Adaptive rescheduling of error monitoring in multitasking

Abstract

The concurrent execution of temporally overlapping tasks leads to considerable interference between the subtasks. This also impairs control processes associated with the detection of performance errors. In the present study, we investigated how the human brain adapts to this interference between task representations in such multitasking scenarios. In Experiment 1, participants worked on a dual-tasking paradigm with partially overlapping execution of two tasks (T1 and T2), while we recorded error-related scalp potentials. The error positivity (Pe), a correlate of higher-level error evaluation, was reduced after T1 errors but occurred after a correct T2-response instead. MVPA-based and regression-based single-trial analysis revealed that the immediate Pe and deferred Pe are negatively correlated, suggesting a trial-wise trade-off between immediate and postponed error processing. Experiment 2 confirmed this finding and additionally showed that this result is not due to credit-assignment errors in which a T1 error is falsely attributed to T2. For the first time reporting a Pe that is temporally detached from its eliciting error event by a considerable amount of time, this study illustrates how reliable error detection in dual-tasking is maintained by a mechanism that adaptively schedules error processing, thus demonstrating a remarkable flexibility of the human brain when adapting to multitasking situations.

Full text

NeuroImage 232 (2021) 117888
Contents lists available at ScienceDirect
NeuroImage
journal homepage: www.elsevier.com/locate/neuroimage
Adaptive rescheduling of error monitoring in multitasking
Robert Steinhauser
∗
, Marco Steinhauser
Department of Psychology, Catholic University of Eichstätt-Ingolstadt, Ostenstraße 25, 85072 Eichstätt, Germany
          
Keywords:
Cognitive control
Error detection
Event-related potentials
Multitasking
       
The concurrent execution of temporally overlapping tasks leads to considerable interference between the subtasks.
This also impairs control processes associated with the detection of performance errors. In the present study, we
investigated how the human brain adapts to this interference between task representations in such multitasking
scenarios. In Experiment 1, participants worked on a dual-tasking paradigm with partially overlapping execution
of two tasks (T1 and T2), while we recorded error-related scalp potentials. The error positivity (Pe), a correlate of
higher-level error evaluation, was reduced after T1 errors but occurred after a correct T2-response instead. MVPA-
based and regression-based single-trial analysis revealed that the immediate Pe and deferred Pe are negatively
correlated, suggesting a trial-wise trade-o between immediate and postponed error processing. Experiment 2
conrmed this nding and additionally showed that this result is not due to credit-assignment errors in which
a T1 error is falsely attributed to T2. For the rst time reporting a Pe that is temporally detached from its
eliciting error event by a considerable amount of time, this study illustrates how reliable error detection in
dual-tasking is maintained by a mechanism that adaptively schedules error processing, thus demonstrating a
remarkable exibility of the human brain when adapting to multitasking situations.
1. Introduction
Adaptive human behavior requires an error monitoring system that
constantly checks whether an error has occurred and initiates adjust-
ments to cognition and behavior accordingly. Numerous studies could
show that such a system exists in the human brain, detecting errors
fast and reliably ( Ullsperger et al., 2014 ). Most studies, however, in-
vestigated error monitoring when single tasks were executed in isola-
tion. In contrast, everyday behavior is characterized by the concurrent
execution of temporally overlapping tasks. As dual-tasking is associ-
ated with decrements in task performance ( Pashler, 1994 ; Tombu and
Jolic œ ur, 2003 ), it is plausible to assume that also error monitoring suf-
fers under these conditions. Indeed, recent evidence suggests that dual-
tasking leads to specic impairments to the neural correlates of error
monitoring ( Klawohn et al., 2016 ; Weißbecker-Klaus et al., 2016 ). In the
present study, we investigate how the brain adapts to these dual-tasking
conditions. By considering error-related brain activity in event-related
potentials (ERPs), we ask how monitoring processes are reorganized to
maintain reliable error detection when the brain is confronted with two
temporally overlapping tasks.
Close temporal succession of more than one task leads to con-
siderable interference between the subtasks, resulting in performance
decrements compared to the execution of single tasks ( Jersild, 1927 ;
Telford, 1931 ). As dual-task interference arises predominantly in cen-
∗ Corresponding author.
E-mail address: robert.steinhauser@ku.de (R. Steinhauser).
tral, decision-related processes, the brain adapts to this interference by
serializing these processes, granting attentional resources to only one
task representation at a time ( Logan and Gordon, 2001 ; Meyer and
Kieras, 1997 ; Pashler, 1994 ). Neuroscientic studies suggested that
this interference as well as the resulting serialization is linked to a
fronto-parietal network that is able to grant access to focused atten-
tion and conscious control to only one task representation, by syn-
chronizing distinct brain regions in a self-amplifying process. A con-
siderable number of ERP studies links this serialization process to the
P300 component ( Dehaene et al., 1998 ; Del Cul, Baillet, and Dehaene,
2007 ; Dell’Acqua et al., 2005 ; Gross et al., 2004 ; Hesselmann et al.,
2011 ; Sergent et al., 2005 ; Sergent and Dehaene, 2004 ; Sigman and De-
haene, 2008 ). For example, Sigman and Dehaene (2008) found in a dual-
tasking paradigm that neural correlates of early attentional processes
were executed in parallel for both tasks, whereas the P300 for the second
task (T2) was delayed until response selection for the rst task (T1), rep-
resented by its own P300, had been completed (see also Dell’Acqua et al.,
2005 ; Hesselmann et al., 2011 ). Recently, Marti et al. (2015) provided
evidence that this serialization of task selection is the result of a com-
petition between task-related processes when they strive for access to
a higher-level attentional workspace. Neural representations of the two
tasks repel each other in a way that initially, only processing of T1 is
enabled, and that of T2 is actively delayed.
It has recently been suggested that also processes involved in the de-
tection and evaluation of performance errors suer from dual-tasking
https://doi.org/10.1016/j.neuroimage.2021.117888
Received 26 October 2020; Received in revised form 11 February 2021; Accepted 15 February 2021
Available online 27 February 2021
1053-8119/© 2021 The Author(s). Published by Elsevier Inc. This is an open access article under the CC BY license ( http://creativecommons.org/licenses/by/4.0/ )

R. Steinhauser and M. Steinhauser NeuroImage 232 (2021) 117888
interference. Research has robustly identied two ERPs associated with
distinct stages of error monitoring, which are commonly reported to be
executed in a xed temporal cascade immediately following the erro-
neous response ( Ullsperger et al., 2014 ) –the error-related negativity
(Ne/ERN, Falkenstein et al., 1991 ; Gehring et al., 1993 ) and the error
positivity (Pe, Overbeek et al., 2005 ). The Ne/ERN, an early fronto-
central negativity on error trials, has been assumed to reect a mis-
match between the correct and the actual response ( Coles et al., 2001 ),
response conict ( Yeung et al., 2004 ), or prediction errors that accom-
pany performance errors ( Holroyd and Coles, 2002 ). The Pe, on the
other hand, is a later positive deection over parietal electrodes, which
is assumed to reect higher-level aspects of error processing that are
associated with, or lead to conscious error detection ( Endrass et al.,
2007 ; Overbeek et al., 2005 ). The Pe has been linked to an accumu-
lation process of evidence that an error has occurred ( Murphy et al.,
2015 ; M. Steinhauser and Yeung, 2010 , 2012 ) but also to metacognitive
concepts such as the graded condence in a correct decision ( Boldt and
Yeung, 2015 ), motivational aspects ( Drizinsky et al., 2016 ; Kim et al.,
2017 ; Moser et al., 2011 ; Schroder and Moser, 2014 ), and the negative
aective response to the error ( Falkenstein et al., 2000 ; van Veen and
Carter, 2002 ).
Based on the shared time course and scalp topographies of the P300
and the Pe ( Leuthold and Sommer, 1999 ; Overbeek et al., 2005 ), it
has been suggested that the stimulus-locked P300 and the response-
locked Pe are eventually based on the same neurocognitive processes
( Ridderinkhof et al., 2009 ). In fact, both components have been linked
to joint activation in bilateral prefrontal and parietal brain regions
( Dehaene et al., 1998 ; Hester et al., 2005 ; Sergent and Dehaene, 2004 ;
Sigman and Dehaene, 2008 ). This can well explain why particularly
the Pe is aected by interference from competing tasks: the Pe inter-
feres with subsequent stimulus processing ( Buzzell et al. 2017 , but see
Beatty et al., 2018 ) and only the Pe has been shown to be impaired when
two temporally overlapping tasks have to be executed ( Weißbecker-
Klaus et al., 2016 ).
Whereas the similarity between the Pe and P300 can explain why
specically the Pe suers from dual-tasking interference, little is known
how the brain maintains the ability to reliably detect errors under dual-
tasking. One solution to this problem would be to serialize not only
task processing but also the accompanying error monitoring processes
( Hochman and Meiran, 2005 ). Processing a T1 error could lead to an
additional deferment of T2 processing, an eect that has indeed been
demonstrated in behavioral studies ( Jentzsch and Dudschig, 2009 ; M.
Steinhauser et al., 2017 ). However, error processing has been shown
to considerably interfere with subsequent stimulus processing, partic-
ularly when there is little time between the tasks ( Beatty et al., 2018 ;
Buzzell et al., 2017 ; Van der Borght, Schevernels, Burle, and Notebaert,
2016 ) and, moreover, error signaling is strongly reduced when an er-
roneous response is quickly followed by another task ( Rabbitt, 2002 ).
Together with the observation of an altogether reduced –not merely
delayed –Pe after the erroneous subtask in dual-tasking ( Weißbecker-
Klaus et al., 2016 ), this suggests that such a serialization cannot fully
protect error monitoring from dual-task interference. Consequently, it
might be necessary to actively defer particularly the more resource-
consuming aspects of error processing to the end of a dual-task scenario.
In this case, processing a T1 error would at least to some degree not oc-
cur until the response to T2. Such an adaptive rescheduling of conscious
error processing would imply that the neural processes that underly the
Pe can be detached from early error signals represented by the Ne/ERN.
The present study investigated this adaptive rescheduling hypoth-
esis, which proposes that the brain adapts to dual-task interference in
error monitoring by deferring higher-level aspects of error processing to
after completion of the dual task. To this end, we focused on T1 errors
but analyzed ERPs following both the T1-response and the T2-response.
We contrasted trials with a suciently long interval between the two
task stimuli that allowed serial task execution (1200 ms) with trials that
required overlapping task execution due to a considerably shorter in-
terval (300 ms). In serial task execution, we expected to see the typi-
cal Ne/ERN and Pe after the erroneous T1-response. Crucially, in over-
lapping task execution, we predicted that the Pe after the T1-response
would be reduced or even absent while a Pe would now be obtained after
the correct T2-response. Such a result would demonstrate that higher-
level aspects of T1 error processing are deferred at least partially until
the completion of T2, and extend previous accounts on an active defer-
ment of component processes in dual-tasking ( Hesselmann et al., 2011 ;
Sigman and Dehaene, 2008 ; R. Steinhauser and Steinhauser, 2018 ) to
the area of performance monitoring.
2. Material and methods
2.1. Participants
We conducted two separate, consecutive experiments with dier-
ent groups of participants. 24 healthy students (2 male; age: M = 22.3
years; SD = 4.5 years; 2 left-handed) participated in Experiment 1, and
24 healthy students (2 male; age: M = 24.0 years; SD = 3.9; 2 left-
handed) participated in Experiment 2. With an assumed correlation of
0.5 among cell means, a sample size of N = 24 should allow for detect-
ing medium-sized eects ( f = 0.25) in the present repeated-measures
design with a statistical power of 0.82. All participants were recruited
from the Catholic University of Eichstätt-Ingolstadt and received course
credit or payment (8 €per hour). Informed consent was provided by all
participants and the study was approved by the ethics committee of the
Catholic University of Eichstätt-Ingolstadt. One subject in Experiment
2 had to be excluded from further analysis due to technical problems
during data acquisition.
2.2. Task and procedure
Experiment 1 . Adopting the PRP paradigm from
Steinhauser et al. (2017) , we combined an error-prone three-choice
color anker task and a two-choice pitch discrimination task. The
anker task stimulus (see Fig. 1 ) consisted of three horizontally ar-
ranged squares of 0.82° edge length. The central target square and the
two anker squares were either red, yellow or blue. While both anker
squares had the same color, this color was always dierent from that
of the target. As three colors were used, it was possible to exclusively
present these more error-prone incongruent stimuli without enabling
participants to infer the target color from the ankers. Stimuli for the
secondary pitch task were low (400 Hz) and high (900 Hz) sine tones.
The trial procedure is depicted in Fig. 1 . Participants rst had to
respond to the anker task (T1) and then to the pitch discrimination
task (T2). Each trial started with the presentation of a xation cross for
500 ms. Following this, the anker task was presented by initially dis-
playing the two ankers for 60 ms and then the target together with
the ankers for 200 ms. The pitch discrimination task started after an
SOA of either 300 ms (short) or 1200 ms (long), which was chosen ran-
domly within the blocks. The stimulus for the pitch task was presented
for 150 ms. After responses to both tasks were given, the xation cross
of the next trial was presented 500 ms after the response. The short SOA
was supposed to create a dual-tasking situation with overlapping task
execution while at the same time preventing excessive response group-
ing, which was observed in a piloting phase with even shorter SOAs. In
contrast, the long SOA was intended to serve as a baseline that allows
for serial task execution.
For the anker task, participants had to indicate the color of the
central target square by pressing the “Y ”, “X ”, or “C ” button on a Ger-
man QWERTZ keyboard (in which Y and Z are exchanged) with their
left hand (the T1-response). For the pitch discrimination task, partic-
ipants entered their response with “arrow down ”for a low pitch and
“arrow up ”for a high pitch (the T2-response). The mapping of colors
to keys was counterbalanced across participants, and for one half of the
participants, the task-to-hand assignment was reversed (resulting in “, ”,
2

R. Steinhauser and M. Steinhauser NeuroImage 232 (2021) 117888
Fig. 1. Time course of two example trials. On each trial, participants had to
respond to two tasks that were separated by an unpredictable stimulus onset
asynchrony of 300 ms (overlapping task execution, A) or 1200 ms (serial task
execution, B). A three-choice color variant of the anker task, in which the color
of the central square had to be indicated (red, yellow, blue), served as Task 1. A
two-choice pitch discrimination task was presented subsequently as Task 2, with
an auditory sine tone stimulus of 400 Hz or 900 Hz. Participants were instructed
to respond as fast as possible to both tasks, in the given order. S1 = Stimulus
1. S2 = Stimulus 2. R1 = Response 1. R2 = Response 2. RT1 = response time
to Task 1. RT2 = response time to Task 2. (For interpretation of the references
to color in this gure legend, the reader is referred to the web version of this
article.)
“. ”, “- ”for the anker task and “A ”and “Y ”for the pitch task). If the
T2-response was given before the onset of the pitch stimulus, written
feedback ( “AUFGABE 2 ZU FRÜH! ”, engl . task 2 too early) was given
immediately and the trial was excluded from data analysis.
The experiment started with a series of practice blocks that ensured
that participants had thoroughly learned the experimental paradigm and
the mappings of colors/pitches to keys. First, in three blocks of 24 trials
the anker task was presented alone. Then, one block of 16 trials served
to practice the pitch task alone, and in one subsequent block of 36 trials,
both tasks were responded to together. Actual testing consisted of 10
blocks of 108 trials each, resulting in a total number of 1080 trials. Oral
feedback to respond faster was given after blocks, if the error rate for
any of the tasks was below 10%.
Experiment 2 . Experiment 2 adopted the experimental method of Ex-
periment 1 but added a trial-wise report of errors. To this end, response
keys to the two tasks were changed ( “A ”, “S ”, “D ”and “L ”, “P ”for one
half of participants, “L ”, “Ö”, “Ä” and “A ”, “Q ”for the other half) and
two additional keys, “alt ”and “alt gr ”, served to report errors. In each
trial, 600 ms after the response to Task 2, an additional instruction
( “FEHLER? ”, engl . error?) appeared for 1000 ms in the center of the
screen. During this time, participants should indicate if they had just
committed an error in the task of their left hand ( “ALT ”) or their right
hand ( “ALT GR ”), or both. Error reports were also counted after the
1000 ms (i.e., during the inter-trial interval), this aected 6% of the
error reports. No response was required if the participants considered
both their responses correct. To account for the increased trial dura-
tion, Experiment 2 was restricted to only the short SOA of 300 ms. With
10 blocks of 84 trials each, this resulted in 840 trials for the condition
with overlapping task execution.
2.3. Data acquisition
The EEG was recorded from 64 electrodes at a sample rate of 512 Hz
with a BioSemi Active-Two system (BioSemi, Amsterdam, The Nether-
lands; channels Fp1, AF7, AF3, F1, F3, F5, F7, FT7, FC5, FC3, FC1, C1,
C3, C5, T7, TP7, CP5, CP3, CP1, P1, P3, P5, P7, P9, PO7, PO3, O1, Iz,
Oz, POz, Pz, CPz, Fpz, Fp2, AF8, AF4, AFz, Fz, F2, F4, F6, F8, FT8, FC6,
FC4, FC2, FCz, Cz, C2, C4, C6, T8, TP8, CP6, CP4, CP2, P2, P4, P6, P8,
P10, PO8, PO4, O2 as well as the left and right mastoid). The CMS (Com-
mon Mode Sense) and DRL (Driven Right Leg) electrodes were used as
reference and ground electrodes. Vertical and horizontal electrooculo-
gram (EOG) was recorded from electrodes above and below the right
eye and on the outer canthi of both eyes. All electrodes were o-line
re-referenced to averaged mastoids.
2.4. Experimental design and statistical analysis
Design. Each trial was assigned to a condition based on the SOA
(short, long) and T1 Correctness (correct, error). T1 Correctness relied
on the post-hoc classication of whether the response in the anker task
(T1) was correct or not. Trials on which an error in the pitch discrim-
ination task (T2) occurred were removed from all analyses of response
times (RTs) and ERP data.
Data analysis . For the analysis of RTs, trials were excluded whose RT
deviated more than three standard deviations from the RT mean of each
condition and participant. Error rates were arcsine-transformed prior
to statistical testing ( Winer et al., 1991 ). All analyses were performed
using custom MATLAB v8.2 (The Mathworks, Natick, MA) scripts and
EEGLAB v12.0 ( Delorme and Makeig, 2004 ) functions. All data and
analysis scripts are publicly available in an online repository hosted
by the Open Science Framework ( https://osf.io/5ub8z/ ). For both Ex-
periments, continuous EEG data was initially band-pass ltered to ex-
clude frequencies below 0.1 Hz and above 40 Hz. Then, epochs were
created for two separate analyses, rst from − 500 ms before to 1000 ms
after the T1-response (T1-response-locked dataset) and, second, from
− 500 ms before to 1000 ms after the T2-response (T2-response-locked
dataset). Epochs in both analyses were baseline-corrected by subtract-
ing mean activity between − 150 ms and − 50 ms before the response,
as neural correlates of performance monitoring were previously found
to emerge slightly before the response button press. Separately for each
dataset, electrodes were interpolated using spherical spline interpola-
tion if they met the joint probability criterion (threshold 5) as well
as the kurtosis criterion (threshold 5) in EEGLAB’s channel rejection
routine (pop_rejchan.m). Epochs were removed that contained activity
exceeding + / − 300 𝜇V in any channel except AF1, Fp1, Fpz, Fp2, AF8
(to prevent exclusion of blink artifacts, which were corrected at a later
stage) and whose joint probability deviated more than 5 standard de-
viations from the epoch mean. To correct for eye blinks and muscular
artefacts, an infomax-based ICA ( Bell and Sejnowski, 1995 ) was com-
puted and components with time courses and topographies typical for
these artefacts were removed after visual inspection.
In both experiments and for each dataset (T1-response-locked
and T2-response-locked), following a considerable number of previ-
ous studies that feature a Pe peaking over parieto-occipital electrodes
( Beatty et al., 2018 ; Endrass et al., 2007 ; Shalgi et al., 2009 ; M.
Steinhauser and Yeung, 2010 , 2012 ), the Pe was quantied by com-
paring mean amplitudes from 200 ms to 400 ms after the respective
response at electrode POz between T1 error trials and correct trials. The
Ne/ERN was quantied by comparing mean amplitudes from 0 ms to
100 ms after the respective response at electrode FCz between T1-error
trials and correct trials. In Experiment 2, these analyses were restricted
to T1-error trials that were reported as T1 errors, and correct trials that
were reported as correct trials.
As we found that T1 errors elicited both an immediate Pe in T1-
response-locked data and a deferred Pe in T2-response-locked data, we
aimed to investigate the relationship between both types of Pe on a
single-trial level. To acquire a robust single-trial estimate of the T1-
response-locked Pe, we used a multivariate pattern analysis (MVPA)
based on the linear integration method introduced by Parra et al.
(2002 , 2005 ). This method has previously been used to quantify error-
related brain activity on a single-trial level ( Boldt and Yeung, 2015 ;
3

R. Steinhauser and M. Steinhauser NeuroImage 232 (2021) 117888
M. Steinhauser and Yeung, 2010 ). Here, we provide only a brief de-
scription of this method while details can be found elsewhere (e.g.,
Steinhauser and Yeung 2010 ). In a rst step, we computed a set of clas-
siers on T1-response-locked data that discriminated optimally between
correct trials and T1 errors. Classiers were constructed for consecutive,
partially overlapping time windows from 0 ms to 400 ms after the T1-
response (width 50 ms, step size 10 ms). All classiers were trained,
separately for each participant, on T1 error trials and the same num-
ber of randomly drawn correct trials. Then, the classier was selected
that featured the highest discrimination sensitivity, as indicated by the
Az score. To prevent overtting, Az was computed using leave-one-out
cross-validation. Using this classier, we calculated prediction values
for each error trial. These prediction values represent single-trial esti-
mates of error-related brain activity in the respective classier window.
Based on these estimates, we tted a linear regression model on the data
of every participant:
𝑃 𝑒
𝑅 2
= 𝛽0
+ 𝛽𝑀𝑉 𝑃 𝐴 _ 𝑃 𝑒
𝑅 1
∗ 𝑃 𝑒
𝑅 1 _ 𝑀𝑉 𝑃𝐴
where Pe
R2
represents the z-scored trial-wise mean amplitude of the T2-
response-locked Pe and Pe
R1_MVPA
represents the (already normalized)
trial-wise amplitude of the immediate Pe as estimated by the MVPA
prediction value. Importantly, the same baseline period was used for the
T2-response-locked ERPs as for T1-response-locked data, i.e., for each
trial, mean activity between − 150 ms and − 50 before the response to T1
was subtracted. Participants’ regression coecients were subsequently
tested against zero by means of a Student’s t -test.
As a second way of investigating a possible trade-o between the
R1-locked Pe and the R2-locked Pe, we computed an analysis based on
linear regression on raw data. To get a suciently robust measure of
the Pe in single trials, we computed mean amplitudes of trial-wise EEG
data in a 100 ms window on a cluster of posterior electrodes (Pz, P1,
P2, POz, P03, P04). This was done in steps of 10 ms on time windows
centering from 0 ms to 400 ms after both responses. For each such com-
bination of single-trial Pe values in R1-locked data and R2-locked data,
the following linear regression model was tted on the data of every
participant:
𝐸 𝐸 𝐺
𝑅 2
= 𝛽0
+ 𝛽𝐸 𝐸 𝐺
𝑅 1
∗ 𝐸 𝐸 𝐺
𝑅 1
+ 𝛽𝐼𝑅𝐼
∗ 𝐼𝑅𝐼
where EEG
R2
represents z-scored posterior response 2-locked EEG activ-
ity, EEG
R1
represents z-scored posterior response 1-locked EEG activity,
IRI represents the z-scored inter-response interval of the current trial.
All z-scores were computed within-subject (i.e., using subject-specic
across-trial mean and standard deviations) and 𝛽i
represent within-
subject regression coecients. Participants’ regression coecients were
subsequently standardized by their SDs and tested against zero by means
of Student’s t-tests. Due to the large number of the resulting t-tests, we
corrected for multiple comparisons by means of a cluster-based permu-
tation test with 100.000 permutations, a cluster inclusion threshold of
p = .01 and an output threshold of p = .05, utilizing the Mass Univariate
ERP Toolbox ( Groppe et al., 2011 ).
3. Results
3.1. Experiment 1
24 healthy adult participants worked on a variant of the psycho-
logical refractory period (PRP) paradigm, which is commonly used to
investigate mutual interference between subtasks in dual-tasking situ-
ations ( Pashler, 1994 ; Tombu and Jolic œ ur, 2003 ). The details of the
experimental paradigm can be found in Fig. 1 .
Behavior . RTs and error rates are depicted in Fig. 2 . To verify that
our paradigm creates a dual-tasking scenario with overlapping task ex-
ecution, we rst examined whether two typical eects of dual-tasking
can be observed in this dataset: First, the so-called PRP eect refers to
the observation that RTs to T2 increase with a decreasing stimulus onset
asynchrony (SOA), and thus indicates a form of dual-task cost. This ef-
fect is typically explained by the idea that, with more overlap between
tasks, T2 execution is delayed ( Pashler, 1994 ) or suers from depleted
resources ( Tombu and Jolic œ ur, 2003 ). Second, it has recently been
shown that T1 errors lead to increased RTs to T2, and this post-error
slowing is larger for short SOAs than for long SOAs ( Steinhauser et al.,
2017 ). This phenomenon presumably reects interference between T1
error monitoring and execution of T2, which again is larger with short
SOAs.
To analyze both eects in our data, we subjected RTs for T1 and T2
separately to repeated measures ANOVAs on the variables SOA (short
vs. long) and T1 Correctness (correct vs. error). Characteristic of PRP
paradigms, no signicant eects were found for RTs to T1, F s(1, 23) ⟨
2.70, p s ⟩0.11, 𝜂p
2 < 0.11. In contrast, RTs to T2 were considerably
slower on trials with short SOA than with long SOA, F (1, 23) = 83.4,
p < .001, 𝜂p
2 = 0.78, reecting the typical PRP eect. In addition,
RTs to T2 were slower when the preceding T1-response was an error,
F (1,23) = 56.8, p < .001, 𝜂p
2 = 0.71, and as expected, this post-error
slowing on T2 was increased with short SOA, F (1, 23) = 25.9, p < .01,
𝜂p
2
= 0.53. Thus, our data replicate well-known signatures of dual-task
interference, which demonstrates that the short-SOA condition induced
overlapping task execution. In addition, an analysis of inter-response in-
tervals (IRIs, the time between the T1 response and the T2 response) in
the short-SOA condition ruled out that participants had grouped their re-
sponses. While IRIs were considerably longer on T1 errors (587 ms) com-
pared to corrects (381 ms) due to the above mentioned post-error slow-
ing, t (23) = 8.35, p < .001, d = 1.25, IRIs in both conditions were far be-
yond the commonly used threshold for response grouping (i.e., simulta-
neously pressing both response buttons) of 50 – 100 ms ( Hommel, 1998 ;
Pashler and Johnston, 1989 ; Welford, 1952 ).
Error rates for T1 were high enough for an analysis of error-related
brain activity. A mean T1 error rate of 8.31% resulted in an average
number of 98.3 trials with T1 errors but correct T2 responses per partic-
ipant. A repeated measures ANOVA on the variables SOA (short vs. long)
and Task (T1 vs. T2) yielded a signicant interaction, F (1, 23) = 8.06,
p = .009, 𝜂p
2
= 0.26, indicating that, whereas T1 errors were equally fre-
quent in short-SOA and long-SOA trials, T2 errors were more frequent in
short-SOA trials as compared to long-SOA trials, t (23) = 2.69, p = .013,
d = 0.50. This again demonstrates increased interference in the short-
SOA condition. Double errors (i.e., trials with incorrect responses to both
subtasks), occurred infrequently with a mean error rate of 1.85%. They
were only slightly more common in short-SOA trials than in long-SOA
trials, t (23) = 2.01, p = .056, d = 0.28.
Ne/ERN and Pe . To investigate our rescheduling hypothesis, we ana-
lyzed error-related brain activity in short-SOA trials, which induce over-
lapping task execution, and long-SOA trials, which allow for serial task
execution. Our central prediction was that, with a short SOA, the Pe as-
sociated with T1 errors should partially be rescheduled to the end of the
dual-task. This would result in a reduced Pe following incorrect T1 re-
sponses but the emergence of a Pe after correct T2 responses. We had no
explicit hypothesis on the Ne/ERN but nevertheless analyzed this com-
ponent to determine whether comparable eects can be found for this
early form of error processing.
We initially analyzed T1-response-locked ERPs to nd out about im-
mediate neural correlates of error processing after T1 errors. A distinct
parietal positivity, the Pe, was clearly observable for both SOA condi-
tions ( Fig 3 A). For an analysis of the Pe, we subjected mean amplitudes
at electrode POz to a repeated measures ANOVA on the variables SOA
(short vs. long) and T1 Correctness (correct vs. error). A Pe was evident
across both SOA conditions, as indicated by a main eect of T1 Correct-
ness, F (1, 23) = 28.6, p < .001, 𝜂p
2
= 0.55, but a signicant interaction
revealed that this Pe was far less pronounced in trials with short SOA
than in trials with long SOA, F (1, 23) = 11.0, p = .003, 𝜂p
2
= 0.32. Raw
ERP waves in Fig. 3 suggest that this interaction is mainly driven by a
dierence in the amplitudes of correct trials. This is likely caused by the
fact that in most trials of the short SOA condition, the stimulus of T2
4

R. Steinhauser and M. Steinhauser NeuroImage 232 (2021) 117888
Fig. 2. Behavioral results of Experiment 1. RTs to Task 1 and Task 2 are depicted in the left and middle panel, respectively. Error rates are presented in the right
panel. Error bars indicate within-subject standard errors of the mean ( Cousineau, 2005 ; Morey, 2008 ). RT = response time. SOA = stimulus onset asynchrony.
T1 = Task 1. T2 = Task 2.
Fig. 3. ERPs locked to the T1-response at posterior (A) and
frontocentral (B) electrodes. Dierence waves are computed
from the respective T1 error minus T1 correct raw ERPs. Scalp
topographies represent these dierence waves. Gray areas in-
dicate the time intervals for statistical testing for the Pe (A)
and the Ne/ERN (B). T1 = Task 1. SOA = stimulus onset asyn-
chrony.
is presented and processed sometime within the time window observed
here, eliciting stimulus-locked ERPs that are superimposed on the raw
ERP waveforms of both correct and error trials in the short SOA con-
dition (see Sigman and Dehaene, 2008 ). For this reason, we quantify
the Pe not from error trials alone but as the dierence between correct
and error trials. To additionally rule out that the observed interaction
may truly be rooted in dierences in correct trials, we correlated the
dierence wave of conditions correct long and correct short with the
dierence wave of the interaction term of the above analysis. We found
no signicant correlation, r = 0.14, p = .50, indicating that the Pe reduc-
tion in trials with short SOA is not a mere consequence of the amplitude
dierence in correct trials.
Fig. 3 B shows that also a clear frontocentral Ne/ERN was observable
in both SOA conditions. Indeed, subjecting mean amplitudes at electrode
FCz to a repeated measures ANOVA of the same variables as above re-
vealed that a signicant Ne/ERN across both SOAs was obtained, F (1,
23) = 22.8, p < .001, 𝜂p
2
= 0.50. In addition, a main eect of SOA demon-
strated that mean amplitudes of correct as well as error trials were more
negative in trials with short SOA, F (1, 23) = 13.5, p = .001, 𝜂p
2
= 0.37.
However, the interaction between both variables did not reach signi-
cance, F (1, 23) = 2.63, p = .12, 𝜂p
2
= 0.10.
Having shown that higher-level aspects of error processing as re-
ected by the Pe were largely impaired in the short-SOA condition,
we subsequently examined neural correlates of T1 error processing af-
ter completion of the whole dual-task, i.e., in T2-response-locked data
( Fig. 4 ). We investigated the possible emergence of such a deferred Pe
( Fig. 4 A) by means of ANOVAs on the variables SOA (short vs. long)
and T1 Correctness (T1 correct vs. T1 error). It must be noted that both
conditions of the variable T1 Correctness here represent trials whose
T2 was answered correctly, i.e., the “error-related ”brain activity re-
ported here was locked to a correct T2-response. Nonetheless, we ob-
tained a signicant interaction between SOA and T1 Correctness on Pe
5

R. Steinhauser and M. Steinhauser NeuroImage 232 (2021) 117888
Fig. 4. ERPs locked to the (correct) T2-response at posterior
(A) and frontocentral (B) electrodes. Dierence waves are com-
puted from the respective T1 error minus T1 correct raw ERPs.
Scalp topographies represent these dierence waves. Gray ar-
eas indicate the time intervals for statistical testing for the Pe
(A) and the Ne/ERN (B). T1 = Task 1. SOA = stimulus onset
asynchrony.
amplitudes, F (1, 23) = 5.55, p = .027, 𝜂p
2
= 0.19, reecting that correct
T2-responses elicited a signicant Pe on short-SOA trials, t (23) = 2.00,
p = .029, d = 0.27, but not on long-SOA trials.
The same analysis on the Ne/ERN amplitudes revealed only a
marginally signicant main eect of T1 Correctness, F (1, 23) = 3.48,
p = .075, 𝜂2
part.
= 0.13, indicating a negativity after T1 errors com-
pared to T1 correct trials, but no signicant interaction, F (1, 23) = 1.73,
p = .20. Visual inspection of the topographies ( Fig. 4 B) revealed that this
eect peaked earlier than the typical Ne/ERN and had a more frontal dis-
tribution than the Ne/ERN in T1-response-locked potentials (however,
the same analysis at electrode Fz showed similar results). This suggests
that, while there appears to be some frontocentral activity related to
the T1 error in T2-response-locked data, this eect lacks robustness and
diers from the usually observed Ne/ERN.
To sum up, we could provide evidence that overlapping task execu-
tion goes along with a reduction of the Pe to the incorrect T1 response
whereas a sizeable Pe emerges after the correct T2-response. We inter-
pret this deferred Pe as a rescheduling of higher-level error processing to
the end of the dual-task trial.
Relationship between immediate and deferred error processing. Although
the immediate Pe that followed the T1 response was strongly reduced in
the short-SOA condition, it was observable there nonetheless. This sug-
gests two possible ways of how higher-level error monitoring is carried
out under conditions of overlapping task execution. On the one hand,
error monitoring after the T2 response could occur predominantly on
those trials that also feature strong error monitoring immediately after
the T1 response. An actual rescheduling mechanism as outlined above,
which defers T1 error processing to after completion of T2, however,
would by denition require the exact opposite pattern: error monitor-
ing after the T2 response would have to occur on error trials that did not
exhibit considerable immediate error processing after T1. To distinguish
between these two possibilities, we examined the trial-wise relationship
between immediate and deferred Pe. This was done in two methodolog-
ically dierent ways, to combine their respective advantages and coun-
tervail their disadvantages.
First, we investigated the inverse relationship of immediate and de-
ferred error processing by deriving single-trial estimates for the R1-
locked and the R2-locked Pe from mean amplitudes in raw data and com-
paring them in a regression-based analysis. This analysis had to be lim-
ited to trials with IRI > 400 ms, however, because raw EEG data within
the same epoch are highly susceptible to autocorrelations in overlapping
time periods. Importantly, this does not aect the remaining analyses of
the present study, because there, EEG data from dierent, individually
baseline-corrected epochs are compared. The restriction to IRIs above
400 ms resulted in M = 31.04 trials per participant. Fig. 5 depicts strong
negative beta weights in a signicant cluster from 290 ms onwards in
R1-locked data and 170 ms in R2-locked data, that is, around the time
of the R1-locked and R2-locked Pe. This is evidence for a trade-o be-
tween the immediate and the delayed Pe, supporting the account that
error processing happens immediately after some erroneous responses,
whereas it is deferred on other trials.
Second, we used multivariate pattern analysis (MVPA) to create a
set of classiers over consecutive time windows that optimally distin-
guished between correct trials and T1 error trials based on T1-response-
locked brain activity. This approach allowed us to examine all trials irre-
spective of their IRI and is furthermore less susceptible to spurious non-
phase-locked activity in the EEG data ( Parra et al., 2002 ; Parra et al.,
2005 ). Mean classier accuracy peaked around 190 ms with an Az of
0.61 but inspection of participants’ individual classier accuracies re-
vealed that one participant (Subject 13) exhibited a remarkably low
classication accuracy of 0.34, which is more than 2.5 standard devia-
tions below the mean classication accuracy across all participants. This
apparently failed attempt to compute a successful MVPA for this partic-
ipant is likely due to the small number of 20 trials that entered the
6

R. Steinhauser and M. Steinhauser NeuroImage 232 (2021) 117888
Fig. 5. Regression weights of the regression-based single-trial analyses for Ex-
periments 1 and 2, in which amplitudes of R1-locked data predict amplitudes
of R2-locked data. Black lines indicate the borders of signicant clusters as re-
vealed by a cluster-based permutation test. Red dotted lines indicate the peak
amplitude of the Pe in the respective dierence waves. A signicant negative
association of the R1-locked Pe and R2-locked Pe becomes evident in both ex-
periments. (For interpretation of the references to color in this gure legend,
the reader is referred to the web version of this article.)
training set (the average training set size of all participants was 120.63
trials). For this reason, data from Subject 13 was excluded from sub-
sequent MVPA-based analyses. Nonetheless, all eects remain signi-
cant also when including that participant (all p s < 0.05). Based on the
peak classier window, we computed the prediction value for each trial.
These prediction values represent the degree to which each trial elicits
error-related brain activity in this time window ( Boldt and Yeung, 2015 ;
Steinhauser and Yeung, 2010 ), and thus represent a single-trial esti-
mate of the T1-response-locked Pe. We utilized these prediction values
in a regression-based analysis on z-scored single-trial amplitudes of the
deferred, T2-response locked Pe.
1
Across all participants, this analysis
yields a considerably negative mean beta weight of − 1.14, t (22) = 2.145,
p = .043, d = − 0.62, indicating larger prediction values (i.e. more pro-
nounced immediate error processing) being associated with smaller de-
ferred Pe amplitudes and vice versa. For better visualization, instead of
continuous data Figs. 6 A and 6 B displays ERPs that were computed from
3 equal-sized bins with small, medium and large immediate Pe (based on
the MVPA’s prediction values). This conceptually replicates the above
ndings on a trade-o between the immediate and deferred Pe from
raw ERP values of the immediate Pe and thus provides additional sup-
port for the idea that the rescheduling of higher-level error processing
occurs predominantly on trials with little immediate error processing.
This nding additionally rules out the above-mentioned alternative in-
terpretation that dierences in the T1-response-locked Pe with regard
to short and long SOA could be rooted only in dierences between the
correct trials in the two conditions. MVPA-based decoding of the degree
of immediate error processing directly modulates Pe amplitudes of error
trials as well ( Fig. 6 A).
The MVPA-based dierentiation of trials with a dierent degree of
immediate error processing eventually also allowed us to investigate
how this dierence aects selecting and executing the response to Task
2. A regression-based analysis of the prediction values of the MVPA (i.e.,
the estimate of the immediate Pe) and IRIs yielded a mean beta weight
of 0.091 across all participants, t (22) = 2.56, p = .018, d = 0.74, which
suggests larger IRIs on trials with a larger immediate Pe and vice versa,
1 Please note that for this particular analysis, baseline correction of T2-
response-locked data was computed on a time interval prior to the individual
trial’s T1-response (see Materials and methods). This was done to rule out the
possibility that the negative correlation of immediate and delayed Pe within the
same condition is due to dierences in the pre-T2-response baseline interval. A
T2-response-locked (deferred) Pe is also found in short-SOA trials in the main
analysis of both Experiments 1 and 2 when such a pre-R1-baseline is used (Exp.
1: t (23) = 4.23, p < .001; Exp. 2: t (22) = 2.09, p = .048). However, to compen-
sate for a possible carry-over of the R1-Pe on the R2Pe with such a baseline, we
utilized a pre-R2-baseline for the main analyses of the present study.
Fig. 6. ERPs at a posterior electrode locked to the T1-response (A) and T2-
response (B) and inter-response intervals after T1 errors (C). ERPs in Panel B
are trial-wise baseline-corrected for the same pre-T1-response baseline as those
in Panel A. For visualization purposes, T1 error trials are divided into three sep-
arate conditions – small, medium, and large immediate Pe –based on a single-
trial estimate of the Pe. Correct trials are presented for comparison (thin line).
A 20 Hz lowpass lter was applied for better visibility. Gray areas indicate the
time intervals for statistical testing for the Pe. T1 = Task 1. T2 = Task 2.
in turn supporting the idea that immediate error processing negatively
aects ecient execution of the second task. Again, for better visual-
ization, Fig. 6 C depicts IRIs based on a threefold bin-separation of the
underlying trials.
3.2. Experiment 2
Individual trials that elicit a deferred Pe with simultaneous omission
of an immediate Pe could also be rooted in the problem of credit as-
signment ( Fu and Anderson, 2008 ; Sutton and Barto, 1998 ; Walsh and
Anderson, 2011 ). It is conceivable that trials with a deferred Pe actually
reect that error monitoring occasionally misattributed internal error
signals to T2, and therefore falsely detected a T2 error. Rather than a
rescheduling of T1-error processing, such a Pe would represent imme-
diate T2-error processing. In a follow-up experiment with 24 dierent
participants (out of which one participant had to be excluded due to
technical diculties during data acquisition), we did not only want to
replicate our initial ndings but also investigated whether this alterna-
tive explanation could account for our ndings.
In Experiment 2, participants worked on the same PRP paradigm but
reported after each trial by key press, whether they had committed an
error in T1, T2, or both. This allowed us to distinguish between correctly
assigned T1 errors (T1 hits) and T1 errors that were mistakenly reported
as T2 errors (T2 false alarms). After presentation of the report cue, errors
were reported with a mean RT of 401 ms (mean SD = 290 ms). Rates of
averaged error reports are depicted in Fig. 7 A. In fact, only 4.9% of all
T1 errors were T2 false alarms (on average 2.1 trials per participant) and
thus occurred very rarely. As only 17.0% of T1 errors were reported to
be correct (T1 misses; on average 6.78 trials per participant), we could
base our further ERP analyses on a solid subset of 73.8% of T1 hits (on
average 31.0 trials per participant).
As in Experiment 1, the analysis of the Ne/ERN in T2-response-
locked data revealed only a marginally signicant negativity for T1 er-
rors relative to correct trials at electrode FCz, t (22) = 1.87, p = .075,
7

R. Steinhauser and M. Steinhauser NeuroImage 232 (2021) 117888
Fig. 7. Error reports (A) and ERP results (B) of Experiment 2.
Error bars in Panel A indicate within-subject standard errors of
the mean ( Cousineau, 2005 ; Morey, 2008 ). Panel B represents
ERPs locked to the T2-response at posterior (left) and fronto-
central (right) electrodes. Dierence waves (dotted lines) are
computed from T1 error minus correct raw ERPs. Scalp to-
pographies represent these dierence waves. Gray areas in-
dicate the time intervals for statistical testing for the Pe (left)
and the Ne/ERN (right). T1 = Task 1. T2 = Task 2.To com-
pensate for the increased trial duration due to trial-wise error
reporting, Experiment 2 was restricted to the short SOA con-
dition. Only T2-response-locked data are reported here as no
immediate Ne/ERN and Pe could be compared between SOA
conditions. Although only T1 hits were included in the anal-
ysis, a clear deferred Pe could be observed after T2-response-
locked data ( Fig. 6 B), t (22) = 2.86, p = .009, d = 0.57. (For
interpretation of the references to color in this gure legend,
the reader is referred to the web version of this article.)
d = 0.51. However, inspection of the topographies ( Fig. 7 B) showed that
the peak of the Ne/ERN was, as in Experiment 1, again at more frontal
electrodes. Correspondingly, additional testing was conducted at elec-
trode Fz and now revealed a signicant deferred Ne/ERN, t (22) = 2.26,
p = .034, d = 0.62. Nonetheless, this eect needs to be treated with
caution due to the post-hoc nature of testing.
The regression-based approach on raw data with IRI > 400 ms
yielded a pattern with close similarity to that in Experiment 1, albeit
somewhat earlier in R2-locked data ( Fig. 5 ). Again, strong negative beta
weights form a signicant cluster from 290 ms onwards in R1-locked
data and between 125 ms and 230 ms in R2-locked data. We tentatively
suggest that this negative correlation between the R1-locked Pe and the
R2-locked Pe is reduced in the R2-locked time window beyond 250 ms
due to the strong inuence of a pronounced overlaying Contingent Neg-
ative Variation in anticipation of the error report (see Fig. 7 B). A robust
MVPA-based bin separation of trials with large vs. medium vs. small
immediate Pe as in Experiment 1 was not possible for Experiment 2 be-
cause classier training did not yield a classication accuracy beyond
the 5% signicance threshold as established by a permutation test, likely
due to the smaller size of the training data set.
Taken together, we could replicate the deferred Pe in Experiment 2.
As this analysis was restricted to those error trials that were correctly
reported as T1 errors, we can rule out that this deferred Pe originates in
occasional T1-error trials that are misattributed as T2-errors.
4. Discussion
In two ERP experiments, we investigated neural correlates of error
monitoring in dual-tasking. Previous studies suggested that increased
interference due to temporally overlapping tasks is countered by an ac-
tive deferment of task components ( Hesselmann et al., 2011 ; Meyer and
Kieras, 1997 ; Sigman and Dehaene, 2008 ). Based on this idea, we hy-
pothesized that because higher-level processing of T1 errors can in-
terfere with the execution of T2 (M. Steinhauser et al., 2017 ), also
performance monitoring, more precisely the more resource-consuming,
higher-level aspects of T1 error processing, are adaptively rescheduled
to after the completion of the whole dual-task. To this end, we ana-
lyzed the Pe, an ERP component that is commonly reported to represent
such higher-level, central processing of the error and involves aspects
such as evidence accumulation ( Murphy et al., 2015 ; M. Steinhauser and
Yeung, 2010 , 2012 ), decision condence ( Boldt and Yeung, 2015 ), mo-
tivation ( Drizinsky et al., 2016 ; Kim et al., 2017 ; Moser et al., 2011 ;
Schroder et al., 2014 ), and aect ( Falkenstein et al., 2000 ; van Veen
and Carter, 2002 ). In both experiments, we found a result pattern that
conrms our prediction. When the execution of two tasks overlaps, er-
rors in the rst task lead to a reduced immediate Pe whereas a deferred
Pe appears after the response to the second task.
Our results are in line with the idea that higher-level error pro-
cessing as reected by the Pe interferes with the execution of a sub-
sequent task ( Buzzell et al., 2017 ; Weißbecker-Klaus et al., 2016 ). In
our study, only the Pe but not the Ne/ERN after T1 errors was reduced
when task execution was overlapping. This conforms with recent nd-
ings by Weißbecker-Klaus et al. (2016) , who found a similar reduction
of the Pe when a anker task was executed concurrently with a seman-
tic task. In both studies, this impairment to immediate error detection
likely originates from interference by T2, which requires cognitive re-
sources for T2-response selection at the same time as T1-error processing
would occur. The decision process for selecting the correct response to
T2 was previously suggested to rely on similar resources as error process-
ing ( Hochman and Meiran, 2005 ). On the neural level, this can well be
explained by recent electrophysiological studies that view also higher-
level error processing as a decision process ( Wessel, 2012 , 2011 ) and
studies that highlight the physiological similarities of the P300 and the
Pe ( Leuthold and Sommer, 1999 ; Overbeek et al., 2005 ). Given that in-
terference between central decision processes is seen as the main origin
of dual-task costs ( Pashler 1994 ; Tombu and Jolic œ ur, 2003 ), this ex-
8

R. Steinhauser and M. Steinhauser NeuroImage 232 (2021) 117888
plains why particularly the Pe as a neural correlate of higher-level error
detection ( Overbeek et al., 2005 ) is impaired in dual-tasking.
In addition to a reduced immediate Pe after the erroneous T1 re-
sponse, we found a deferred Pe after completion of the dual-task even
though the second response itself was correct. MVPA-based single-trial
estimates of the Pe as well as a regression-based approach on raw data
suggest that the deferred Pe occurred predominantly on trials with lit-
tle immediate error processing. This temporal detachment of higher-
level error processing from the early error signals that are represented
by the Ne/ERN –on average by more than 580 ms –goes beyond a
mere serialization as it was previously found for the P300 in response
selection ( Hesselmann et al., 2011 ; Marti et al., 2015 ; Sigman and De-
haene, 2008 ). In the present study, at least some aspects of higher-level
error detection appear to be suspended until a whole additional task
has been executed. Analysis of IRIs furthermore showed that the am-
plitude of the immediate Pe is linked to post-error slowing of the re-
sponse to T2. Accordingly, the deferment of error processing was as-
sociated with faster responses to T2, suggesting reduced dual-tasking
interference on such trials, in line with Buzzell et al.’s (2017) nd-
ing that a smaller Pe is also linked to improved sensory processing in
the subsequent trial. This is strong support for the adaptive reschedul-
ing account of error detection in dual-tasking, which suggests such a
mechanism to reduce interference between the subtask representations
and hence to maintain the ability to detect and evaluate errors. No-
tably, time course and scalp topography of the deferred Pe indicate
that this phenomenon aects what has previously been described as
the posterior “late Pe ”, not the earlier frontocentral aspects of the Pe
( Endrass et al., 2007 ; Ullsperger et al., 2014 , see also Steinhauser and
Yeung, 2010 ).
Previous research on error detection as an evidence accumulation
process ( Murphy et al., 2015 ; M. Steinhauser and Yeung, 2010 , 2012 )
may help to approach the neural basis of this deferment mechanism. The
evidence accumulation account considers error detection as a decision
process (M. Steinhauser and Yeung, 2010 ) and is thus well compatible
with the serialization of central, resource-consuming response selection
processes in dual-tasking ( Pashler, 1994 ; Tombu and Jolic œ ur, 2003 ).
The idea that the Pe represents the ongoing accumulation of evidence
that an error has occurred by mirroring the internal weights of evi-
dence can consequently be linked to the processes that form the ba-
sis of the stimulus-locked P300 (see also Ridderinkhof et al., 2009 ).
Marti et al. (2015) recently showed that the brain networks linked to
the emergence of the P300 bring about a competition between the neu-
ral representations of the subtasks in dual-tasking, so that only one of
those representations can be active at a time and a deferment of the T2-
related P300 is induced (see also Hesselmann et al., 2011 ; Sigman and
Dehaene, 2008 ). Shared neural generators of the Pe and the P300 could
therefore also lead to a deferment of the evidence accumulation process
for the T1 error until completion of T2, thus constituting the deferred
Pe that was observed in the present two experiments.
Analyzing brain activity in a PRP paradigm bears the risk that ac-
tivity related to T1 is accidentally attributed to T2 ( Sigman and De-
haene, 2008 ). For instance, what appears to be a deferred Pe after the
T2-response could simply be a carry-over of the immediate Pe after the
T1-response. However, there are several reasons why this cannot ac-
count for the present results. First, the deferred Pe shows a distinct time
course, emerging about 200 ms after the T2 response, which is compara-
ble to the latency of the immediate Pe as well as of the Pe in other studies
( Overbeek et al., 2005 ). Second, even with a short SOA of 300 ms, the
average interval between the T1-response and T2-response on error tri-
als was 587 ms. That is, the onset of the immediate Pe is clearly prior to
the T2-response, and any sustained dierences between T1 errors and
correct trials emerging prior to the T2-response are controlled by the
pre-response baseline. Finally, and most importantly, our results indi-
cate an inverse correlation between immediate Pe and deferred Pe. If
the deferred Pe reected a carry-over of activity from the immediate
Pe, both eects should be positively correlated.
In Experiment 2, we could also rule out that the deferred Pe results
from internal misattribution of occasional T1 errors as T2 errors, which
would imply a credit assignment problem ( Fu and Anderson, 2008 ;
Sutton and Barto, 1998 ; Walsh and Anderson, 2011 ) as the origin of our
ndings. Trial-wise error-reports in Experiments 2 revealed that credit-
assignment errors occurred extremely rarely in our tasks. Moreover, we
obtained a deferred Pe even on the subset of correctly assigned T1 er-
rors. This suggests that even when higher-level processing of the T1 er-
ror is postponed, these errors are still detected and correctly assigned to
their corresponding task. One could speculate that the proposed schedul-
ing mechanism even contributes to solving the credit assignment prob-
lem. On the one hand, rescheduling counteracts interference between
T1 error processing and T2 response selection which otherwise would
increase the risk of confusing error signals from T1 and T2 responses. On
the other hand, when higher-level error evaluation occurs after both re-
sponses were executed, the involved decision process could utilize cues
from both tasks to correctly assign error signals (like post-response con-
ict) to their corresponding tasks.
Though far less robust than the deferred Pe, we also obtained fronto-
central activity related to T1 errors following the T2 response. This could
reect a deferred Ne/ERN, although this activity had a slightly dierent
time course and spatial distribution as the immediate Ne/ERN after T1-
responses. However, this deferred Ne/ERN diered in a crucial aspect
from the deferred Pe as it was not associated with a reduced immediate
Ne/ERN. Without this trade-o between immediate and deferred error
processing, this eect cannot reect a scheduling mechanism. It is also
implausible that the deferred Ne/ERN reects post-response conict in-
duced by a corrective response tendency ( Yeung et al., 2004 ) or a reward
prediction error ( Holroyd and Coles, 2002 ) as both these mechanisms
are temporally linked to the erroneous response. Instead, the deferred
Ne/ERN could represent a negative aective signal to the T2-response.
Several studies could show that the Ne/ERN involves early aspects of
aective processing of an error ( Aarts et al., 2013 ; Maier et al., 2016 ).
The deferred Ne/ERN hence could reect that committing an error in
one task leads to a devaluation also of the second task, indicating that
the whole dual-task trial is processed as erroneous.
Taken together, our results suggest that reliable error detection in
dual-tasking is maintained by a mechanism that adaptively resched-
ules higher-level aspects of error processing to after the completion of
the whole dual-task. Our ndings indicate that the exible reorganiza-
tion of componential processes under dual-tasking is not only a viable
strategy to prevent interference between decision processes involved in
task execution ( Hesselmann et al., 2011 ; Marti et al., 2015 ; Meyer and
Kieras, 1997 ; Sigman and Dehaene, 2008 ), but also serves to reduce in-
terference between task execution and performance monitoring.
Data and code availability statement
Also included in the “Material and Methods ”section of the
manuscript: “All data and analysis scripts are publicly available
in an online repository hosted by the Open Science Framework
( https://osf.io/5ub8z/ ). ”
Author statement
Steinhauser, Robert: Conceptualization, Methodology, Software,
Formal Analysis, Investigation, Writing – Original Draft, Visualization
Steinhauser, Marco : Conceptualization, Writing –Review & Edit-
ing, Supervision, Project administration, Funding acquisition
Acknowledgments
This work was supported by a grant within the Priority Program, SPP
1772 from the German Research Foundation (Deutsche Forschungsge-
meinschaft, DFG, Grant No. STE 1708/4-1 ). The open access publication
9

R. Steinhauser and M. Steinhauser NeuroImage 232 (2021) 117888
of this article was supported by the Open Access Fund of the Catholic
University of Eichstätt-Ingolstadt.
References
Aarts, K. , Houwer, J.De , Pourtois, G , 2013. Erroneous and correct actions have a dierent
aective valence: evidence from ERPs. Emotion 13, 960–973 .
Beatty, P.J. , Buzzell, G.A. , Roberts, D.M. , McDonald, C.G. , 2018. Speeded response errors
and the error-related negativity modulate early sensory processing. Neuroimage 183,
112–120 .
Bell, A.J. , Sejnowski, T.J. , 1995. An information-maximization approach to blind separa-
tion and blind deconvolution. Neural Comput. 7, 1129–1159 .
Boldt, A. , Yeung, N. , 2015. Shared neural markers of decision condence and error detec-
tion. J. Neurosci. 35, 3478–3484 .
Buzzell, G.A. ,
Beatty, P.J. , Paquette, N.A. , Roberts, D.M. , McDonald, C.G. , 2017. Error-in-
duced blindness: error detection leads to impaired sensory processing and lower ac-
curacy at short response–stimulus intervals. J. Neurosci. 37, 2895–2903 .
Coles, M.G.H. , Scheers, M.K. , Holroyd, C.B. , 2001. Why is there an ERN /Ne on cor-
rect trials ? Response representations, stimulus-related components, and the theory of
error-processing. Biol. Psychol. 56, 173–189 .
Cousineau, D. , 2005. Condence intervals in within-subject designs: a simpler solution to
Loftus and Masson’s method. Tutor. Quant. Methods Psychol. 1, 42–45 .
Dehaene, S. , Kerszberg, M. , Changeux, J.-.P. , 1998. A neuronal model of a global
workspace in eortful cognitive tasks. Proc. Natl. Acad. Sci. 95, 14529–14534 .
Del Cul, A. , Baillet, S. , Dehaene, S. , 2007. Brain dynamics underlying the nonlinear thresh-
old for access to consciousness. PLoS Biol. 5, 2408–2423 .
Dell’Acqua, R. , Jolicoeur, P. , Vespignani, F. , Toanin, P. , 2005. Central processing overlap
modulates P3 latency. Exp. Brain Res. 165, 54–68 .
Delorme, A. , Makeig, S. , 2004. EEGLAB: an open source toolbox for analysis of single-trial
EEG dynamics including independent component analysis. J. Neurosci. Methods 134,
9–21
.
Drizinsky, J. , Zülch, J. , Gibbons, H. , Stahl, J. , 2016. How personal standards perfection-
ism and evaluative concerns perfectionism aect the error positivity and post-error
behavior with varying stimulus visibility. Cogn. Aect. Behav. Neurosci. 876–887 .
Endrass, T. , Reuter, B. , Kathmann, N. , 2007. ERP correlates of conscious error recognition:
aware and unaware errors in an antisaccade task. Eur. J. Neurosci. 26, 1714–1720 .
Falkenstein, M. , Hohnsbein, J. , Hoormann, J. , Blanke, L. , 1991. Eects of crossmodal di-
vided attention on late ERP components. II. Error processing in choice reaction tasks.
Electroencephalogr.
Clin. Neurophysiol. 78, 447–455 .
Falkenstein, M. , Hoormann, J. , Christ, S. , Hohnsbein, J. , 2000. ERP components on reac-
tion errors and their functional signicance: a tutorial. Biol. Psychol. 51, 87–107 .
Fu, W.T. , Anderson, J.R. , 2008. Solving the credit assignment problem: explicit and im-
plicit learning of action sequences with probabilistic outcomes. Psychol. Res. 72,
321–330 .
Gehring, W.J. , Goss, B. , Coles, M.G.H , 1993. A neural system for error detection and com-
pensation. Psychol. Sci. 4, 385–390 .
Groppe, D.M. , Urbach, T.P. , Kutas, M. , 2011. Mass univariate analysis of
event-related
brain potentials/elds II: simulation studies. Psychophysiology 48, 1726–1737 .
Gross, J. , Schmitz, F. , Schnitzler, I. , Kessler, K. , Shapiro, K. , Hommel, B. , Schnitzler, A. ,
2004. Modulation of long-range neural synchrony reects temporal limitations of vi-
sual attention in humans. Proc. Natl. Acad. Sci. 101, 13050–13055 Retrieved from
papers3://publication/uuid/D192C02A-B83A-4F02-AEA5-BC0E51C10D82 .
Hesselmann, G. , Flandin, G. , Dehaene, S. , 2011. NeuroImage probing the cortical network
underlying the psychological refractory period : a combined EEG –fMRI study. Neu-
roimage 56, 1608–1621 .
Hester, R. , Foxe, J.J. , Molholm, S. , Shpaner, M. , Garavan, H.
, 2005. Neural mechanisms in-
volved in error processing: a comparison of errors made with and without awareness.
Neuroimage 27, 602–608 .
Hochman, E.Y. , Meiran, N. , 2005. Central interference in error processing. Mem. Cognit.
33, 635–643 .
Holroyd, C.B. , Coles, M.G.H , 2002. The neural basis of human error processing: rein-
forcement learning, dopamine, and the error-related negativity. Psychol. Rev. 109,
679–709 .
Hommel, B. , 1998. Automatic stimulus-response translation in dual-task performance. J.
Exp. Psychol. Hum. Percept. Perform. 24, 1368–1384 .
Jentzsch, I. , Dudschig, C. , 2009. Why do we slow down after an error? Mechanisms un-
derlying the eects of posterror slowing. Q. J. Exp. Psychol. 62, 209–218 .
Jersild, A.T. , 1927. Mental set and shift. Arch. Psychol. 14, 81 .
Kim, M.H. , Marulis, L.M. , Grammer, J.K. , Morrison, F.J. , Gehring, W.J. , 2017. Motiva-
tional processes from expectancy –value theory are associated with variability in the
error positivity in young children. J. Exp. Child. Psychol. 155, 32–47 .
Klawohn, J. , Endrass, T. , Preuss, J. , Riesel, A. , Kathmann, N. , 2016. Modulation of hyper-
active error signals in obsessive–compulsive disorder by dual-task demands. J. Ab-
norm. Psychol. 125,
292–298 .
Leuthold, H. , Sommer, W. , 1999. ERP correlates of error processing in spatial S-R com-
patibility tasks. Clin. Neurophysiol. 110, 342–357 .
Logan, G.D. , Gordon, R.D. , 2001. Executive Control of Attention in Dual-Task Situations.
Psychol. Rev. .
Maier, M.E. , Scarpazza, C. , Starita, F. , Filogamo, R. , Ladavas, E. , 2016. Error monitoring
is related to processing internal aective states. Cogn. Aect. Behav. Neurosci. 16,
1050–1062 .
Marti, S. , King, J.-.R. , Dehaene, S. , 2015. Time-resolved decoding of two processing chains
during dual-task interference. Neuron .
Meyer, D.E. , Kieras, D.E. , 1997. A computational theory of executive cognitive processes
and multiple-task performance: part 1. Basic mechanisms. Psychol. Rev. 104, 3 .
Morey, R.D. , 2008. Condence intervals from normalized data: a correction to Cousineau
(2005). Tutor. Quant. Methods Psychol. 4, 61–64 .
Moser, J.S. , Schroder, H.S. , Heeter, C. , Moran, T.P. , Lee, Y. , 2011. Mind your errors : ev-
idence for a neural mechanism linking growth mind-set to adaptive posterror adjust-
ments. Psychol. Sci. 22, 1484–1489 .
Murphy, P.R. , Robertson, I.H. , Harty, S. , O’Connell, R.G , 2015. Neural evidence accumu-
lation persists after choice to inform metacognitive judgments. Elife 4, 1–23 .
Overbeek, T.J.M. , Nieuwenhuis, S. , Ridderinkhof, K.R , 2005. Dissociable components of
error processing: on the functional signicance of the Pe vis-à-vis the ERN/Ne. J.
Psychophysiol. 19, 319–329 .
Parra, L.C. , Alvino, C. , Tang, A. , Pearlmutter, B. , Yeung, N. , Osman, A. , Sajda, P. , 2002.
Linear Spatial Integration for Single-Trial Detection in Encephalography. Neuroimage
17, 223–230 .
Parra, L.C. ,
Spence, C.D. , Gerson, A.D. , Sajda, P. , 2005. Recipes for the linear analysis of
EEG. Neuroimage 28, 326–341 .
Pashler, H. , 1994. Dual-task interference in simple tasks: data and theory. Psychol. Bull.
116, 220–244 .
Pashler, H. , Johnston, J.C. , 1989. Chronometric evidence for central postponement in
temporally overlapping tasks. Q. J. Exp. Psychol. Sec. A 41, 19–45 .
Rabbitt, P. , 2002. Consciousness is slower than you think Consciousness is slower than
you think. Q. J. Exp. Psychol.: Sec. A 55, 1081–1092 .
Ridderinkhof, K.R. , Ramautar, J.R. , Wijnen, J.G. , 2009. To PE or
not to PE: a P3-like
ERP component reecting the processing of response errors. Psychophysiology 46,
531–538 .
Schroder, H.S. , Moran, T.P. , Donnellan, M.B. , Moser, J.S. , 2014. Mindset induction eects
on cognitive control : a neurobehavioral investigation. Biol. Psychol. 103, 27–37 .
Schroder, H.S. , Moser, J.S. , 2014. Improving the study of error monitoring with consider-
ation of behavioral performance measures. Front. Hum. Neurosci. 8, 1–4 .
Sergent, C. , Baillet, S. , Dehaene, S. , 2005. Timing of the brain events underlying access to
consciousness during the attentional blink. Nat. Neurosci. 8, 1391–1400 .
Sergent, C.
, Dehaene, S. , 2004. Neural processes underlying conscious perception: exper-
imental ndings and a global neuronal workspace framework. J. Physiol. Paris 98,
374–384 .
Shalgi, S. , Barkan, I. , Deouell, L.Y. , 2009. On the positive side of error processing: er-
ror-awareness positivity revisited. Eur. J. Neurosci. 29, 1522–1532 .
Sigman, M. , Dehaene, S. , 2008. Brain mechanisms of serial and parallel processing during
dual-task performance. J. Neurosci. : O. J. Soc. Neurosci. 28, 7585–7598 .
Steinhauser, M. , Ernst, B. , Ibald, K.W. , 2017. Isolating component processes of posterror
slowing with the psychological refractory period paradigm.
J. Exp. Psychol.: Learn.
Mem. Cogn. 43, 653–659 .
Steinhauser, M. , Yeung, N. , 2010. Decision processes in human performance monitoring,
30. J. Neurosci. §.
Steinhauser, M. , Yeung, N. , 2012. Error awareness as evidence accumulation: eects of
speed-accuracy trade-o on error signaling. Front. Hum. Neurosci. 6, 1–12 .
Steinhauser, R. , Steinhauser, M. , 2018. Preparatory brain activity in dual-tasking. Neu-
ropsychologia 114, 32–40 .
Sutton, R.S. , Barto, a G , 1998. Reinforcement learning: an introduction. In: IEEE Trans-
actions on Neural Networks /a Publication of the IEEE Neural Networks Council, 9,
p. 1054 .
Telford, C.W. , 1931. The refractory phase of voluntary and associative responses. J. Exp.
Psychol. 14, 1–36 .
Tombu, M. , Jolic œ ur, P. , 2003. A central capacity sharing model of dual-task performance.
J. Exp. Psychol. Hum. Percept. Perform. 29, 3–18 .
Ullsperger, M. , Fischer, A.G. , Nigbur, R. , Endrass, T. , 2014. Neural mechanisms and tem-
poral dynamics of performance monitoring. Trends Cogn. Sci. 18, 259–267 .
Van der Borght, L. , Schevernels, H. , Burle, B. , Notebaert, W. , 2016. Errors disrupt subse-
quent early attentional processes. PLoS ONE 11, e0151843 .
van Veen, V. , Carter,
C.S. , 2002. The timing of action-monitoring processes in the anterior
cingulate cortex. J. Cogn. Neurosci. 14, 593–602 .
Walsh, M.M. , Anderson, J.R. , 2011. Modulation of the feedback-related negativity by in-
struction and experience. PNAS 108, 19048–19053 .
Weißbecker-Klaus, X. , Ullsperger, P. , Freude, G. , Schapkin, S.A. , 2016. Impaired Error
Processing and Semantic Processing During Multitasking. J. Psychophysiol. 1–12 .
Welford, A.T. , 1952. The “psychological refractory period ”and the timing of high-speed
performance - a review and a theory. Br. J. Psychol. Gen. Sec. 43, 2–19 .
Wessel, J.R. , 2012. Error awareness and the
error-related negativity: evaluating the rst
decade of evidence. Front. Hum. Neurosci. 6, 88 .
Wessel, J.R. , Danielmeier, C. , Ullsperger, M. , 2011. Error awareness revisited: accumula-
tion of multimodal evidence from central and autonomic nervous systems. J. Cogn.
Neurosci. 23, 3021–3036 .
Winer, B.J. , Brown, D.R. , Michels, K.M. , 1991. Statistical Principles in Experimental De-
sign, 3rd ed. McGraw-Hill .
Yeung, N. , Botvinick, M.M. , Cohen, J.D. , 2004. The neural basis of error detection: conict
monitoring and the error related nagativity. Psychol. Rev. 11, 931–959 .
10