Electrophysiological correlates of residual switch costs

cortex 46 (2010) 1138–1148

available at www.sciencedirect.com

journal homepage: www.elsevier.com/locate/cortex

Research report

Electrophysiological correlates of residual switch costs Patrick D. Gajewski*, Thomas Kleinsorge and Michael Falkenstein Leibniz Research Centre for Working Environment and Human Factors at the Technical University of Dortmund, Germany

article info

abstract

Article history:

Switching among cognitive tasks results in switch costs which are only partly reduced even

Received 25 June 2008

after sufficient task preparation. These residual switch costs are frequently explained in terms

Reviewed 1 October 2008

of interference of simultaneously active task representations that delays selection of a correct

Revised 3 March 2009

response. Recent studies showed that the benefit of a task- and response-set repetition can

Accepted 28 July 2009

also explain residual costs. We aimed to extend the findings by clarifying the mechanisms

Action editor Ray Johnson

underlying task- and response-mode repetition benefit as well as costs arising by switch of one

Published online 5 August 2009

or both dimensions. To this end we used a combination of task-switching and go/no-go paradigm during an electrophysiological recording. Particularly, we focused on the fronto-

Keywords:

central N2, which has been usually related to conflict, but also to response selection. The

Task switching

behavioral results replicate previous findings of lack of residual switch costs due to slower

Residual switch costs

responses in task repetitions (TRs) following no-go relative to go trials. This indicates elimi-

No-go

nation of TR benefit when in a previous trial no response was selected and prepared. In other

ERP

words, task sets clearly benefits from repetition of response mode whereas interference seems

Response selection

to occur whenever the task-set, the response mode or both were switched. Trial incongruity

N2

increased reaction times. The event-related potentials (ERPs) revealed a frontocentral N2 in all

Conflict

conditions which followed the same pattern as the reaction times (RTs), showing smaller

Interference

amplitude and peaking earlier when both the task and response mode were repeated relative

Congruency

to the three switching conditions. Similar to the behavioral data, the N2 increased as a function of incongruity. Finally, both the N2 amplitude and latency correspond closely to the residual switch costs. This finding suggests that task-set or response mode switching intensify and delay response selection, relative to the repetition of both dimensions. ª 2009 Elsevier Srl. All rights reserved.

1.

Introduction

In everyday life, people are confronted with number of changing cognitive tasks, which have to be managed in a close temporal succession. The ability to switch among different tasks necessitates balance between activation of the relevant and deactivation of the irrelevant task and has been assumed

to be a crucial executive function in humans (e.g., Goschke, 2002). A common observation is the slowing of a response to the task that has to be switched relative to the repetition of the same task, which has been termed switch costs. It was proposed that switch costs reflect a time consuming process of re-adjustment of internal task representations (Rogers

* Corresponding author. Leibniz Research Centre for Working Environment and Human Factors, Ardeystr. 67, D-44139 Dortmund, Germany. E-mail address: [email protected] (P.D. Gajewski). URL: http://www.ifado.de 0010-9452/$ – see front matter ª 2009 Elsevier Srl. All rights reserved. doi:10.1016/j.cortex.2009.07.014

cortex 46 (2010) 1138–1148

and Monsell, 1995). A core finding was that switch costs are significantly reduced when people have the opportunity to prepare for a task switch (TS) in advance. However, the switch costs are often not eliminated by advance preparation. The remaining portion of reaction time costs has been denoted residual switch costs (Allport, et al., 1994; De Jong, 2000; Meiran, 1996, 2000; Rogers and Monsell, 1995). Residual switch costs were systematically analyzed by Rogers and Monsell (1995). They varied the response-to-stimulus interval (RSI) between blocks of predictable sequence of tasks and observed reduction of switch costs when the interval increased from 150 to 600 msec. A further increase did not diminish the costs. The authors explained the residual costs in terms of intrinsic limitations to complete the switch process during preparation. Allport et al. (1994) originally attributed the residual switch costs to ‘‘task-set inertia’’ that means persisting activation of competing task-set from the previous trial that decays only slowly over time. They argued that during switching between tasks that require the same stimulusset, the stimulus-response mapping of the recently executed but no-longer relevant task persists until the next trial. When a TS is required, the negative transfer of this no-longer relevant mapping interferes with the currently relevant one. Due to this proactive interference, some additional time is needed to select the adequate response and this is reflected in the residual switch cost (Allport and Wylie, 2000). Mayr and Keele (2000) hypothesized that proactive interference consists of persisting inhibition of task sets: inhibition from a task-set that was abandoned in a previous trial hampers the reactivation of the same task, when it is relevant later. The authors used three tasks A, B, and C and compared switch costs in the last trial of ABA with CBA – sequences. They found larger switch costs in the ABA (n  2 task repetition costs) relative to CBA – sequence. This finding, termed ‘backward inhibition’ implies that the more recently a task has been switched away from, the harder it is to switch back to. The authors suggested that residual switch costs may be also a consequence of persisting inhibition of task sets. Schuch and Koch (2003) investigated the functional mechanism underlying backward inhibition and its contribution to the residual switch costs. In particular they tested the hypothesis that backward inhibition occurred as a function of response selection and not task preparation. This prediction was supported using a combination of taskswitching and go/no-go paradigm. The participants were asked to switch among two tasks presented in a random sequence and to withhold their response in 25% of trials. Indeed, after no-go trials no backward inhibition and no switch costs have been observed. Schuch and Koch (2003) interpreted the residual switch costs as a consequence of persisting inhibition of currently irrelevant task-set arising during response selection. Moreover, the authors show that response preparation (i.e., selection) but not response execution is the crucial process for backward inhibition and switch costs to occur (but see Philipp et al., 2007 for findings regarding contribution of response execution to switch costs).

1139

However, the elimination of residual switch costs after nogo trials yielded in Schuch and Koch’s study was not due to a decrease of RTs in switch trials but to increase of RTs in TR trials. This observation was tentatively interpreted in terms of (negative) priming of a no-go decision from trial n  1 to n. In a further study Koch, Gade and Philipp (2004) investigated inhibitory processes during switching among response modes and found substantial costs for switch among simple (unconditional double-press) versus choice responses. Although this study did not explicitly investigate the no-go – go transition effects across trials, the general increase of reaction times (RTs) after a no-go trial may be due to the same mechanism, namely interference between two response modes. In a later study Koch and Philipp (2005) investigated the response-selection account of residual switch costs in more detail and found that the increase of task repeat RT after no-go trials is due to lack of a TR benefit. This benefit only occurs when in a previous trial a particular response has been selected. In other words, response selection in n  1 leads to an activation bias favoring TR in the following trial. This result emphasized the role of persisting activation of task-set after a response has been selected but at the same time it did not diminish the role of the inhibition of irrelevant task. In sum, Schuch and Koch (2003) and Koch and Philipp (2005) concluded that the absolute activation level of task-set (defined as a category–response – C–R association; e.g., digits smaller than 5 – left key) may explain slowing in switching trials: the closer the activation level is to the activation threshold, the faster the selection of a response. This activation level is decreased by persisting inhibition from previously abandoned or elevated by previously selected task, suggesting that the timing to activate a correct response in the current trial is determined by the persisting activation level of C–R associations. The effect of switching between task sets can be additionally modulated by the overlapping stimulus-response assignments (Meiran, 1996; Rogers and Monsell, 1995). This so called congruency effect occurred when the stimuli are assigned to different responses than the same one. For example if the digits are mapped according to the task rules: smaller than 5 or even – left key and larger than 5 and odd – right key, the digits 2, 4, 7, 9 are mapped onto the same keys in both tasks (2 and 4 to the left and 7 and 9 to the right key). In contrast the remaining digits are mapped onto opposite responses (e.g., the digit 1 requires the left button press in the smaller/larger than 5 – task and the right button press in the odd/even task). These so called incongruent trials presumably reflect additional activation of the concurrent task resulting in performance decrements relative to trials that called for the same response (Allport and Wylie, 2000). Kiesel et al. (2007) supported this hypothesis showing that the congruency effect is a consequence of activation of concurrent stimulus-response associations. According to previous findings of Koch and coworkers (e.g., Koch and Philipp, 2005) that showed that residual switch costs arise as a function of response related processes and are largely independent from task preparation, we focused mainly on mechanisms during implementation of a switching task that is on processes that occurred after presentation of a target. In particular, the aim of the present study was to test the response-selection account of residual switch costs during

1140

cortex 46 (2010) 1138–1148

task- and response-mode switching, by elucidating the relationship between task activation and inhibition in more detail. In order to do this, we employed the above described hybrid task switching – go/no-go paradigm hybrid task-switching – and go/no-go paradigm during an electrophysiological recording. Electrophysiological studies yielded a neural correlate of interference – a negative event-related potential (ERP) deflection with frontocentral maximum at about 300 msec after onset of a target in choice reaction time tasks (cf. N2c Kopp et al., 1996a, 1996b; van Veen and Carter, 2002a). This negative wave has been originally described by Klinke et al. (1968) and systematic analyzed by Simson et al. (1976, 1977) who proposed that N2 reflects stimulus classification (e.g., target selection). Further studies showed consistent correlations between the latency of the N2 and RTs during single trial analyses (Ritter et al., 1979, 1982; Renault and Lesevre, 1979; Renault et al., 1980), leading to the conclusion that the N2 reflects a directly measure of the absolute timing of certain decision processes (Ritter et al., 1979). The peak latency of the N2 varies as a function of task difficulty and precedes and correlates with the timing of discriminative behavioral responses (see Ritter et al., 1982, 1983 for an overview). In recent investigations the frontocentral N2 was interpreted in terms of mismatch or monitoring and control of motor responses (see Folstein and Van Petten, 2008 for review) or conflict detection during response selection (Yeung and Cohen, 2006; van Veen and Carter, 2002a, 2002b) because it is especially prominent in situations when two incompatible response tendencies are simultaneously active. In our previous study (Gajewski et al., 2008) we investigated the N2 in a compatibility task. In this task participants prepared a response in advance on the basis of a pre-cue but the pre-cue predicted the following response in 20% in the first, in 50% in the second and in 80% in the third block of all trials incorrectly, producing interference between the expected and the actually required response. Participants were not informed about the block by block decrease of cue validity. A strong compatibility effect on responses in the first block was observed, which diminished in the second and reversed in the last block, leading to increased RTs and error rates in unexpected, although correctly cued trials. Accordingly, in the ERP-data a N2 occurred which was most pronounced in case of unexpected cue-target associations regardless of trial compatibility. Importantly, a reliable positive correlation between N2 latency and RTs was obtained in all conditions, suggesting that N2 occurs both in compatible (non-conflict) and incompatible (conflict) trials and varies as function of RTs. These observations led us to conclude that N2 may reflect the process of response selection which is intensified and prolonged when response selection becomes more difficult, e.g., by unexpected stimulus-response (S-R) relations. Source analysis studies found the neural generator of this frontocentral N2 – component in the anterior cingulate cortex (ACC; Carriero et al., 2007; Chen et al., 2007; Swainson et al., 2003; van Veen and Carter, 2002a, 2002b). It has been previously reported that the corresponding ACC activity in conflict tasks correlates positively with the amount of interference observed behaviourally (Botvinick et al., 1999; MacDonald et al., 2000).

A number of electrophysiological studies obtained the frontocentral, target-locked N2 in context of task switching. For example, Jackson et al. (2001, 2004) reported a study of language switching in bilingual speakers. They observed an increase of the N2 in switch compared to repetition trials and interpreted this finding in terms of a suppression of a habitual response. Swainson et al. (2003) observed a N2 – effect when participants switched between go- (immediately response) and wait- (withholding response until stimulus offset) modes during the performance of a simple categorization task. Furthermore, an enhanced post-target N2 in switch trials was also obtained in a number of studies employing standard taskswitching paradigms (Gehring et al., 2003; Karayanidis et al., 2003; Kieffaber and Hetrick, 2005; Kotchoubey et al., 1997; Nicholson et al., 2005, 2006; Rushworth et al., 2002, 2005; Wylie et al., 2003) but was often not further considered. As the frontocentral N2 has been frequently associated with processing of response conflict or interference, we intend to investigate the relationship between the negativities and interference during switching among task sets and response modes (no-go – go). We extend the concept of ‘interference’ by ‘interference during response selection’ and define response selection as cognitive process of activation of a particular stimulus category (e.g., smaller than 5) that is mapped on a particular response category (e.g., left). More specifically, we hypothesized that N2 may be involved in the selection process which is modulated by persisting activation or inhibition of task sets and response-mode sets. This assumption led to following predictions. 1. If the frontocentral N2 is susceptible to a selection process that is modulated by task interference, we expect to observe an enhanced N2 in TS relative to repetition trials after go trials (i.e., in a condition which elicits residual switch costs). No N2 difference between TS and repetition trials and consequently no residual switch costs are expected after no-go trials. Moreover, the residual switch costs should be related to the corresponding N2 effect. 2. If switching among response modes (go after no-go) elicits similar interference as switching among tasks, we expect an increase of the N2 relative to the repetition of the response mode (go after go). 3. If incongruent trials reflect interference due to the increased co-activation of a concurrent task, the N2 should be more pronounced relative to congruent trials. 4. If task-set interference produces prolongation of the selection process, a relationship between the timing in the N2 effect and the switch costs should be observed.

2.

Materials and methods

2.1.

Participants

Seventeen undergraduate students (6 male), aged 20–35 (mean age 24 years) from the University of Dortmund participated in the study for course credit. All subjects were right handed and had normal or corrected-to-normal vision. They had no prior exposure to the paradigm and were not informed about the aim of the study.

1141

cortex 46 (2010) 1138–1148

2.2.

Stimuli and tasks

go

nogo Response

The experiment was conducted with subjects seated in a dimly illuminated, electrically shielded room in front of a computer monitor at a distance of 80 cm. The stimuli were displayed on a black background in the center of the screen and consisted of white frames (square or diamond) indicating type of task and green or red digits from the range 1–9 excluding 5 as imperative stimuli. The sides of the frames were 4 cm in length. The digits were 1 cm in height and 1 cm in width. The participants were required to respond according to the numerical magnitude (more or less than five) or parity (odd or even) of the digit. The relevant task was indicated by the frame that surrounded the digit. A square indicated the magnitude task, a diamond the parity task. Hence the frame served as cue, which indicated the relevant task and allowed task preparation in advance. Responses consisted of pressing one of two buttons which were mounted in two joystick-like vertical bars. The buttons were operated with the right or the left thumb. The stimulus-response mapping of both tasks was overlapping, that is responses according to ‘smaller than five’ and ‘even’ were assigned to the left key and ‘larger than five’ or ‘odd’ to the right key. This assignment was counterbalanced across participants. 50% of trials were congruent that is the stimuli required the same response, 50% trials were incongruent. Twenty-five percent of the trials were no-go trials, with the restriction that, first, two no-go trials could not appear in a close succession and, second, the frequency of TR and TS was the same in the trials following go and no-go trials. Green digits indicated go, red digits no-go trials.

2.3.

Design and procedure

A schematic example of a sequence of two trials is shown in Fig. 1. A trial started with a presentation of the cue frame. After 1000 msec (cue-target interval – CTI)1 a green digit appeared inside the cue frame for go trials and stayed until a response was given. In case of an incorrect response, an error message ‘Wrong key’ (in German) was displayed for 500 msec. The response-target interval (RTI), which included the CTI msec was 1600 msec. Thus, the responsecue interval (RCI) was 600 msec. During RCI the screen was blank. In the no-go trials both the red digit and the frame stayed on the screen for 1000 msec when the participant correctly refrained from responding. After stimulus offset the screen was blank for 1100 msec before the next cue was presented, resulting in a global cue – cue interval of 3100 msec. When a response was given during a no-go trial, a feedback ‘Don’t press any key!’ (in German) appeared on the screen for 500 msec. The experiment consisted of 6 blocks with 128 trials each. A randomly selected block served as a practice block. 1

We employed also a short CTI ¼ 100 msec in order to control the preparation efficiency and effects of no-go trials following a short interval. However, as we are mainly interested in residual switch costs we report data from a long preparation interval only.

No Response

8

3 Cue

Time

Target RCI 600 ms

CTI 1000 ms

1000 ms

RTI 1600 ms

Fig. 1 – A schematic illustration of a go and a no-go trial. A cue frame was presented for 1000 msec. Then, a green (go) or red (no-go) digit (target) appeared within the frame. In a go trial the target was displayed until a response occurred. The RTI amounted to 1600 msec. Thus, the RCI amounted to 600 msec. In a no-go trial, the target stayed on the screen for 1000 msec. After that, the target disappeared and the next cue was presented 1100 msec later.

The sequence of tasks resulted from a factorial combination of type of task in n, type of task in n  1, required response hand in n and required response hand in n  1. All factors were randomized within the experiment and for each participant separately. The participants were given a written instruction that explained the task. The instruction encouraged making use of the preparatory interval and responding quickly and accurately.

2.4.

ERP recordings

Electroencephalogramm (EEG) was recorded continuously from 63 scalp electrodes using electrodes positioned according to the extended 10–20 system (Jasper, 1958) and mounted on an elastic cap (EASYCAP GmBH). The montage included 9 midline electrodes and 27 sites over each hemisphere. ERPs were referenced to the electrode Cz, than re-referenced offline to both mastoids. The ground electrode was placed at the nasion. The horizontal electrooculogramm (EOG) was recorded bipolarly from electrodes at the outer canthi of both eyes; the vertical EOG was recorded from electrodes above and below the right eye. Eye movement artifacts were corrected using the correction algorithm of Gratton et al. (1983). Electrode impedance was kept below 5 kU. Amplifier bandpass was .01–140 Hz. EEG and EOG were sampled continuously with a rate of 500 Hz. Epochs in which the amplitude exceeded  100 mV were rejected. The ERPs were filtered digitally with a 17 Hz low pass.

2.5.

Data analysis

The design was applied only to go trials in the current trial (n) and included three within-subject factors: response mode in n  1 (go vs no-go), task transition (TR vs TS) and congruency (congruent vs incongruent). With this design reaction times and error rates were compared. The practice block, first trial of each test block, trials with incorrect responses, trials immediately following incorrect responses, trials immediately following false alarms in no-go trials, and trials associated with a response faster than 100 msec or slower than 2500 msec were excluded from behavioral and ERP analyses.

1142

cortex 46 (2010) 1138–1148

cue-locked

[µV] -10

target-locked

[µV] -10

Fz

TR after go TS after go

Fz

-5

-5

0

0

5

5

10

10

TR after no-go TS after no-go

CSD: N2 at 328 msec 0

500

1000

1500

[msec]

0

200

400

600

800 [msec]

-10

-10

FCz

FCz -5

-5

N2

N2

0

0

5

5

TR after go

TS after go

10

10 0

500

1000

1500

[msec]

0

200

400

600

800 [msec]

-10

-10

Cz

-5

Cz

-5

0

0

5

5

TR after no-go

TS after no-go

10

10 0

500

1000

1500

[msec]

0

200

400

600

800 [msec]

-10 µV/m²

0

10 µV/m²

Fig. 2 – ERP – waveforms in go trials as a function of task transition (TR vs TS) and response mode in n L 1 (go vs no-go). Left column: Cue-locked ERPs at electrode sites Fz, FCz, Cz. Dashed vertical lines indicate cue (0 msec) and target onset (1000 msec). Right column: Target-locked waveform at electrode sites Fz, FCz and Cz. Dashed line indicates target onset. The Current Density Maps (CSD) illustrates activity distribution at the time point of 328 msec (N2) after target onset in each of the four conditions.

The ERPs were analyzed in go trials following go or no-go trials in the time range of 1000 msec after target onset. Targetlocked N2 was assessed by measuring the most negative deflection in the time range 250–400 msec at Fz, FCz and Cz relative to a 100 msec baseline prior to target onset. Peak amplitudes and latencies were measured fully automatically and inspected visually. In order to avoid boundary hits, only local maxima or minima within the interval were considered. As the peak detectability was sufficiently reliable, no further peak adjustments were necessary. The analysis of the N2 amplitude was conducted for following factors: electrode (Fz, FCz, Cz), response mode in n  1 (go, no-go) and task-set transition (TR, TS). In two post-hoc analyses restricted to the electrode position Fcz, additional factors response hand transition (repetition vs switch) and congruity (congruent vs incongruent trials) were conducted. In order to avoid systematic effects from overlapping components from the precuing interval, we analyzed the N2 relative to 100 msec pre-cue baseline and conducted additionally a P2N2 peak-to-peak measure (Gajewski et al., 2008; Picton et al., 2000). The P2 was assessed as the most positive peak in the time range 150–250 msec after target onset.

The relationship between behavioral and electrophysiological parameters was assessed by correlation analysis using the Pearson correlation coefficient (two-tailed). A regression analysis was conducted to estimate the relationship between electrophysiological and behavioral parameters in more detail. To test specific effects or interactions, additional repeated measures analysis of variance (ANOVA) and planned comparisons were employed.

3.

Results

3.1.

Behavioral data

The mean reaction times and error rates are shown in Fig. 4a. 1% of trials were responded faster than 100 or slower than 2500 msec. ANOVA revealed a main effect of task-set transition, suggesting longer RTs in switch than repetition trials [825 msec, SD ¼ 301 msec vs 778 msec, SD ¼ 280 msec; F(1, 16) ¼ 16.8, p < .001, h2 ¼ .51]. Furthermore, the responses were longer after no-go than go trials (820 msec, SD ¼ 296 msec vs 783 msec, SD ¼ 286 msec) resulting in an

1143

cortex 46 (2010) 1138–1148

[µV] -10

[µV] -10

Fz

Fz

-5

-5

0

0

5

5

TR after go TS after go TR after no-go TS after no-go

10

10 0

200

400

600

800 [msec]

0

200

400

600

800 [msec]

600

800 [msec]

600

800 [msec]

-10

-10

FCz

FCz -5

-5

N2

N2

0

0

5

5

10

10 0

200

400

600

800 [msec]

0

200

400

-10

-10

Cz

-5

Cz

-5

0

0

5

5

10

10 0

200

400

600

800 [msec]

0

200

400

Fig. 3 – Target-locked ERP – waveforms in go trials as a function of trial congruency (congruent vs incongruent), task transition (TR vs TS) and response mode in n L 1 (go vs no-go) at electrode sites Fz, FCz and Cz. Left column: congruent trials, right column: incongruent trials. Dashed line indicates target onset.

main effect of response mode in n  1 [F(1, 16) ¼ 7.3, p ¼ .015, h2 ¼ .31] and in incongruent trials than in congruent trials leading to a simple effect of congruency [858 msec, SD ¼ 202 msec vs 751 msec, SD ¼ 180 msec; F(1, 16) ¼ 50.9, p < .0001, h2 ¼ .76]. More important for the present purpose is the impact of the previous response mode on task-set transition. While there were substantial costs after go trials (90 msec), these were virtually absent (6 msec) after no-go trials, resulting in an interaction of the factors response mode in n  1 and task-set transition [F(1, 16) ¼ 27.7, p < .0001, h2 ¼ .63]. The RTs in switch trials were longer than in repetitions after go (828 msec, SD ¼ 303 msec vs 738 msec, SD ¼ 269 msec; p < .0001) but not after no-go trials (823 msec, SD ¼ 299 msec vs 817 msec, SD ¼ 292, ns.). This interaction was not further modulated by the trial congruency [F(1, 16) ¼ 3.3, p ¼ .09]. Taken together, this pattern of results shows that the absence of switch costs in trials following no-go trials is due to the elimination of a task repetition benefit. Errors were made in 3.8% of trials. Participants committed significantly more errors when the task was switched than repeated [4.2% vs 3.1%, F(1, 16) ¼ 8.2, p ¼ .011, h2 ¼ .34] and in

incongruent than in congruent trials [5.2% vs 1.2%, F(1, 16) ¼ 60.4, p < .0001, h2 ¼ .73]. Finally, error rates were higher when participants switched after go trials (5% vs 2.4%) than after no-go trials (3.4% vs 3.8% for TS and TR, respectively), resulting in a significant interaction response mode in n  1  task-set transition [F(1, 16) ¼ 29.27, p < .0001, h2 ¼ .65].

3.2.

Target-locked ERPs2

The ERPs as a function of task-set transition and response mode in n  1 for pre-cue and pre-target baseline at the electrode sites Fz, FCz, Cz are depicted in Fig. 2. For all conditions the waveforms were characterized by the emergence of a negative peak with a maximum at about 2 We analyzed also cue-locked and other target-locked ERPs. Although we found several task-set and response set related effects (e.g., in the contingent negative variation (CNV) and the P3), a consistent interaction between both factors reflecting the pattern obtained in the behavioral data was found in the reported N2 only. From this reason, we refrain from reporting other effects than the N2.

1144

cortex 46 (2010) 1138–1148

RT [ms] 850

go in n-1

N2 latency [ms]

no-go in n-1

a

b

0,5

330

825 800

1,5 2,0

ER [%] 10

750

320

2,5

8 6

725

4

3,0

315

2

700 TR

d

3,5 TR

TS

y = -19.6x + 75.2

Amplitude of the N2 – effect [µV]

c

1,0

325

775

Residual switch costs [ ms]

N2 - amplitude [µV]

e

TS

y = 1.2x + 69.2

Latency of the N2 – effect [ms]

TR

f

TS

y = 1.2x + 1.4

Latency of the N2 – effect [ms]

Fig. 4 – Top: Reaction times, error rates (a), amplitudes (b) and latencies (c) of the N2 at FCz as a function of task transition (repetition: TR vs switch: TS) and response mode in n L 1 (go vs no-go). Go in n L 1 – solid lines, no-go in n L 1 – dashed lines. Bottom: linear regression for amplitude (d) and latency (e, f) of the N2 effect at FCz (difference between TS and TR trials) and residual switch costs for go trials following go trials (d, e) and following no-go trials (f).

320 msec after target onset. Latency, topography and morphology of this peak suggest that it represents a N2. The peak was shifted to the positive range because the N2 was superimposed on the subsequent P3. ANOVA revealed a main effect of electrode [F(2, 32) ¼ 5.7, p < .01, h2 ¼ .26], suggesting a similar N2 at Fz and FCz (1.6 mV vs 1.9 mV, p ¼ .9) and a less negative N2 at Cz (3.0 mV; p < .01, for the contrast with FCz, and Fz). N2 amplitude was also strongly affected by the response mode in n  1 [F(1, 16) ¼ 19.8, p < .0001, h2 ¼ .55] that was due to a more negative N2 in trials following no-go (1.3 mV) than go trials (3.0 mV). This effect was additionally modulated by electrode [F(2, 32) ¼ 4.6, p < .05, h2 ¼ .22], being stronger at Fz and FCz (2.1 and 1.8 mV) than at Cz (1.1 mV, p’s < .05). There was no main effect of task-set transition [F(1, 16) ¼ 2.4, p ¼ .14]. However, task-set transition  electrode interacted significantly [F(2, 32) ¼ 9.1, p < .005, h2 ¼ .36], indicating an increasing difference between TS and TR for more posterior electrodes (0 mV at Fz, .2 mV at FCz and .5 mV at Cz). Most important however, was a significant interaction between response mode in n  1 and task-set transition [F(1, 16) ¼ 12.7, p < .005, h2 ¼ .44] that was not modulated by the electrode position (F < 1). This result suggests that the effect of task-set transition was influenced by the response mode in n  1 and was relatively broadly distributed over frontocentral areas. As the current source density maps (CSDs, Fig. 2) suggest a clear frontocentral maximum of the N2 at Fcz, all subsequent analyses were conducted at this electrode position only.

A post-hoc ANOVA yielded again a significantly more negative N2 after no-go than go trials [1.0 mV vs 2.8 mV; F(1, 16) ¼ 19.3, p < .0001, h2 ¼ .55] and the above reported interaction response mode in n  1  task-set transition at Fcz [F(1, 16) ¼ 12.8, p ¼ .003, h2 ¼ .44], suggesting more negative N2 in switch relative to repetition trials after go trials [2.3 mV (SD ¼ 3 mV, SE ¼ .7 mV) vs 3.3 mV (SD ¼ 3 mV, SE ¼ .7 mV); p < .0001], and no significant difference after no-go trials [.8 mV (SD ¼ 3 mV, SE ¼ .7 mV) vs 1.3 mV (SD ¼ 3 mV, SE ¼ .7 mV); p ¼ .16, see Fig. 4c]. In order to substantiate this effect and avoid the influence from the negative going slow wave after no-go trials (see Fig. 2 left), we additionally measured the N2 relative to a 100 msec pre-cue baseline and relative to the preceding P2 (peak-topeak measure). Both ANOVAs conducted at FCz confirmed the interaction between task-set transition and response mode transition [F(1, 16) ¼ 8.9, p < .01, h2 ¼ .36 and F(1, 16) ¼ 11.0, p < .005, h2 ¼ .41, for the N2 measured relative to pre-cue baseline and relative to the P2, respectively]. The observed impact of task switching on the N2 amplitude may originate not from switching tasks per se, but could be an effect of switching response hands among successive trials. To meet this objection, we added the factor ‘response hand transition’ (response switch vs response repetition) into the analysis. The ANOVA revealed a marginally significant main effect of response hand transition [F(1, 16) ¼ 4.0, p ¼ .059, h2 ¼ .20], i.e., the N2 was overall slightly larger for response hand switches (1.5 mV) than for response hand repeats (1.9 mV) but all

cortex 46 (2010) 1138–1148

interactions including this factor were far from significance (all F’s < 1), indicating no impact of response hand switch or repetition on the observed N2 effects. Finally, we investigated the impact of trial congruity on the N2 amplitude (Fig. 3). To this end, we conducted an ANOVA including the factors response mode transition, task-set transition and congruency at Fcz and found a main effect of trial congruency [F(1, 16) ¼ 8.4, p < .01, h2 ¼ .35], indicating more negative N2 in incongruent than congruent trials (.3 mV vs 1.0 mV). As the congruency effect did not interact with other factors it suggests an unspecific amplification of the N2 whenever two concurrent S-R assignments are simultaneously active. The mean N2 latency was 323 msec. ANOVA with the factors response mode in n  1 and task-set transition revealed a significant interaction of both factors [F(1, 16) ¼ 4.43, p < .05, h2 ¼ .22)], indicating no difference between repetitions and switches after no-go trials [325 msec (SD ¼ 23 msec, SE ¼ 5.5 msec) vs 323 msec (SD ¼ 27 msec, SE ¼ 6.5 msec); ns.] but a significant N2 delay in TSs relative to TRs after go trials 331 msec [(SD ¼ 29 msec, SE ¼ 6.9 msec) vs 313 msec (SD ¼ 23 msec, SE ¼ 5.5 msec); p < .05, see Fig. 4b]. As we also measured intensity of a button press using force sensors and did not find any differences between conditions, we can exclude the possibility that the reported effects are consequence of differences in motor execution processes.

3.3.

Relationship between N2 and switch costs

In order to assess the relationship between the interference indexed by the N2 and residual switch costs, we related the N2 amplitude and latency differences between switch and repetition trials to the corresponding RT-difference (residual switch costs). A significant negative correlation was found between RT-costs and N2 amplitude effects in go trials following go trials (r ¼ .48, p < .05). Moreover positive correlations were found between RT-costs and the corresponding N2 latency effects for trials following both go (r ¼ .60, p < .01) and no-go (r ¼ .55, p < .02) trials. Note, the negative correlation for the amplitude indicates that the more negative the N2 difference the higher the residual switch costs, while the positive correlation for the latency suggests that the larger the delay of the N2 in switch versus repetition trials the larger the switch costs. Fig. 3 (bottom) illustrates the relationship between N2 amplitude (3d), latency (3e, 3f) and residual switch costs. This significant relationship between N2 amplitude and switch costs in trials following go trials was lost after removal of the two outliers with residual costs of about 200 msec (r ¼ .45, p ¼ .07), the correlation between N2 latency effects and switch costs however, remained significant (r ¼ .57, p < .05) in this reduced data set. We conducted also a correlational analysis for the P2-N2 peak-to-peak amplitude measure at Fz and reaction times for each condition separately. This analysis yielded a consistent positive relationship for all of the four conditions: TR after no-go (r ¼ .66, p < .005), TS after no-go; (r ¼ .57, p < .02), TR after go (r ¼ .52, p < .05) and TS after go; (r ¼ .63, p < .01). This suggests that the more negative the P2-N2 index, the faster the response.

1145

Regression analyses conducted over N2 latency and residual costs over all participants yielded the same slope for trials following go and no-go trials ( y ¼ 1.2x, see Fig. 4e, f). This may imply a more general relationship between the N2 latency difference (between switches and repeats) and switch costs,3 which seems not to be influenced by response execution in the previous trial. The same topography of the frontocentral N2 (Fig. 2 right) in all four conditions indicates also a single component on switch as well as repeat trials that seems to be associated with the subsequent response. In summary, when the previous trial was a go trial the frontocentral N2 was enhanced and delayed for TSs compared to TRs. This effect was specific for task transition and was not modulated by response hand transition or trial congruency. Incongruent trials elevated the N2 amplitude. When the previous trial was a no-go trial, the N2 was generally more negative than after a go trial, and it did not differ statistically between switch and repetition trials. When both the task and response mode were repeated, the N2 was smaller and peaked earlier than in remaining switch conditions. Moreover, residual switch costs correlated negatively with the corresponding difference in amplitude after go trials, and positively with the latency effect of the frontocentral N2. The P2-N2 amplitude at Fz was correlated with RTs.

4.

Discussion

In the present study, we tested the hypothesis that residual shift costs arise as a result of interference between overlapping task sets during selection of an appropriate response. This approach was tested using a combined task-switching and go/no-go paradigm which required occasionally suppressing overt responses during switching or repeating tasks. In particular we investigated modulation of an electrophysiological correlates of interference: the N2.

4.1.

Behavior

We replicated the behavioral findings previously obtained within this paradigm (Astle et al., 2006; Kleinsorge and Gajewski, 2004; Koch and Philipp, 2005; Schuch and Koch, 2003) that indicated large RT – and error switch costs in trials following a go trial, and an absence of those costs after no-go trials. It is important to note however, that the RT level after no-go trials corresponds to the RT level on switch trials after go trials. Markedly faster responses were observed only in TR trials following another go trial. Thus, this pattern should be interpreted as a repetition benefit of a repeated and executed task, which is eliminated when either the task or response mode (go vs no-go) changed, suggesting that all other conditions may reflect interference either between response modes (no-go vs go) or task sets. At first glance our finding of elimination of residual switch costs due to lack of response selection in n  1 seems 3

Although in trials following a no-go trial the switch costs amounted to 3 msec and were of course non significant, we used here the term switch costs to denote the RT-difference between switch and non-switch trials.

1146

cortex 46 (2010) 1138–1148

incompatible to the findings of Verbruggen et al. (2007) and Monsell and Mizon (2006). For example Verbruggen and coworkers observed reduction of residual switch costs by enhancing participant’s efficiency to prepare the next task. However, as acknowledged by the authors, the reduction of residual costs may be due to a combination of different factors (e.g., a relatively long RSI, which may abolish the effect of task activation or inhibition from previous trial combined with a phasic increase of effort to prepare for the next task etc.). Importantly, as already mentioned by Philipp et al. (2007) there may be a different mechanism underlying the elimination of shift costs in our or Philipp et al.’s (2007) and Verbruggen et al.’s (2007) study: the reduction of residual switch costs was mainly due to faster responses in switch trials in Verbruggen’s study, whereas this effect was clearly due to slower responses in repetition trials after no-go trials in the present study. This is consistent with the notion that task preparation may selectively affect switch trials, whereas no-go trials led to abolishing of the TR benefit that seems to be uninfluenced by preparation. In sum, our finding did not deny the possibility that residual switch costs could be diminished by advanced preparation but the detailed mechanisms how preparation modulates the residual costs are largely unknown. However, the bulk of research shows that residual switch costs are relative insensitive to preparation processes (e.g., Meiran, 2000; Mayr and Keele, 2000; Rogers and Monsell, 1995). According to the modified activation/inhibition account of switch costs suggested by Koch and Philipp (2005) and Philipp et al. (2007), in switch trials response selection can take place when competition between simultaneously active C–R rules (from the previous and the current trial) has been resolved. The competition is resolved when the activation level of the currently relevant category exceeds the response threshold, whereas the irrelevant C–R has been successfully inhibited. When the C–R rule from the previous trial is repeated, its activation is refreshed leading to a repetition benefit of an already selected and performed task. If a response mode has to be switched, similarly to switching a task, the inappropriate response mode has to be inhibited and the relevant one activated. This leads to costs of switching among response modes (Koch et al., 2004). Following this reasoning the residual switch costs are due to a repetition benefit of a repeated task and response mode relative to conditions where the task, the response mode or both have to be revised. It is important to note that switching among response modes and task sets seems to proceed in parallel because switching of both dimensions did not take more time than switching of only one of them (see also Kleinsorge et al., 2005).4 A second possibility may be an unspecific aftereffect of an inhibited response in n  1 which overshadows any task-specific effect from n  1. Unfortunately, we cannot decide between these possibilities on the basis of our data.

4.2.

ERPs

Overall there was an N2 with frontocentral maximum in all conditions, not only in switch trials. Moreover we observed a significant enhancement of the N2 in response to TSs 4

We thank an anonymous reviewer for stressing this point.

relative to TRs after go trials. This difference was absent when the previous trial was a no-go trial. An equivalent pattern appeared in the N2 latency, suggesting that not only the strength of activity but also the timing of this effect corresponded to the behavioral data. Furthermore, a statistical relationship between residual switch costs and the N2 amplitude- and latency-effects implies that both parameters are correlates of the observed residual switch costs. These findings confirm the hypothesis that the frontocentral N2 reflects a decision process (Ritter et al., 1979, 1982), e.g., the mapping of a stimulus onto a response category, which is intensified and prolonged when interference between C–R associations has to be resolved. In the additional analysis we attempted to disentangle the possible confound between task transition and response hand transition, which could be responsible for the observed data pattern. This analysis showed that the N2 switch effect was not influenced by repeating versus switching response hands. This suggests that the N2 effect does not reflect conflict at the stimulus level, nor at the level of an individual response (e.g., van Veen and Carter, 2002a) but rather at the level of C–R – sets. This hypothesis was additionally supported by the finding that incongruent trials that is trials with overlapping S-R mappings consistently increase the N2. These observations indicate susceptibility of the N2 to interference at an abstract level of information processing. It might be argued that the N2 is unlikely to reflect a decision process such as response selection because it peaks much earlier than the overt response, and the switch costs are larger for RT than for N2 peak latency. However, the decision process reflected in the N2 may well last longer than the N2 peak, which is also reflected in the prolonged enhancement in the switch versus task repetition ERPs in the N2 range after the peak (cf. Fig. 2). Also motor activation may well take several hundred milliseconds before the response threshold is reached (Fiedler et al., 2009). Finally, for more difficult decisions (e.g., switch) the response threshold may be temporarily increased, which would prolong RT more than the decision itself. Another intriguing point deserves attention: Why was the N2 altogether more negative in trials following no-go relative to go trials? A plausible explanation is that this unspecific N2 enhancement reflects also interference among two response modes (no-go and go) in successive trials that was also reflected in the RT-data. Alternatively, the N2 may be superimposed on the long lasting slow wave developing as a consequence of response inhibition in n  1 and/or advanced preparation for the following trial. In a similar vein, the N2 may be larger after no-go trials because a response has to be definitely given in the following trial. It is important to emphasize the fact that a N2 was also present in repetition trials. The ‘‘repeat’’-N2 was smaller and occurred slightly earlier than a switch – N2 after go trials but had the same topography as the ‘‘switch’’ N2, suggesting that it reflects the same functional process. Support for this notion stems first from the significant relationship between the N2 amplitude and N2 latency – and RT – differences for go trials following go as well as no-go trials and second from the significant correlations between the P2-N2 index and RTs in each condition. A similar relationship between the N2 and RTs in conflict and non-conflict trials was also found in our

cortex 46 (2010) 1138–1148

previous study (Gajewski et al., 2008). The finding that compatible and incompatible trials both evoked an N2, and that its latency correlated with RTs in compatible trials as well suggests that the N2 is not reflecting interference per se but may be related to the more general process of selection, which is modulated, i.e., more strongly activated, by interference. Therefore its peak is increased and delayed due to a longer activation time of the correct response until a selection threshold is exceeded in incompatible or switch compared to compatible or repetition trials. This explanation nicely fits with the interpretation of residual switch costs proposed by Schuch and Koch (2003) who concluded that the absolute activation level of C–R associations may explain slowing in switching trials: the further the activation level is to the activation threshold, the slower the response selection.

5.

Conclusions

The inhibition and activation account of task sets which delays response selection proposed by Allport et al. (1994) offers a simple and plausible mechanism, which may underlie switch costs. Supported by our data, we propose that these costs could be sufficiently explained by this approach. Thus, the central message of the present study is that differences in task activation levels during selection of a correct response, reflected in the timing and the amplitude of the N2 may be the main source of residual switch costs.

Acknowledgements We thank Claudia Wipking for running the experiment and assistance in analyzing the data and Ludger Blanke for providing the software and technical support. The research reported in this article was supported by grant Kl 1205/3-3 of the Deutsche Forschungsgemeinschaft to the second author (TK). We thank three anonymous reviewers for many helpful comments on an earlier draft of the manuscript.

references

Allport DA, Styles EA, and Hsieh S. Shifting intentional set: Exploring the dynamic control of tasks. In Umilta` C and Moscovitch M (Eds), Attention and Performance XV. Hillsdale, NJ: Erlbaum, 1994: 421–452. Allport DA and Wylie GR. Task switching, stimulus-response bindings and negative priming. In Monsell S and Driver J (Eds), Control of Cognitive Processes: Attention and Performance XVIII. Cambridge, MA: MIT Press, 2000: 35–70. Astle DE, Jackson GM, and Swainson R. Dissociating neural indices of dynamic cognitive control in advance task-set preparation: An ERP study of task-switching. Brain Research, 1125: 94–103, 2006. Botvinick MM, Nystrom LE, Fissell K, Carter CS, and Cohen JD. Conflict monitoring versus selection-for-action in anterior cingulate cortex. Nature, 402: 179–181, 1999.

1147

Carriero L, Zalla T, Budai R, and Battaglini PP. Inhibition of wrong responses and conflict resolution: An electroencephalogram study. NeuroReport, 28: 793–796, 2007. Chen A, Xu P, Wang Q, Luo Y, Yuan J, Yao D, et al. The timing of cognitive control in partially incongruent categorization. Human Brain Mapping, 29: 1028–1039, 2007. De Jong R. An intention-activation account of residual switch costs. In Monsell S and Driver J (Eds), Control of Cognitive Processes: Attention and Performance XVIII. Cambridge, MA: MIT Press, 2000: 357–376. Fiedler A, Schro¨ter H, and Ulrich R. No evidence for a late locus of task switch effects. Brain Research, 1253: 74–80, 2009. Folstein JR and Van Petten C. Influence of cognitive control and mismatch on the N2 component of the ERP: A review. Psychophysiology, 45: 152–170, 2008. Gajewski PD, Stoerig P, and Falkenstein M. ERP-correlates of response selection in a response conflict paradigm. Brain Research, 1189: 127–134, 2008. Gehring WJ, Bryck RL, Jonides J, Albin RL, and Badre D. The mind’s eye, looking inward? In search of executive control in internal attention shifting. Psychophysiology, 40: 572–585, 2003. Goschke T. Voluntary action and cognitive control from a cognitive neuroscience perspective. In Maasen S, Prinz W, and Roth G (Eds), Voluntary Action: An Issue at the Interface of Nature and Culture. Oxford: Oxford University Press, 2002. Gratton G, Coles MGH, and Donchin E. A new method for off-line removal of ocular artifact. Electroencephalography and Clinical Neurophysiology, 55: 468–484, 1983. Jackson GM, Swainson R, Cunnington R, and Jackson SR. ERP Correlates of executive control during repeated languageswitching. Bilingualism: Language and Cognition, 4: 169–178, 2001. Jackson GM, Swainson R, Mullin A, Cunnington R, and Jackson R. ERP correlates of receptive language switching. Quarterly Journal of Experimental Psychology, 57A: 223–240, 2004. Jasper HH. The 10–20 electrode system of the International Federation. Electroencephalography and Clinical Neurophysiology, 10: 370–375, 1958. Karayanidis F, Coltheart M, Michie PT, and Murphy K. Electrophysiological correlates of anticipatory and poststimulus components of task switching. Psychophysiology, 40: 329–348, 2003. Kieffaber PD and Hetrick WP. Event-related potential correlates of task switching and switch costs. Psychophysiology, 42: 56–71, 2005. Kiesel A, Wendt M, and Peters A. Task switching: On the origins of response congruency effects. Psychological Research, 71: 117–125, 2007. Kleinsorge T and Gajewski PD. Preparation for a forthcoming task is sufficient to produce subsequent shift costs. Psychonomic Bulletin and Review, 11: 302–306, 2004. Kleinsorge T, Gajewski PD, and Heuer H. Task sets under reconstruction: Effects of partially incorrect precues. Quarterly Journal of Experimental Psychology, 58a: 521–546, 2005. Klinke R, Fruhstorfer H, and Finkenzeller P. Evoked responses as a function of external and stored information. Electroencephalography and Clinical Neurophysiology, 25: 119–122, 1968. Koch I, Gade M, and Philipp AM. Inhibition of response mode in task switching. Experimental Psychology, 51: 52–58, 2004. Koch I and Philipp AM. Effects of response selection on the task repetition benefit in task switching. Memory and Cognition, 33: 624–634, 2005. Kopp B, Mattler U, Goertz R, and Rist F. N2, P3 and the lateralized readiness potential in a no-go task involving selective response priming. Electroencephalography and Clinical Neurophysiology, 99: 19–27, 1996a. Kopp B, Rist F, and Mattler U. N200 in the flanker task as a neurobehavioral tool for investigating executive control. Psychophysiology, 33: 282–294, 1996b.

1148

cortex 46 (2010) 1138–1148

Kotchoubey B, Wascher E, and Verleger R. Shifting attention between global features and small details: An event-related potential study. Biological Psychology, 46: 25–50, 1997. MacDonald AW, Cohen JD, Stenger VA, and Carter CS. Dissociating the role of the dorsolateral prefrontal and anterior cingulated cortex in cognitive control. Science, 288: 1835–1838, 2000. Mayr U and Keele SW. Changing internal constraints on action: The role of backward inhibition. Journal of Experimental Psychology: General, 129: 4–26, 2000. Meiran N. The reconfiguration of processing mode prior to task performance. Journal of Experimental Psychology: Learning, Memory and Cognition, 22: 1423–1442, 1996. Meiran N. Reconfiguration of stimulus task sets and response task sets during task switching. In Monsell S and Driver J (Eds), Control of Cognitive Processes: Attention and Performance XVIII. Cambridge, MA: MIT Press, 2000: 377–399. Monsell S and Mizon GA. Can the task-cueing paradigm measure an ‘‘endogenous’’ task-set reconfiguration process? Journal of Experimental Psychology: Human Perception and Performance, 32: 493–516, 2006. Nicholson R, Karayanidis F, Poboka D, Heathcote A, and Michie PT. Electrophysiological correlates of anticipatory taskswitching processes. Psychophysiology, 42: 540–554, 2005. Nicholson R, Karayanidis F, Davies A, and Michie PT. Components of task-set reconfiguration: Differential effects of ‘switch-to’ and ‘switch-away’ cues. Brain Research, 1121: 160–176, 2006. Philipp AM, Jolicoeur P, Falkenstein M, and Koch I. Response selection and response execution in task switching: Evidence from a go-signal paradigm. Journal of Experimental Psychology: Learning, Memory and Cognition, 33: 1062–1075, 2007. Picton TW, Bentin S, Berg P, Donchin E, Hillyard SA, Johnson Jr R, et al. Guidelines for using human event-related potentials to study cognition: Recording standards and publication criteria. Psychophysiology, 37: 127–152, 2000. Renault B and Lesevre N. A trial-by-trial study of the visual omission response in reaction time situations. In Lehmann D and Callaway E (Eds), Human Evoked Potentials. New York: Plenum Press, 1979: 317–329. Renault B, Ragot R, and Lesevre N. Correct and incorrect responses in a choice reaction time task and the endogenous components of the evoked potential. In Kornhuber HH and Deecke L (Eds), Motivation, Motor and Sensory Processes of the Brain Electrical Potentials, Behavior and Clinical Use. Amsterdam: Elsevier, 1980: 647–654. Ritter W, Simson R, Vaughan Jr HG, and Friedman D. A brain event related to the making of a sensory discrimination. Science, 203: 1358–1361, 1979.

Ritter W, Simson R, Vaughan Jr HG, and Macht M. Manipulation of event-related potential manifestations of information processing stages. Science, 218: 909–911, 1982. Ritter W, Simson R, and Vaughan Jr HG. Event-related potential correlates of two stages of information processing in physical and semantic discrimination tasks. Psychophysiology, 20: 168–179, 1983. Rogers RD and Monsell S. The costs of a predictable switch between simple cognitive tasks. Journal of Experimental Psychology: General, 124: 207–231, 1995. Rushworth MF, Passingham RE, and Nobre AC. Components of switching intentional set. Journal of Cognitive Neuroscience, 14: 1139–1150, 2002. Rushworth MF, Passingham RE, and Nobre AC. Components of attentional set-switching. Experimental Psychology, 52: 83–98, 2005. Schuch S and Koch I. The role of response selection for inhibition of task sets in task shifting. Journal of Experimental Psychology: Human Perception and Performance, 29: 92–105, 2003. Simson R, Vaughan Jr HG, and Ritter W. The scalp topography of potentials associated with missing visual and auditory stimuli. Electroencephalography and Clinical Neurophysiology, 40: 33–42, 1976. Simson R, Vaughan Jr HG, and Ritter W. The scalp topography of potentials in auditory and visual discrimination tasks. Electroencephalography & Clinical Neurophysiology, 42: 528–535, 1977. Swainson R, Cunnington R, Jackson GM, Rorden C, Peters AM, Morris PG, et al. Cognitive control mechanisms revealed by ERP and fMRI: Evidence from repeated task-switching. Journal of Cognitive Neuroscience, 15: 785–799, 2003. van Veen V and Carter CS. The anterior cingulate as a conflict monitor: fMRI and ERP studies. Physiology and Behavior, 77: 477–482, 2002a. van Veen V and Carter CS. The timing of action-monitoring processes in the anterior cingulate cortex. Journal of Cognitive Neuroscience, 14: 593–602, 2002b. Verbruggen F, Liefooghe B, Vandierendonck A, and Demanet J. Short cue presentations encourage advance task preparation: A recipe to diminish the residual switch cost. Journal of Experimental Psychology: Learning, Memory and Cognition, 33: 342–356, 2007. Wylie GR, Javitt DC, and Foxe JJ. Task switching: A high-density electrical mapping study. NeuroImage, 20: 2322–2342, 2003. Yeung N and Cohen JD. Impact of cognitive deficits on conflict monitoring. Predictable dissociations between the errorrelated negativity and N2. Psychological Science, 17: 164–171, 2006.