
(M = 986 ms, SD = 43) as on incorrect judgments
(M = 974 ms, SD = 209), Wilcoxon signed ranks
z (13) = .85, ns. Also, participants went through about
as many correct trials (M = 33 trials, SD = 3.25) as error
trials (M = 31 trials, SD = 3.25), t (13) = 1.64, ns, replicating the findings by Miltner et al. (1997). The average
window width on the basis of which responses were
judged to be correct or not was ±132 ms. This is also
similar to the study by Miltner et al.; the average window widths in their study were ±132 ms in the auditory
condition, ±138 ms in the somatosensory condition, and
±124 ms in the visual condition.
We analyzed whether subjects used the feedback to
guide their time estimations in the same way that Miltner
et al. (1997) analyzed their data. That is, we calculated the
absolute RT difference between each trial n and subsequent trial n + 1 (ignoring, of course, the last trial of each
block). Then we analyzed that difference as a function of
whether subjects had received correct or error feedback
on trial n. Replicating the findings by Miltner et al.
(1997), participants changed their time estimates more
following error feedback (M = 202 ms, SD = 86) than following correct feedback (M = 126 ms, SD = 49), Wilcoxon signed ranks z (13) = 3.23 (p = .001). These results
strongly suggest that participants performed the task in
a similar way as the participants of Miltner et al. did; they
used the feedback to guide and adjust their behavior.
3.2. Stroop task performance data
Mean RTs in the congruent condition (M = 612 ms,
SD = 151) were significantly faster than RTs in the incongruent condition (M = 719 ms, SD = 203), t (13) = 6.5,
p < .001. Mean accuracy for the congruent condition
(M = 94%, SD = 7) was significantly higher than for the
incongruent condition (M = 85%, SD = 12), Wilcoxon
signed ranks z (13) = 2.94, p = .003. These findings replicate typical Stroop results (MacLeod, 1991): incongruent
trials elicit slower and less accurate responses.
3.3. fMRI data
First, we investigated whether error feedback stimuli
in the time estimation task elicited a greater hemodynamic response than correct feedback stimuli. No area
in the brain displayed this result (see Table 1), not even
at a decreased statistical threshold (p < .2). Results did
272
V. van Veen et al. / Brain and Cognition 56 (2004) 267–276
show (see Fig. 1A–C) that correct feedback stimuli, compared to error feedback stimuli, engaged a distributed
network of brain areas consisting of the bilateral caudate
nuclei, the right putamen, the right rostral and posterior
cingulate cortices, bilateral middle and inferior prefrontal gyri, bilateral superior temporal gyri, and the right
lateral and medial occipital cortex (see Table 1, Fig. 1).
To verify whether we were simply underpowered or
otherwise unable to detect any activity of the ACC to error feedback, we investigated whether response conflict
and action slips in the Stroop task led to ACC activity.
We will only report ACC activity here; additional analysis of the results of this Stroop task will be reported
elsewhere (van Veen et al., submitted). Two contrasts
were analyzed: incongruent versus congruent trials,
and errors versus correct trials. Results of the first con-
trast did include a region of the caudal ACC (485 mm3;
x = 0, y = 25, z = 28). Results of the second contrast resulted in three active regions in the ACC. One was located caudally (379 mm3, x = 1, y = 26, z = 33), and
overlapped with the area engaged by response conflict
(see Fig. 1D). A second region was located more rostrally (347 mm3, x = 1, y = 40, z = 22). A third region
was located more posterior and dorsally, and extended
into the supplementary motor areas (3672 mm3, x = 4,
y = 7, z = 58). These results clearly show that our participants were quite able to activate their ACC, and that we
were quite able to detect this activation.
A last set of tests involved performing post-hoc statistics using paired t tests on the three ACC regions of
interest (ROI) taken from the Stroop test. For the rostral ACC region, beta estimates were 5.03 (SD = 8.53)
Fig. 1. Areas of activation. (A) Activation of the right putamen in the time estimation task (y = 6), correct feedback > error feedback. (B) Activation
of the medial wall, including rostral and posterior cingulate cortices in the time estimation task (x = 5), correct feedback > error feedback. (C)
Activation of the bilateral caudate nuclei and rostral anterior cingulate in the time estimation task (z = 6), correct feedback > error feedback. (D)
Activation in the anterior cingulate cortex in the Stroop task (x = 2), obtained by a conjunction analysis using both incongruent > congruent and
errors > correct. p < .005, voxel contiguity = 100 mm3.
V. van Veen et al. / Brain and Cognition 56 (2004) 267–276
Table 1
Regions showing greater activity for correct feedback than error
feedback (p < .005, voxel contiguity = 100 mm3)
Area
Volume
(mm3)
r. Caudate nucleus
l. Caudate nucleus
r. Putamen
r. Rostral cingulate cortex (BA 24/32)
r. Paracentral lobule (BA 31)
r. Middle frontal gyrus (BA 9)
l. Middle frontal gyrus (BA 9)
r. Inferior frontal gyrus (BA 45)
l. Inferior frontal gyrus (BA 45)
r. Superior temporal gyrus (BA 22)
l. Superior temporal gyrus (BA 22/39)
r. Occipital lobe (BA 19)
r. Occipital lobe (BA 19)
r. Lingual gyrus (BA 19)
r. Lingual gyrus (BA 19)
1917
1836
512
143
130
123
103
265
252
136
1443
446
171
541
387
x
y
12
10
20
6
4
30
35
54
49
38
34
26
34
24
11
z
19
23
6
38
26
35
17
21
25
39
47
65
72
63
61
8
9
0
6
45
30
28
18
10
16
18
18
1
3
3
for correct feedback and 2.98 (SD = 5.12) for error feedback, (t (13) = .7, ns). For the caudal ACC region overlapping with the conflict area, beta estimates were 7.88
(SD = 3.94) for correct feedback and 6.08 (SD = 5.45)
for error feedback, (t (13) = 1.44, ns). For the caudal
ACC region extending into the SMA, beta estimates
were 10.11 (SD = 5.63) for correct feedback and 8.01
(SD = 5.74) for error feedback, (t (13) = 1.26, ns). Thus,
for each ROI, activity was in fact higher to correct feedback than to error feedback, although not significantly
so. This, again, suggests we were not merely underpowered; the fact that the signal was in the opposite direction argues against this possibility.
3.4. Discussion
Performance data clearly replicated the findings obtained by Miltner et al. (1997); the adjustable time window ensured that subjects received approximately equal
amounts of correct and error feedback. Moreover, following error feedback, participants adjusted their
behavior. These results strongly suggest that participants performed the task in a similar way as the participants of Miltner et al. did. Nevertheless, we did not
observe increased activation of the ACC to error feedback in the absence of conflict that might be caused by
expectancy violation, even at extremely low statistical
thresholds. This contrasts with previous results using dipole modeling of the error feedback negativity (Miltner
et al., 1997). Because our task design was almost identical to that used by Miltner et al., as were our performance results, we expect similar neural processes to be
engaged during our task and during the study of Miltner
et al. However, we failed to provide support for the dipole models obtained in that study. This result suggests
that the error feedback negativity may be generated by
areas other than the ACC.
273
It is, of course, possible that we were simply underpowered to detect error feedback-induced ACC activity.
Typically, the ERN is a very large signal, and the error
feedback negativity is much smaller, so it is possible that
the ACC signal following error feedback was simply too
small to be reliably detected using fMRI. However,
there are several reasons why we believe that this is
not likely to be the case. During the Stroop task, which
subjects performed during the same scanning session,
both response conflict and errors engaged the ACC
(see Fig. 1D), replicating earlier findings showing that
response conflict and errors engage the same area in
the ACC (Carter et al., 1998; Kerns et al., 2004). It is
notable that in ERPs, ACC activity during correct
incongruent trials in the Stroop task is thought to be reflected in the so-called N450 (Liotti, Woldorff, Perez, &
Mayberg, 2000; West, 2003; West & Alain, 2000), and
this component appears to have a similar amplitude as
the error feedback negativity does. Thus, if the error
feedback negativity reflects ACC activity, this activity
should be similar in magnitude to ACC activity during
correct incongruent Stroop trials. Since we were not
underpowered to detect ACC activity during such correct incongruent Stroop trials, we believe that it is unlikely we were underpowered to detect such activity
associated with negative feedback.
Analysis of data obtained from the Stroop task indicated that slips activated three areas in the ACC: one
rostral, one caudal and overlapping with an area
engaged by response conflict, and one area caudal
and extending into the supplementary motor areas.
Post-hoc tests were performed on these three areas to
investigate whether one of them would respond to error
feedback. However, results from these tests indicated
that, though not significant, the ACC activation was
in fact larger following correct feedback than following
error feedback. The fact that activation was not even
in the right direction is another argument against the
possibility that we were simply underpowered.
Results did, however, show a relative increase in
activity of the basal ganglia to correct feedback (see
Figs. 1A and C), consistent with research showing basal
ganglia activity to rewarding feedback stimuli (e.g., Delgado, Nystrom, Fissell, Noll, & Fiez, 2000; Knutson,
Westdorp, Kaiser, & Hommer, 2000). This is consistent
with the RL theory of Holroyd and Coles (2002), which
predicts increased basal ganglia activity following
unpredicted positive or rewarding events.
As noted in the introduction, in several previous
fMRI reports of ACC activity elicited by error feedback
(Holroyd et al., 2004; Monchi et al., 2001; Ullsperger &
von Cramon, 2003), a broad conflict-based explanation
of this activity in these studies cannot be ruled out. For
instance, Monchi et al. (2001) observed ACC activity to
error feedback in the WCST, which clearly involves task
set reconfiguration or expectancy violation. Ullsperger
274
V. van Veen et al. / Brain and Cognition 56 (2004) 267–276
and von Cramon (2003) and Holroyd et al. (2004) used
rather different paradigms than the present one; in their
tasks, participants had to make a two-button forced
choice response, and feedback was provided shortly following the response. It is therefore possible that the
feedback stimulus evoked the tendency to make a corrective response, leading to the ACC activation. If this
is the case, this ACC activation with feedback might
either reflect either the premotor functions of this region
of the brain (low levels of activation are seen during
compatible trials in response interference tasks), or conflict between engagement of the corrective response and
an online representation of the executed response maintained in working memory (cf. Jeannerod & Frak, 1999).
One way to test these possibilities would be to replicate
the tasks used by Ullsperger and von Cramon, or Holroyd et al., measuring muscle force or the lateralized
readiness potentials while varying the response-feedback
SOA. It is unlikely that something similar to this would
have occurred in the present study and in the study by
Miltner et al. (1997); first, the response-feedback SOA
was too long in our study, and second, the accuracy of
the response was determined by its timing, so that it
would not be possible to ‘‘correct’’ it. Finally, Bush
et al. (2002) observed increased ACC activity following
feedback indicating reduced reward. They had participants press one of two buttons following a cue, after
which they received a feedback stimulus. In 80% of trials, participants received a ‘‘reward’’ feedback stimulus,
indicating they earned 15 cents, and had to push the
same button on the next trial. In 10% of trials, the feedback stimulus instructed the participants to switch, that
is, push the other button on the next trial. In the remaining 10% of trials, they received a ‘‘reduced reward’’ feedback, indicating that they earned only 9 cents that trials,
and had to push the other button on the next trial. ACC
activity was highest to the reduced reward feedback,
somewhat lower to the switch feedback, and lowest to
the constant reward feedback. Based on these results,
Bush et al. proposed that the ACC plays a role in what
they refer to as ‘‘reward-based decision making,’’ rather
than merely responding to the amount of conflict that
occurs in the information processing stream. Note, however, that conflict theory can still account for these findings. Trials during which the participants received a
stimulus instructing them to switch occurred on 20%
of the trials (switch trials plus reduced reward trials),
while participants reward was reduced on 10% of the total amount of trials. Thus, the increased ACC activity
during these trial types might be considered as related
to novelty; the same goes for other studies (Monchi
et al., 2001). Alternatively, it is possible that the ACC
activity during such trials still reflected response conflict,
between the response established under previously
unambiguously rewarded conditions and the preparation of the alternative response. Because of these ambi-
guities in the literature we believe that it remains an
open question as to how feedback engages the ACC
and whether it does so independently of the conflict
which it may elicit. Further studies will be needed to resolve this question.
It is worth emphasizing again that the concept of
conflict, by the present account, goes beyond response
conflict. In the papers by Botvinick et al. (2001), Jones
et al. (2003), and Yeung et al. (in press), conflict was
defined as Hopfields energy calculated over the response layer of a PDP network. Previous studies have
indicated that the ACC appears to respond selectively
to response conflict (Milham et al., 2001; Van Veen
et al., 2001). More recent evidence, however, has suggested that the ACC appears to respond also to conflict
between other types of representations as well (Badre &
Wagner, 2004; Milham et al., 2003) including novelty
or expectancy violation (Clark et al., 2000; Dien
et al., 2003; Downar et al., 2000; Kiehl et al., 2001).
In fact, Botvinick et al. theoretically left it an open
question as to which kinds of conflict engage the
ACC (Botvinick et al., 2001, pp. 630, 646). Perhaps
novel modeling studies are needed to evaluate the
implications of this slightly expanded view, to generate
novel predictions and to prevent the theory from
becoming unfalsifiable.
As stated earlier, the fact that we failed to provide
support for the dipole models by Miltner et al. (1997)
suggests that the error feedback negativity is generated
by areas other than the ACC. However, in our study,
no brain area was engaged to a stronger degree by error feedback than by correct feedback. Because ERPs
cannot distinguish between relative increases or decreases in activity, it is possible that the error feedback
negativity in fact reflects the increased engagement of
one or more of the areas to correct feedback. Perhaps,
then, the error feedback negativity is better conceived
of as a superimposed ‘‘correct feedback positivity.’’
One possible way to test this would be to investigate
correlations between the amplitude of participants error feedback negativity and the fMRI signals associated with positive and error feedback. The hypothesis
outlined above predicts that the amplitude of the correct-error feedback difference wave should be correlated with correct feedback-related activity of a
generator somewhere in the brain other than the caudal
ACC.
In summary, in the present study using event-related
fMRI and a long delay we were unable to provide evidence that the ACC is involved in error feedback-detection. This finding therefore fails to provide support for
the RL theory of ACC function and the dipole models
on which it is, in part, based. It is therefore possible that
the error feedback negativity is not generated by the
ACC but by other elements of systems evaluating performance and feedback. But while this negative result
V. van Veen et al. / Brain and Cognition 56 (2004) 267–276
appears consistent with conflict theory, it is based upon
a null finding. It is conceivable that the small differences
between the current study and the study by Miltner et al.
(1997) were responsible for the lack of results. A possible factor is the difference in timing; the current study
used a very long cue to feedback SOA, and it is possible
that, despite the fact that performance measures indicated that participants executed the task in the same
way, they processed the feedback in a different way. Another possibility is that ACC activity might only come
into play at shorter response-feedback SOAs, either because there is less salience for delayed feedback or because there is less conflict. If this is true, then we
would expect the error feedback negativity to disappear
with increasing response-feedback SOAs and the systematic manipulation of salience and conflict in such
an experiment might help to elucidate the mechanisms
underlying this response. A third possibility is that there
is no one-to-one correspondence between neural activity
recorded at the scalp (as measured in the Miltner et al.
study) and blood flow (as measured in the present experiment), thus, it is possible that the ACC activity that
produces the error feedback negativity cannot be picked
up by fMRI. However, this seems somewhat unlikely, as
we were able to measure ACC activity on errors and
conflict trials during the Stroop task. Lastly, it is possible, even likely, that generation of the error feedback
negativity is a more complex process, to which multiple
brain areas might contribute; contribution of the ACC
to this process might be more paradigm-dependent.
Therefore we conclude that the involvement of the
ACC in processing error feedback remains an open
question that merits further careful investigation.
Acknowledgments
This work was supported by MH64190 from the
N1MH and a Translational Clinical Scientist Award
by the Burroughs-Wellcome Foundation to Dr. Carter.
We thank Mylaina L. Gordon for helping with the data
collection.
References
Alain, C., McNeely, H. E., He, Y., Christensen, B. K., & West, R.
(2002). Neurophysiological evidence of error-monitoring deficits in
patients with schizophrenia. Cerebral Cortex, 12, 840–846.
Badgaiyan, R. D., & Posner, M. I. (1998). Mapping the cingulate
cortex in response selection and monitoring. NeuroImage, 7,
255–260.
Badre, D., & Wagner, A. D. (2004). Selection, integration, and conflict
monitoring: Assessing the nature and generality of prefrontal
cognitive control mechanisms. Neuron, 41(3), 473–487.
Beauregard, M., Lévesque, J., & Bourgouin, P. (2001). Neural
correlates of conscious self-regulation of emotion. Journal of
Neuroscience, 21(RC165), 161–166.
275
Botvinick, M. M., Braver, T. S., Barch, D. M., Carter, C. S., & Cohen,
J. D. (2001). Conflict monitoring and cognitive control. Psychological Review, 108(3), 624–652.
Bush, G., Vogt, B. A., Holmes, J., Dale, A. M., Greve, D., Jenike, M.
A., et al. (2002). Dorsal anterior cingulate cortex: A role in rewardbased decision making. Proceedings of the National Academy of
Sciences of the United States of America, 99(1), 523–528.
Butterfield, B., & Mangels, J. A. (2003). Neural correlates of error
detection and correction in a semantic retrieval task. Cognitive
Brain Research, 17(3), 793–817.
Carter, C. S., Botvinick, M. M., & Cohen, J. D. (1999). The
contribution of the anterior cingulate cortex to executive processes
in cognition. Reviews in the Neurosciences, 10(1), 49–57.
Carter, C. S., Braver, T. S., Barch, D. M., Botvinick, M. M., Noll, D.
C., & Cohen, J. D. (1998). Anterior cingulate cortex, error
detection, and the online monitoring of performance. Science,
280(5364), 747–749.
Carter, C. S., MacDonald, A. W., III, Ross, L. L., & Stenger, V. A.
(2001). Anterior cingulate cortex activity and impaired self-monitoring of performance in patients with schizophrenia: An eventrelated fMRI study. American Journal of Psychiatry, 158(9),
1423–1428.
Clark, V. P., Fannon, S., Lai, S., Benson, R., & Bauer, L. (2000).
Responses to rare visual target and distractor stimuli using eventrelated fMRI. Journal of Neurophysiology, 83(5), 3133–3139.
Cohen, J. D., Botvinick, M. M., & Carter, C. S. (2000). Anterior
cingulate and prefrontal cortex: Whos in control?. Nature Neuroscience, 3(5), 421–423.
Coles, M. G. H., Scheffers, M. K., & Fournier, L. (1995). Where did
you go wrong? Errors, partial errors, and the nature of human
information processing. Acta Psychologica, 90(1–3), 129–144.
Coles, M. G. H., Scheffers, M. K., & Holroyd, C. B. (2001). Why is
there an ERN/Ne on correct trials? Response representations,
stimulus-related components, and the theory of error-processing.
Biological Psychology, 56(3), 173–189.
Cools, R., Clark, L., Owen, A. M., & Robbins, T. W. (2002). Defining
the neural mechanisms of probabilistic reversal learning using
event-related functional magnetic resonance imaging. Journal of
Neuroscience, 22(11), 4563–4567.
Dehaene, S., Posner, M. I., & Tucker, D. M. (1994). Localization of a
neural system for error detection and compensation. Psychological
Science, 5(5), 303–305.
Delgado, M. R., Nystrom, L. E., Fissell, C., Noll, D. C., & Fiez, J. A.
(2000). Tracking the hemodynamic responses to reward and
punishment in the striatum. Journal of Neurophysiology, 84(6),
3072–3077.
Della-Maggiore, V., Chau, W., Peres-Neto, P. R., & McIntosh, A. R.
(2002). An empirical comparison of SPM preprocessing parameters
to the analysis of fMRI data. NeuroImage, 17(1), 19–28.
Dien, J., Spencer, K. M., & Donchin, E. (2003). Localization of
the event-related potential novelty response as defined by
principal components analysis. Cognitive Brain Research, 17(3),
637–650.
Downar, J., Crawley, A. P., Mikulis, D. J., & Davis, K. D. (2000). A
multimodal cortical network for the detection of changes in the
sensory environment. Nature Neuroscience, 3(3), 277–283.
Falkenstein, M., Hoormann, J., Christ, S., & Hohnsbein, J. (2000).
ERP components on reaction errors and their functional significance: A tutorial. Biological Psychology, 51(2–3), 87–107.
Gehring, W. J., Goss, B., Coles, M. G. H., Meyer, D. E., & Donchin,
E. (1993). A neural system for error detection and compensation.
Psychological Science, 4(6), 385–390.
Gehring, W. J., & Willoughby, A. R. (2002). The medial frontal cortex
and the rapid processing of monetary gains and losses. Science,
295(5563), 2279–2282.
Gruber, O., & Goschke, T. (2004). Executive control emerging from
dynamic interactions between brain systems mediating language,
276
V. van Veen et al. / Brain and Cognition 56 (2004) 267–276
working memory and attentional processes. Acta Psychologica,
115(2–3), 105–121.
Holroyd, C. B., & Coles, M. G. H. (2002). The neural basis of human
error processing: Reinforcement learning, dopamine, and the errorrelated negativity. Psychological Review, 109(4), 679–709.
Holroyd, C. B., Dien, J., & Coles, M. G. H. (1998). Error-related scalp
potentials elicited by hand and foot movements: Evidence for an
output-independent error-processing system in humans. Neuroscience Letters, 242(2), 65–68.
Holroyd, C. B., Nieuwenhuis, S., Yeung, N., Nystrom, L. E., Mars, R.
B., Coles, M. G. H., et al. (2004). Dorsal anterior cingulate cortex
is sensitive to internal and external sources of error information: A
functional magnetic resonance imaging study. Nature Neuroscience,
7(5), 497–498.
Hopfield, J. J. (1982). Neural networks and physical systems with
emergent collective computational abilities. Proceedings of the
National Academy of Sciences of the United States of America, 79,
2554–2558.
Jeannerod, M., & Frak, V. (1999). Mental imaging of motor activity in
humans. Current Opinion in Neurobiology, 9(6), 735–739.
Jones, A. D., Cho, R. Y., Nystrom, L. E., Cohen, J. D., & Braver, T. S.
(2003). A computational model of anterior cingulate function in
speeded response tasks: Effects of frequency, sequence, and conflict.
Cognitive, Affective, & Behavioral Neuroscience, 2(4), 300–317.
Kerns, J. G., Cohen, J. D., MacDonald, A. W., III, Cho, R. Y., Stenger,
V. A., & Carter, C. S. (2004). Anterior cingulate conflict monitoring
and adjustments in control. Science, 303(5660), 1023–1026.
Kiehl, K. A., Laurens, K. R., Duty, T. L., Forster, B. B., & Liddle, P.
F. (2001). An event-related fMRI study of visual and auditory
oddball tasks. Journal of Psychopysiology, 15(4), 221–240.
Kiehl, K. A., Liddle, P. F., & Hopfinger, J. B. (2000). Error processing
and the rostral anterior cingulate: An event-related fMRI study.
Psychophysiology, 33, 282–294.
Knutson, B., Westdorp, A., Kaiser, E., & Hommer, D. (2000). FMRI
visualization of brain activity during a monetary incentive delay
task. NeuroImage, 12(1), 20–27.
Liotti, M., Woldorff, M. G., Perez, R., III, & Mayberg, H. S. (2000).
An ERP study of the temporal course of the Stroop color-word
interference effect. Neuropsychologia, 38(5), 701–711.
Luu, P., Tucker, D. M., Derryberry, D., Reed, M., & Poulsen, C.
(2003). Electrophysiological responses to errors and feedback in the
process of action regulation. Psychological Science, 14(1).
MacLeod, C. M. (1991). Half a century of research on the iStroop
effect: An integrative review. Psychological Bulletin, 109(2),
163–203.
Mathalon, D. H., Whitfield, S. L., & Ford, J. M. (2003). Anatomy of
an error: ERP and fMRI. Biological Psychology, 64(1–2), 119–141.
Menon, V., Adleman, N. E., White, C. D., Glover, G. H., & Reiss, A.
L. (2001). Error-related brain activation during a go/nogo response
inhibition task. Human Brain Mapping, 12(3), 131–143.
Milham, M. P., Banich, M. T., & Barad, V. (2003). Competition for
priority in processing increases prefrontal cortexs involvement in
top-down control: An event-related fMRI study of the Stroop task.
Cognitive Brain Research, 17(2), 212–222.
Milham, M. P., Banich, M. T., Webb, A., Barad, V., Cohen, N. J.,
Wszalek, T., et al. (2001). The relative involvement of anterior
cingulate and prefrontal cortex in attentional control depends on
nature of conflict. Cognitive Brain Research, 12, 467–473.
Miltner, W. H. R., Braun, C. H., & Coles, M. G. H. (1997). Eventrelated brain potentials following incorrect feedback in a timeestimation task: Evidence for a generic neural system for error
detection. Journal of Cognitive Neuroscience, 9(6), 788–798.
Monchi, O., Petrides, M., Petre, V., Worsley, K., & Dagher, A. (2001).
Wisconsin Card Sorting revisited: Distinct neural circuits participating in different stages of the task identified by event-related
functional magnetic resonance imaging. Journal of Neuroscience,
21(19), 7733–7741.
Nieuwenhuis, S., Holroyd, C. B., Mol, N., & Coles, M. G. H. (2004).
Reinforcement-related brain potentials from medial frontal cortex:
Origins and functional significance. Neuroscience & Biobehavioral
Reviews, 28(4), 441–448.
Nieuwenhuis, S., Ridderinkhof, K. R., Talsma, D., Coles, M. G. H.,
Holroyd, C. B., Kok, A., et al. (2002). A computational account of
altered error processing in older age: Dopamine and the errorrelated negativity. Cognitive, Affective, & Behavioral Neuroscience,
2(1), 19–36.
Peters, M., & de Munck, J. (1990). On the forward and the inverse
problem for EEG and MEG. In F. Grandori, M. Hoke, & G. L.
Romani (Eds.). Auditory evoked magnetic fields and electric
potentials. Advances in audiology (Vol. 6, pp. 70–102). Basel:
Karger.
Robertson, D. A., Gernsbacher, M. A., Guidotti, S. J., Robertson, R.
R. W., Irwin, W., Mock, B. J., et al. (2000). Functional
neuroanatomy of the cognitive process of mapping during
discourse comprehension. Psychological Science, 11(3), 255–260.
Ruchsow, M., Grothe, J., Spitzer, M., & Kiefer, M. (2002). Human
anterior cingulate cortex is activated by negative feedback:
Evidence from event-related potentials in a guessing task. Neuroscience Letters, 325, 203–206.
Ruff, C. C., Woodward, T. S., Laurens, K. R., & Liddle, P. F. (2001).
The role of the anterior cingulate cortex in conflict processing:
Evidence from reverse Stroop interference. NeuroImage, 14(5),
1095–9572.
Scheffers, M. K., Coles, M. G. H., Bernstein, P., Gehring, W. J., &
Donchin, E. (1996). Event-related brain potentials and errorrelated processing: An analysis of incorrect responses to go and nogo stimuli. Psychophysiology, 33(1), 42–53.
Scherg, M. (1990). Fundamentals of dipole source potential analysis.
In F. Grandori, M. Hoke, & G. L. Romani (Eds.). Auditory evoked
magnetic fields and electric potentials. Advances in audiology (Vol. 6,
pp. 40–69). Basel: Karger.
Srinivasan, R., Tucker, D. M., & Murias, M. (1998). Estimating the
spatial Nyquist of the human EEG. Behavior Research Methods,
Instruments and Computers, 30, 8–19.
Stroop, J. R. (1935). Studies of interference in serial verbal reactions.
Journal of Experimental Psychology, 18, 643–662.
Ullsperger, M., & von Cramon, D. Y. (2003). Error monitoring using
external feedback: Specific roles of the habenular complex, the
reward system, and the cingulate motor area revealed by functional
magnetic resonance imaging. Journal of Neuroscience, 23(10),
4308–4314.
Van Veen, V., & Carter, C. S. (2002a). The anterior cingulate as a
conflict monitor: fMRI and ERP studies. Physiology & Behavior,
77(4–5), 477–482.
Van Veen, V., & Carter, C. S. (2002b). The timing of actionmonitoring processes in the anterior cingulate cortex. Journal of
Cognitive Neuroscience, 14(4), 593–602.
Van Veen, V., Cohen, J. D., Botvinick, M. M., Stenger, V. A., &
Carter, C. S. (2001). Anterior cingulate cortex, conflict
monitoring, and levels of processing. NeuroImage, 14(6),
1302–1308.
Weissman, D. H., Giesbrecht, B., Song, A. W., Mangun, G. R., &
Woldorff, M. G. (2003). Conflict monitoring in the human anterior
cingulate cortex during selective attention to global and local object
features. NeuroImage, 19(4), 1361–1368.
West, R. (2003). Neural correlates of cognitive control and conflict
detection in the Stroop and digit-location tasks. Neuropsychologia,
41(8), 1122–1135.
West, R., & Alain, C. (2000). Effects of task context and fluctuations of
attention on neural activity supporting performance of the Stroop
task. Brain Research, 873, 102–111.
Yeung, N., Botvinick, M. M., & Cohen, J. D. (in press). The neural
basis of error detection: Conflict monitoring and the error-related
negativity. Psychological Review.