Investigating the role of conflict resolution in memory updating by means of the one-back choice RT task

ORIGINAL ARTICLE

Investigating the role of conflict resolution in memory updatingby means of the one-back choice RT task

Arnaud Szmalec Æ Jelle Demanet ÆAndre Vandierendonck Æ Frederick Verbruggen

Received: 13 November 2007 / Accepted: 29 March 2008 / Published online: 24 April 2008

� Springer-Verlag 2008

Abstract The current study is inspired by recent findings,

which suggest that conflict is involved in the updating of

memory representations. It directly addresses the relation

between memory updating and conflict resolution by

means of the one-back choice reaction time (RT) task, an

updating task, which requires participants to postpone their

response to stimulus n until the subsequent stimulus n + 1

has been presented. In three experiments, a more detailed

analysis of the one-back choice RT task is presented in

order to further identify the role of conflict resolution in

memory updating. The findings demonstrate that the one-

back choice RT task, which allows motor preparation just

like a simple RT task, is in fact performed slower than a

simple RT task because it additionally involves conflict

resolution. It is further shown that also the response–

stimulus interval of the one-back task involves processes

that affect the amount of conflict in the task. In the

‘‘General discussion’’, the theoretical relevance of these

findings for the concept of updating is discussed.

Introduction

In order to explain goal-directed behaviour, cognitive sci-

entists have proposed constructs such as the Central

Executive (Baddeley & Hitch, 1974), executive attention

(e.g., Engle, Kane, & Tuholski, 1999) or executive func-

tioning (e.g., Rabbitt, 1997). These notions suggest that

specialized cognitive mechanisms intervene in order to

achieve a goal but very often, the mediating entity is so

poorly specified that it seems as if a homunculus performs

the critical interventions. Because this state of affairs is

scientifically not tenable, efforts have been made to frac-

tionate or decompose these intervening homunculi into a

number of autonomous executive functions (e.g., Baddeley,

1996; Miyake et al., 2000). While a conception proposing a

multiplicity of executive functions for achieving behav-

ioural goals (e.g., Miyake et al., 2000) may be considered

as a progress in comparison to a framework assuming a

single executive control entity (Baddeley, 1996), it is likely

that even more specific processes or sequences of processes

(such as control loops, e.g., Botvinick, Braver, Barch,

Carter, & Cohen, 2001) have to be specified in order to

completely get rid of the homunculus idea. Within this

context, it is important to critically analyze the often pos-

tulated executive functions such as inhibition, updating,

mental shifting, planning or dual-task coordination, with

the aim of achieving a more precise characterization of the

control processes that are involved. In the present paper,

we distinguish the terms ‘control processes’ and ‘executive

functions’, based on the view that specific control pro-

cesses, which are the object of this study, may underlie

executive functions. The introduction of this paper is

structured as follows. First, we go more deeply into exec-

utive function research and focus on a number of studies

that we have conducted earlier and that are relevant in

order to understand the current experimental approach.

Next, the concept of memory updating is introduced, along

with our recently developed updating paradigm, i.e., the

one-back choice RT (CRT-1) paradigm. Finally, we

A. Szmalec (&) � J. Demanet � A. Vandierendonck �F. Verbruggen

Department of Experimental Psychology,

Ghent University, Henri Dunantlaan 2, 9000 Ghent, Belgium

e-mail: [email protected]

F. Verbruggen

Department of Psychology, Vanderbilt University,

Nashville, TN, USA

123

Psychological Research (2009) 73:390–406

DOI 10.1007/s00426-008-0149-3

present the goal of this study, which is to further explore

the relation between updating and conflict resolution, by

using the CRT-1 paradigm.

One important distinction to keep in mind when ana-

lyzing executive functions is the distinction between an

executive function and an executive task. In fact, the tax-

onomy of executive functions has initially been defined

mainly by differentiating a number of tasks that were

assumed—often at face validity—to measure concepts like

inhibition, shifting or updating. However, this differentia-

tion does not automatically imply that these tasks also

involve functionally separable executive functions. One

theoretically plausible alternative is that these functions are

not autonomous, but rather task requirements that are met

by a common control process or control loop (see e.g.,

Braver, Gray, & Burgess, 2007). Therefore, efforts to

investigate executive control must go hand in hand with

attempts to better understand the processes and mecha-

nisms involved in tasks that are assumed to operationalize

popular executive functions, an approach that has already

proven to be fruitful before (Miyake et al., 2000; Rabbitt,

1997; Szmalec, Vandierendonck, & Kemps, 2005) and that

will also be applied in the current study.

There are two further important concerns with respect to

the use of executive tasks. One is that tasks that involve

executive control always seem to require some form of

decision or choice process in the sense that a response must

be chosen among a number of alternatives (i.e., response

selection, see e.g., Hegarty, Shah, & Miyake, 2000).

Because response selection also involves executive control

(e.g., Szmalec et al., 2005), one should be aware that any

task that requires response selection, also measures execu-

tive control to some degree (Hegarty et al., 2000). The other

concern is that each of the popular executive tasks calls on a

multitude of processes, of which only some may be relevant

to the issue of executive control. Therefore, to the extent

that short-term memory processes are also involved in

executive tasks, it is important that modality-specific (e.g.,

verbal short-term memory) processing does not obscure the

executive nature of the task. This is particularly true for

dual-task paradigms, in which secondary tasks are often

used to investigate executive control involvement (e.g.,

Kemps, Szmalec, Vandierendonck, & Crevits, 2005). In

particular, it has been shown within the working memory

framework of Baddeley (1986) that an executive demand-

ing secondary task that also calls on one of the modality-

specific slaves (e.g., the Phonological Loop) may interfere

with a primary memory task because it also requires the

same slave system. This way, it is possible that two tasks

interfere in the absence of any controlled processing.

In previous research, we have tried to minimize such

modality-specific interference effects by using tasks with a

minimal memory load (Szmalec et al., 2005). On the one

hand, we used a standard choice RT task (CRT task). A

CRT task requires that a response is selected among a

number of alternatives (e.g., choosing a left or right

response to the identity of a stimulus, such as a low or high

tone). On the other hand, we used a simple RT task (SRT

task). A SRT task does not involve a response selection

stage (e.g., Schubert, 1999) and the same response can be

produced, irrespective of the identity of the stimulus. We

examined the interference due to response selection by

comparing the patterns of dual-task interference of both

these RT tasks with a number of established working

memory tasks that involve both executive and domain-

specific processing. We showed that merely the fact that

response choices had to be produced was sufficient to

interfere with working memory resources. This interfer-

ence proved to be consistent with an executive interference

pattern and was completely different from the signature of

interference produced at the level of working memory’s

slave systems. This suggests that executive control must

have been involved in response selection, consistent with

the hypothesis of many authors (e.g., Hegarty et al., 2000;

Rowe, Toni, Josephs, Frackowiak, & Passingham, 2000;

Smyth & Scholey, 1994).

In the present paper, we focus on the executive function

of memory updating, which is the act of modifying the

current status of a representation in memory to accommo-

date new input (Morris & Jones, 1990). During the last

decade, the n-back task has proven to be a useful opera-

tionalization of updating (e.g., Miyake et al., 2000; Smith &

Jonides, 1999; Van der Linden et al., 1999). Building on the

same methodology as Szmalec et al. (2005), i.e., minimal

memory load and usage of the choice reaction time para-

digm, we recently developed an updating task, namely the

one-back choice reaction time task, which differs from a

standard choice RT task in the requirement to delay the

choice response to stimulus n - 1 until the subsequent

stimulus n has been presented (Szmalec & Vandierendonck,

2007). Figure 1 illustrates the one-back choice RT task in

comparison to a simple RT task and a choice RT task. The

one-back choice RT task can be seen as a choice reaction

time task variant of the n-back memory updating procedure

(Smith & Jonides, 1997). The classical n-back task requires

participants to determine whether each item (e.g., a letter) in

a list matches the item that was presented n positions back.

Therefore, they are required to remember a specified

number (n) of the most recent items serially (n-back). This

implies that, while the task evolves and new items are

presented, participants have to update the memorized string

of n most recent items: they need to drop the oldest item in

the string and add the most recent one.

In a similar way, participants have to keep an active

representation of the information that will be necessary in

order to make a correct response to stimulus n - 1 when

Psychological Research (2009) 73:390–406 391

123

stimulus n is presented in the one-back choice RT task.

This means that as soon as stimulus n is presented, the

response to the previous stimulus n - 1 is executed and

the active representation is updated or refreshed with the

information that will be necessary to make a correct

response to stimulus n when stimulus n + 1 will be pre-

sented, and so on. Therefore, a representation of task-

relevant information must be kept active throughout the

task and this representation must be updated each time new

information (i.e., a new stimulus) is presented. By analogy

with Szmalec et al. (2005), we recently compared the one-

back choice RT task with a standard choice RT task in

terms of dual-task interference with a number of estab-

lished domain-specific and executive working memory

measures (Szmalec & Vandierendonck, 2007). We dem-

onstrated that the additional working memory demands of

the one-back task, compared to the choice RT task, selec-

tively interfered with measures of executive control and

thus not with measures of modality-specific (i.e., verbal or

visuospatial) processing. On the basis of these findings, we

concluded that updating taxes executive resources. Mean-

while, the one-back choice RT task has been used as a

measure of updating in other dual-task studies, for example

in order to investigate the role of memory updating in

mental arithmetic (Deschuyteneer, Vandierendonck, &

Muyllaert, 2006).

The present study is directed towards the relation

between updating and conflict resolution. A number of

recent findings namely suggest that conflict resolution is

involved in updating the contents of memory. More pre-

cisely, several studies with the n-back updating task used

so-called n + x and n - x ‘lure’ trials in order to increase

the executive control demands of the n-back task (an

example of an n + 1 lure trial in a 2-back task is B–F–L–B,

e.g., Kane, Conway, Miura, & Colflesh, 2007). Such lure

trials are assumed to augment the amount of interference in

the n-back task (see Jonides and Nee, 2006) because these

trials induce conflict1 between the activation of new mem-

ory contents and the suppression of old memory contents

throughout the updating procedure. The observation that

conflict occurs in the n-back task suggests that updating the

contents of memory involves the resolution of conflict at the

level of activations in memory. This interpretation is of

particular relevance in the debate about whether different

executive tasks also measure different executive functions.

Suppose that the executive demands of memory updating

can be entirely grasped by the resolution of conflict between

competing representations. This would mean that memory

contents can be refreshed relatively effortlessly when there

is no conflict between the working memory representations.

By contrast, executive control would be involved when

conflict arises between different working memory repre-

sentations. This would suggest that memory updating itself

is not an autonomous executive function, but a task demand

that, in particular circumstances, is achieved by a control

process that resolves conflicts at the level of memory rep-

resentations. Similar concerns have been raised by Bunting,

Cowan and Saults (2006), who distinguish two different

ways of refreshing memory representations in the running

memory span task (another well-known measure of updat-

ing) based on temporal presentation parameters, namely a

passive, relatively automated and an active, more controlled

way. This distinction between relatively automated and

controlled ways of refreshing memory contents also chal-

lenges the functional autonomy of updating within

executive functioning.

Fig. 1 Illustration of the one-

back choice RT task (CRT-1),

compared to a simple RT task

(SRT) and a regular choice RT

task (CRT). An SRT task

always requires the same

response, irrespective of the

identity of the stimulus (i.e., a

high or low pitch tone). A CRT

task requires a response choice

(i.e., high or low response) as a

function of the identity of the

stimulus. In the CRT-1 task, on

each trial (except the first one),

participants respond to the

identity of the previous stimulus

(one-back)

1 Note that we prefer the term conflict resolution over interference

because it offers a broader and, as we will argue further, more

appropriate description of the processes involved in updating

392 Psychological Research (2009) 73:390–406

123

The present study was designed to examine whether

conflict is also involved in updating tasks that have no

distinct verbal or visuospatial requirements. In order to

achieve this aim, the recently developed CRT-1 updating

paradigm is used to directly address the relation between

updating and conflict resolution. The observation that the

one-back choice RT task shows no measurable involve-

ment of the working memory slave systems (Szmalec &

Vandierendonck, 2007), whereas de n-task does (e.g.,

Smith & Jonides, 1997), yields an important reason for

addressing this issue by using the one-back choice RT task.

This will enable us to understand whether the relation

between conflict resolution and updating is mediated by

modality-specific processing in one particular type of

updating task or whether it is rather a general property of

updating which is not dependent on the type of updated

materials.

Our goal will be pursued by presenting an in-depth

analysis of the cognitive processes involved in the one-

back choice RT task (with focus on conflict resolution),

inspired by two earlier observations with this task. The first

observation is that the one-back choice RT task is per-

formed slower than a simple RT task, which does not

require a response choice (see Fig. 1), but only a response

irrespective of the stimulus category. Whereas simple

reactions are quite fast (on average less than 300 ms, e.g.,

Szmalec et al., 2005), one-back choice reactions appear to

be much slower (on average 450 ms, Szmalec & Van-

dierendonck, 2007). This finding is remarkable given that

execution of the response in the one-back task is in essence

a simple reaction because the response can be selected

prior to stimulus presentation. This RT difference between

the SRT and CRT-1 task leads us to assume that there is

some additional processing going on between stimulus

presentation and response execution in the one-back choice

RT task, resulting in the observed RT increase. The second

observation is that one-back choice reactions are on aver-

age 50 ms faster than regular choice reactions, whereas the

one-back choice RT task causes more dual-task interfer-

ence than a standard choice RT task (Szmalec &

Vandierendonck, 2007). This suggests that part of the

processing in the one-back choice RT task is not captured

by the reaction time, possibly because it occurs after the

response has been executed.

In three experiments, we further investigate the relation

between updating and conflict resolution, using the CRT-1

task. In ‘‘Experiment 1’’, we offer a more detailed analysis

of the CRT-1 task by distinguishing four different types of

one-back trials, based on stimulus/response repetition/

alternation. In ‘‘Experiment 2’’, we further tested the

hypothesis that the CRT-1 task is performed slower than a

simple RT task because it additionally involves resolution

of conflict. Finally, ‘‘Experiment 3’’ was designed to

demonstrate that the greater dual-task interference of the

CRT-1 task is due to processing in the response–stimulus

interval (i.e., the interval between the execution of the

response and the next stimulus) and that these inter-trial

processes influence the degree of conflict in the CRT-1

task. In essence, the manipulations in ‘‘Experiments 1–3’’

are directed towards the occurrence of conflict in the one-

back choice RT task. They are aimed to understand whe-

ther the differences in executive control demands between

the one-back task and other RT tasks, which have initially

been ascribed to updating requirements, can be attributed

to conflict resolution.

Experiment 1

In the one-back choice RT task, stimulus n is a signal for

the response to stimulus n - 1. Usually, the presentation

time of stimulus n (200 ms in Szmalec & Vandierendonck,

2007) is shorter than the observed average RT to stimulus

n - 1, which means that the new stimulus has already

disappeared when the response to the previous one is

executed. Thus in order to proceed with the task, a repre-

sentation of stimulus n should be available to the

participant after response n - 1 was executed. This means

that stimulus n is likely to be processed to some extent

before the response to stimulus n - 1 is terminated. In

other words, there is an overlap between the processing of

two successive trials. An important question in relation to

this overlap is to what extent stimulus n is processed.

Imagine that stimulus n - 1 was a low tone, which

requires a left response, whereas stimulus n is a high tone,

which requires a right response. If stimulus n is already

processed to a relatively late (response-related) stage, this

could give rise to competition between both responses.

This is in line with the post-hoc analyses of the CRT-1 task

reported in Szmalec and Vandierendonck (2007). These

analyses showed that stimulus repetitions (i.e., stimulus

n - 1 and n are the same) are processed much faster than

stimulus alternations, which suggests that early processing

of the trigger signal n can produce competition or facili-

tation with the final processing of stimulus n - 1.

In fact, whereas stimulus repetition/alternation auto-

matically involves response repetition/alternation in a

standard CRT task, stimulus and response repetition/alter-

nation can be separated in the CRT-1 task. More precisely,

four types of possible trials can be distinguished, as a

function of repetition and alternation of both task features

(i.e., the stimulus and the response). They are represented

in Fig. 2. The first type involves a repetition of stimulus

and response (henceforth referred to as S=R= trials). In the

second type, the stimulus is repeated while the response is

alternated (S=R=). In the third type, the stimulus is


123

alternated but the response repeated (S=R=). In the fourth

type, both task features are alternated (S=R=). In the first

experiment, participants were asked to perform the CRT-1

task with equal proportions of the four trial types described

above. Performance on these trials was compared with

performance on a regular CRT task.

Method

Participants and design

Twelve students (nine females; mean age 21 years; range

18–23) enrolled at the Faculty of Psychology and Educa-

tional Sciences at Ghent University were paid €10 for

taking part in the study. Each participant performed both

the CRT task and the CRT-1 task. The order of the tasks

was counterbalanced across subjects.

Materials and procedure

All experimental procedures reported in this paper were

administered using the Tscope C/C++ programming library

(Stevens, Lammertyn, Verbruggen, & Vandierendonck,

2006). For both tasks, we used two easily discriminable

tones with a frequency of 262 (C1 note) and 524 Hz (C1

plus one octave), presented through a closed headphone

(Sennheiser HD 265-1) at 60 dB. Each tone lasted 150 ms.

Participants were required to hit the right key or the left key

of a response box connected to the parallel port, as fast and

accurately as possible after they heard the low or the high

frequency tone, respectively. They were instructed to rest

their index fingers on the response keys to avoid that RTs

become contaminated with movement time. In the CRT-1

task, participants were required to delay their response until

the next stimulus occurred. In order to discourage antici-

pations in the CRT-1 task, the inter-stimulus interval of both

RT tasks was either 1,000 or 2,000 ms, randomly chosen

with the constraint that no more than three consecutive

intervals were of equal duration.

Prior to the counterbalanced conditions, participants

practiced the CRT task and the CRT-1 task until they met a

criterion of 20 consecutively correct (one-back) choice

reactions. In the counterbalanced conditions, 480 trials

were presented for each RT task, divided in four blocks of

120 trials. Stimuli in both tasks were pseudo-randomized

so that a sound could not be repeated more than four times

in a row. In the CRT-1 task, the four types of stimulus/

response alternation/repetition (cf. Fig. 2) were equally

distributed (i.e., 120 trials per type). The entire procedure

lasted approximately 60 min.

Results

We first conducted a within-participant data trimming

procedure in which all RTs that deviated more than 2.5 SDs

from the mean, were discarded. This resulted in a data

reduction of 1.8%. Only RTs of correct trials were included.

Performance was analyzed by means of a one-way ANOVA

with the factor Task (CRT task, CRT-1 task). The RT data

show that the CRT-1 task (M = 379, SD = 74) was per-

formed faster than the CRT task (M = 432, SD = 88),

F(1,11) = 5.57, np2 = 0.34, p \ 0.05. Although the

Fig. 2 The four different types

of trials in the one-back choice

RT task (CRT-1) as a function

of stimulus and response

repetition. S=R= involves both

stimulus and response

repetition, S=R= stimulus

repetition and response

alternation, S=R= stimulus

alternation and response

repetition, and S=R= both

stimulus and response

alternation


123

proportion of correct responses in the CRT-1 task

(M = 0.86, SD = 0.08) was lower than in the CRT task

(M = 0.90, SD = 0.06), the difference fell short from sta-

tistical significance, F(1,11) = 3.80, np2 = 0.26, p = 0.08.

We further broke down performance on the choice RT

task and the CRT-1 task based on the inter-stimulus

interval (1,000 or 2,000 ms). This was done in order to

explore whether differences between both RT tasks also

occur at the level of processing in the inter-stimulus

interval (see Discussion of ‘‘Experiment 1’’). We con-

ducted a post hoc comparison of (CRT and CRT-1) trials

that followed the 1,000 ms ISI, with trials that followed the

2,000 ms ISI. In the choice RT task, the difference between

the 1,000 ms condition (M = 421, SD = 69) and the

2,000 ms ISI condition (M = 442, SD = 102) was not

reliable, F(1,11) = 2.29, np2 = 0.17, p [ 0.10. In the CRT-

1 task, performance was slower for trials following the

1,000 ms ISI (M = 414, SD = 98), compared to trials

following the 2,000 ms ISI (M = 344, SD = 54),

F(1,11) = 11.52, np2 = 0.51, p \ 0.01.

Next, we conducted a two-way ANOVA with the factors

Task (CRT, CRT-1) and Trial (S=R=, S=R=, S=R=,

S=R=). The means for this analysis are displayed in

Fig. 3. Note that the ANOVA also includes performance on

the four trial types for the CRT task, whereas these types

do by definition not occur in the CRT task. In fact, trial

type values for the CRT task show the average RTs on the

same stimulus transitions as those used to determine the

different types in the CRT-1 task. For example, imagine a

sequence of stimuli that makes up a trial of type S=R= in

the CRT-1 task: high tone–high tone–low tone. An S=R=

trial cannot occur in a standard choice RT task, so what

Fig. 3 represents is CRT task performance on the (last

stimulus of the) triplet high tone–high tone–low tone. We

believe that comparing performance on the four different

CRT-1 task trials with performance on the identical stim-

ulus transition in the CRT task offers a better baseline for

comparison than calculating overall CRT task perfor-

mance. The reason is that it allows us to understand to

which extent the difference between the four trial types of

the CRT-1 task can be attributed merely to sequential

effects that also occur in a standard CRT task. Because the

trial types are determined by a sequence of stimuli and

responses, we also removed the stimulus that follows an

error from our dataset in order to obtain completely correct

transitions and hence avoid post-error effects (Rabbitt,

1968). In total, an additional 1.8% of error and post-error

trials were excluded. We observed a significant main effect

of Task, F(1,11) = 5.62, np2 = 0.34, p \ 0.05, a significant

main effect of Trial, F(3,33) = 50.74, np2 = 0.82,

p \ 0.001, and a significant interaction between both fac-

tors, F(3,33) = 14.86, np2 = 0.57, p \ 0.001.

Planned comparisons show that in the choice RT task,

S=R= trials (e.g., low–low–low; see Fig. 2) were per-

formed faster than S=R= trials (e.g., high–low–low),

F(1,11) = 128.13, np2 = 0.92, p \ 0.001. S=R= trials

were in turn faster than S=R= trials (e.g., low–low–high),

F(1,11) = 140.24, np2 = 0.93, p \ 0.001. Also S=R=

trials (e.g., low–high–low) produced faster RTs than trials

of the type S=R=, F(1,11) = 81.27, np2 = 0.88, p \ 0.001.

With respect to the CRT-1 task, only the S=R= trials

yielded reliably slower RTs than the other three trial types,

F(1,11) = 67.90, np2 = 0.86, p \ 0.001. Also worth men-

tioning in the context of possible speed-accuracy trade-offs

is that the slowest one-back trial (S=R=) also yielded

lower accuracies than the other three types: 78% for S=R=

trials versus 95, 86, and 83% for the S=R=, S=R= and

S=R= trials, respectively.

Discussion

The results of ‘‘Experiment 1’’ show that performance in

the CRT-1 task is determined by an interaction between the

stimulus and response features that repeat/alternate during

the task. This interaction indicates that the trial types S=R=,

S=R= and S=R= are performed nearly equally fast,

whereas S=R= trials are performed much slower than the

others in the CRT-1 task. Next, we discuss two possible

origins of additional slowing in S=R= trials.

First, the cognitive implications of a dissociation of

stimulus and response features can possibly be related to

feature binding (Hommel, 1998). According to feature

binding theory, both the stimulus and the response features

of a stimulus–response (S–R) trial are integrated in one

event. Hence, when both features are repeated or alternated

Fig. 3 Mean RTs in milliseconds for the four different trial types in

the one-back choice RT task (CRT-1) and the standard choice RT task

(CRT), in ‘‘Experiment 1’’. Vertical bars denote standard errors


123

on the next trial, performance is better than when only one

of both features is repeated (i.e., partial repetitions). The

reason is that re-viewing the same (or a different) stimulus

activates the associated (or alternative) response, which has

become incorrect on partial repetition trials. Therefore,

trials with stimulus repetition and response alternation or

stimulus alternation with response repetition (such as types

S=R= and S=R= in the one-back choice RT task) are

likely to cause interference between the automatically

activated incorrect response and the required response.

However, whereas the current findings regarding trial types

S=R=, S=R= and S=R= could be accounted for by fea-

ture binding, the results of trial type S=R= do not

correspond to the predictions derived from this theory.

A second possible explanation for the current pattern of

results is that in the one-back choice RT task, conflict

arises between the response required for stimulus n and the

response that is prepared for stimulus n - 1. This possi-

bility is supported by Szmalec and Vandierendonck (2007)

who demonstrated that in the CRT-1 task, stimulus repe-

titions (the present types S=R= and S=R=) are faster than

stimulus alternations (the present types S=R= and

S=R=). However, this second account also fails to

explain the entire pattern of results, since we observed in

‘‘Experiment 1’’ that the conclusions drawn by Szmalec

and Vandierendonck (2007) are true only for one type of

the stimulus alternations trials, i.e., the one type that

coincides with a response repetition (i.e., S=R=). Hence, it

remains uncertain whether performance on S=R= trials

indeed reflects conflict between response representations.

This concern is even more pertinent because the difference

between the four trial types is to some extent also mea-

surable in the standard choice RT task (see Fig. 3). This

observation indicates that part of the RT differences

between the trial types in the one-back task are not due to

conflict between response representations (because such

conflict does not occur in a standard choice RT task) but

rather reflect sequential effects of the various stimulus

transitions (i.e., bottom-up), which are task-independent.

Therefore, ‘‘Experiment 2’’ examines more directly whe-

ther S=R= trials are performed slower than the other types

because they additionally involve conflict.

Another point that deserves some attention concerns the

finding that RTs were slower for ISI = 1,000 ms than for

ISI = 2,000 ms in the one-back choice RT task, whereas

no such difference was observed in the choice RT task. In

the introduction, we argued that RTs are faster in the CRT-

1 task than in the CRT task, whereas the CRT-1 task causes

more dual-task interference than the CRT task (Szmalec &

Vandierendonck, 2007). This could indicate that part of the

processing in the CRT-1 task occurs after the one-back

response has been executed. In this view, the present post-

hoc finding that one-back reactions are slower following a

shorter ISI may indicate that the one-back choice RT task

differs from a standard choice RT task, also in terms of the

processing that occurs in the interval between the one-back

response and the presentation of the subsequent stimulus

(i.e., response–stimulus interval). This assumption will be

further tested in ‘‘Experiment 3’’.

A final remark concerns the observation that responses

in the CRT-1 task are faster but also less accurate than

those in the CRT task. In our view, this does not neces-

sarily reflect a speed-accuracy trade-off, for the following

reasons. First, although the CRT-1 task is clearly more

difficult than a regular CRT task (as evidenced by stronger

dual-task interference, Szmalec & Vandierendonck, 2007,

and by participants’ experience), this additional difficulty is

not reflected in the RTs because, as we will argue further in

this study, part of the CRT-1 task processing occurs in the

response–stimulus interval and is thus not reflected in the

RT, by which we observe a pattern of speed and accuracy

that undeservedly resembles that of a trade-off. Second, our

position is further strengthened by the finding that within

the CRT-1 task, the slowest trials (S=R=) also produced

most errors.

Experiment 2

The previous experiment demonstrated that four types of

stimulus transition trials could be distinguished in the one-

back choice RT task. We showed that performance was

slower for S=R= trials than for the three other trial types

in the CRT-1 task and we hypothesized that this detri-

mental effect on performance reflects conflict between the

processing of two successive stimuli, n - 1 and n.

‘‘Experiment 2’’ is aimed to test the conflict hypothesis

more directly and to investigate whether this hypothesis

can account for the reaction time differences observed

between a simple RT task and the one-back choice RT task.

As we mentioned in the introduction, reaction times in

the one-back choice RT task are on average slower than

those in a simple RT task (Szmalec & Vandierendonck,

2007). This finding is remarkable knowing that both RT

tasks allow that motor responses are prepared prior to

stimulus presentation and are thus expected to yield com-

parable reaction times. We distinguish two possible

accounts for this observation. The first one is the conflict

hypothesis, which, as explained before, assumes that the

additional slowing of the CRT-1 task is a result of a

competition between the processing of two successive

stimuli. The second hypothesis explains the slowing of the

one-back choice RT task, relative to a simple RT task, in

terms of response selection. In the one-back choice RT

task, stimulus n is merely a signal for response n - 1.

However, stimulus n must at least be processed to some


123

extent in order to enable participants to proceed with the

task and perform a correct response when the subsequent

signal stimulus n + 1 is presented. One could argue that

this initial treatment of stimulus n, which occurs prior to

execution of response n - 1, already reaches the stage of

response selection, a stage that does not occur in the simple

RT task (e.g., Donders, 1868; Schubert, 1999; Szmalec

et al., 2005). According to the latter hypothesis, slower RTs

in the CRT-1 task than in a SRT task can be attributed to

the response selection for signal stimulus n, which could

occur before the response to the previous stimulus n - 1

was executed.

The current experiment was designed to test, which one

of both hypotheses (conflict or response selection hypoth-

esis) can account for the additional slowing in the CRT-1

task, compared to an SRT task. To this end, the stop-signal

paradigm was used. In the stop-signal paradigm, partici-

pants are required to perform a reaction time task while

occasionally a stop signal is presented, which instructs the

participants to withhold their response (Logan, 1994;

Logan & Cowan, 1984). In order to account for perfor-

mance in the stop-signal paradigm, Logan and Cowan

(1984) proposed a race between two stochastically inde-

pendent processes: a go process and a stop process.

According to the model, if the stop process is completed

before the go process, participants will successfully inhibit

their response (signal-inhibit). By contrast, if the go pro-

cess is terminated before the stop process, participants will

fail to inhibit the response (signal-respond). The model

allows estimating the covert latency of stopping, as indexed

by the stop signal reaction time (SSRT).

Several studies have demonstrated that SSRTs are

affected by distracting information in conflict tasks, like

Stroop, Simon and flanker tasks (e.g., Chambers et al.,

2007; Ridderinkhof, Band, & Logan, 1999; Verbruggen,

Liefooghe, & Vandierendonck, 2004, 2006; Verbruggen,

Liefooghe, Notebaert, & Vandierendonck, 2005). These

findings indicate that the suppression of distracter infor-

mation in conflict tasks and response inhibition rely on

common (inhibitory) mechanisms (Ridderinkhof et al.,

1999; Verbruggen et al., 2004). This conclusion is further

supported by Friedman and Miyake (2004), who observed

a strong correlation between measures of prepotent

response inhibition (e.g., Stop Signal, task, Antisaccade

task) and measures of resistance to distracter interference

(e.g., Eriksen flanker task). They attributed this overlap to

the common requirement of maintaining a task goal acti-

vated while more dominant but less appropriate responses

or distracting stimuli occur in the environment.

Furthermore, it has been demonstrated before that a

standard choice RT task yields longer SSRTs than a simple

RT task (e.g., Logan, Cowan, & Davis, 1984), suggesting

that response selection—which is involved in the choice

RT task but not in the simple RT task—competes for the

same resources as response inhibition. The interaction

between response selection and response inhibition was

made explicit in a recent neurophysiological study by

Szmalec et al. (2008). Szmalec and colleagues argued that

in response-selection tasks, the different choice stimuli

(e.g., high or low tone) are usually not completely unam-

biguous in terms of perceptual discriminability. They

showed that dependent on the degree of stimulus ambigu-

ity, also the erroneous response becomes activated in

choice RT tasks, and concluded that elementary variants of

response selection (such as in our standard choice RT task)

involve the requirement to suppress an erroneous response

tendency (see also Chambers et al., 2007).

The rationale of ‘‘Experiment 2’’ is the following:

because stop-signal inhibition on the one hand, and the

resolution of conflict and response selection on the other

hand, rely on a common cognitive mechanism, it should

be possible to further examine the role of conflict and

response selection in the CRT-1 task by comparing stop-

signal performance in the one-back choice RT task, the

choice RT task and the simple RT task. Based on the

assumption that conflict resolution and response selection

rely on the same mechanism as response inhibition in the

stop-signal paradigm, the following predictions can be

made. If the RT slowing in the one-back choice RT task,

compared to the simple RT task, is due to response

selection, then we expect that SSRTs for all four types of

one-back trials will be higher than SSRTs in the simple

RT task (since response selection does not occur in the

simple RT task). Moreover, based on the response-selec-

tion account, we anticipate that SSRTs of the one-back

choice RT task will be comparable to those of the choice

RT task, because also in the latter task, response selection

occurs in the interval between stimulus presentation and

response execution. By contrast, if the RT slowing in the

one-back choice RT task compared to the simple RT task

is due to conflict resolution, in line with our conflict

hypothesis, we predict that SSRTs in the one-back choice

RT task will augment selectively in the type(s) of trials

that involve conflict (S=R=, S=R=) and that SSRTs in

the non-conflict trials will be comparable to those of a

simple RT task.

Method


Eighteen students (ten females, mean age 21 years, range

19–24) of the Faculty of Psychology and Educational

Sciences at Ghent University were paid €10 for taking part

in the study. None of them had participated in ‘‘Experiment

1’’. We used a within-subject design with the factor Task


123

(SRT task, CRT task, CRT-1 task). The order of the tasks

was counterbalanced across subjects.


For the RT tasks, we used a white square and a white

circle, each 64 cm2, displayed on a black background of a

17 in. monitor. The stop signal was a 700 Hz auditory

signal, presented through a closed headphone (Sennheiser

HD 265-1) at 60 dB. The stop signal lasted 75 ms. In the

CRT task and CRT-1 task, participants were required to

hit the right key or the left key of a response box as fast

and accurately as possible when they saw the square or

the circle, respectively. They were instructed to rest their

index fingers of both hands on the response keys to avoid

that RTs become contaminated with movement time. In

the CRT-1 task, participants were required to delay their

response until the next stimulus occurred. In the SRT

task, they were asked always to respond with the same

hand (hand was counterbalanced across subjects), irre-

spective of the identity of the stimulus. The stimuli

remained on the display until the participant responded,

with a maximum response time of 1,500 ms. After the

response was executed, 1,500 ms elapsed until presenta-

tion of the next stimulus. In the CRT-1 task, we used

equal proportions of the four different trial types that

were presented in ‘‘Experiment 1’’. In the CRT task, we

used equal proportions of stimulus repetitions and alter-

nations. We also made sure that for all RT tasks, a

stimulus could not be repeated more than four times in a

row.

Participants were tested individually in a quiet room.

Prior to the counterbalanced conditions, they practiced the

CRT task and the CRT-1 task until they met a criterion of

20 consecutively correct (one-back) choice reactions. They

also practiced 20 SRT task trials. Participants performed

800 CRT-1 task trials, 400 CRT task trials and 200 SRT

task trials, each divided in blocks of 100 trials. These

numbers of trials were calculated so that each RT task

complied with the required proportion of stop signal trials

(see further). In total, each participant thus performed 1400

RT task trials. Because of the relatively high number of

trials, participants were asked to take a 10 min rest

between the tasks, during which they were not allowed to

perform intellectual activities. The entire procedure lasted

approximately 90 min.

On 25% of the trials, a stop signal was presented. The

SSD was initially set at 150 ms and continuously adjusted

according to a tracking procedure in order to obtain a

probability of stopping of 0.50 (Levitt, 1970). Each time a

participant responded to the stimulus in the presence of

a stop signal, SSD decreased by 25 ms. When inhibition

succeeded, SSD increased by 25 ms. The four different

trial types of the one-back choice RT task were tracked

independently (i.e., SSD adjustment for one trial type did

not influence SSD tracking of the other types). In the

choice RT task, stimulus repetitions and alternations were

tracked independently. In order to avoid waiting strate-

gies, participants were informed about the tracking

procedure and about the fact that stopping is expected to

succeed on approximately 50% of the stop signal trials.

At the end of each block, participants received feedback

about the percentage of suppressed trials and about the

mean reaction time for the trials without stop signal (i.e.,

no-signal trials). Per trial type, 50 stop signals were

presented. This means that 200 of the 800 CRT-1 task

trials were signal trials. In the CRT task, stimulus repe-

titions and alternations were tracked separately, implying

that 100 out of the 400 CRT task trials included a stop

signal. For the SRT task, in which stimulus alternatives

are not relevant and only one response alternative is

involved, a stop signal was presented on 50 of the 200

trials.

The SSRTs were estimated following the method pro-

posed by Logan and Cowan (1984). To this end, the RTs of

no-signal trials are first rank ordered. The fastest (left) part

of this RT distribution is assumed to correspond to the

distribution of RTs on signal trials for which participants

failed to suppress the response. The finishing time of the

stop process corresponds to the nth reaction time of the no-

signal trial RT distribution, where n is the result of mul-

tiplying the total number of no-signal trials by the

probability of responding when a signal is presented, given

a certain SSD. As the start of the inhibition process (i.e.,

mean SSD) and the finishing time are known, the SSRT can

then be estimated by subtracting the mean SSD from the

nth RT.

Results

One female participant was excluded from the sample

because she still adopted a waiting strategy, as evidenced

by a stopping probability of over 70%. The data of the

remaining participants (mean age 20 years, range 19–24)

were subjected to the same trimming procedure as in

‘‘Experiment 1’’. This resulted in a data reduction of 2.5%.

Performance was analyzed on this reduced dataset. RTs

were only calculated for the correct trials. In total, an

additional 3.2% of error and post-error trials were

excluded.

No-signal trials

Performance on the no-signal trials was analyzed by means

of a one-way ANOVA with the factor Task. The analysis

shows that the SRT task (M = 269.28, SD = 48.17) was


123

performed faster than the CRT-1 task (M = 321.38,

SD = 79.97), F(1,16) = 13.36, np2 = 0.45, p \ 0.01, and

that the CRT-1 task was in turn performed faster than the

CRT task (M = 378.32, SD = 63.13), F(1,16) = 15.01,

np2 = 0.48, p \ 0.01. With respect to the accuracy data, the

proportions of correct choice reactions in the CRT task

(M = 0.95, SD = 0.05) and in the CRT-1 task (M = 0.95,

SD = 0.03) were equal, F \ 1.

Next, we conducted a two-way ANOVA with the factors

Task (SRT, CRT, CRT-1) and Trial (S=R=, S=R=, S=R=,

S=R=). For the same reason as delineated in ‘‘Experiment

1’’, values for the different trial types in the SRT and CRT

tasks show the average RTs on the same stimulus transitions

as those used to determine the different types in the CRT-1

task. The mean RTs for the analysis are displayed in Fig. 4.

Stimuli following an error were discarded from the analysis.

The results show significant main effects of Task,

F(2,32) = 32.33, np2 = 0.50, p \ 0.001, and of Trial,

F(3,48) = 44.94, np2 = 0.48, p \ 0.001, and a significant

interaction, F(6,96) = 16.00, np2 = 0.14, p \ 0.001. Plan-

ned comparisons indicate that CRT-1 task performance

differed as a function of Trial, F(3,48) = 29.67, np2 = 0.38,

p \ 0.001. S=R= trials were performed equally fast as

S=R= and S=R= trials, all F’s\1. Only S=R= trials were

performed significantly slower than S=R=, F(1,16) = 44.85,

np2 = 0.74, p \ 0.001, S=R=, F(1,16) = 29.75, np

2 = 0.65,

p \ 0.001, and S=R= trials, F(1,16) = 38.62, np2 = 0.71,

p \ 0.001. CRT task performance also differed as a func-

tion of Trial, F(3,48) = 23.40, np2 = 0.33, p \ 0.001.

S=R= trials were performed faster than S=R= trials,

F(1,16) = 29.32, np2 = 0.65, p \ 0.001, and those of the

type S=R= were in turn performed faster than type S=R=,

F(1, 16) = 20.48, np2 = 0.56, p \ 0.001. Also S=R= trials

lead to faster RTs than S=R= trials, F(1,16) = 17.19,

np2 = 0.52, p \ 0.001. In the simple RT task, there was no

difference between the four different trial types, F \ 1.

Signal trials

Stopping data are represented in Table 1. We compared

stopping performance of seven conditions (SRT task, CRT

task repetitions, CRT task alternations, and the four types

of CRT-1 task trials) in one ANOVA. The SSRTs of CRT-

1 trial types S=R=, S=R= and S=R=, did not differ

reliably from the SSRT of the simple RT task, all p’s

[0.20. Only the SSRT of CRT-1 type S=R= was longer

than the SSRT of the simple RT task, F(1, 16) = 11.80,

np2 = 0.42, p \ 0.01. The SSRT of S=R= trials was

comparable to that of repetitions in the standard choice RT

task, F \ 1, and it did also not differ reliably from alter-

nations in the same task, F(1,16) = 2.09, np2 = 0.11,

p = 0.17.

Discussion

The purpose of the second experiment was to understand

why the one-back choice RT task yields slower RTs than

a simple RT task, bearing in mind that both RT tasks

allow preparation of the motor response prior to stimulus

presentation. The observation that the simple RT task is

performed faster than the one-back choice RT task sug-

gests that additional processing in the one-back task slows

down the prepared response. We proposed two possible

explanations for this additional slowing. The first is the

conflict hypothesis, which explains the RT differences

between simple and one-back reactions in terms of con-

flict (in the CRT-1 task) between the initial processing of

stimulus n and the final processing of stimulus n - 1. The

Fig. 4 Mean RTs in milliseconds, as a function of Task (SRT, CRT,

CRT-1) and Trial (S=R=, S=R=, S=R=, S=R=) in ‘‘Experiment

2’’. Vertical bars denote standard errors

Table 1 Observed signal-

respond RTs (SigRT), stop

signal reaction times (SSRT),

probabilities of responding

given a stop signal (Presp), and

stop signal delays (SSD) in

‘‘Experiment 2’’

SDs are in parentheses

Task SRT CRT CRT-1

Repet. Alter. S=R= S=R= S=R= S=R=

SigRT 237 (71) 362 (84) 395 (100) 295 (92) 303 (129) 365 (142) 297 (119)

SSRT 135 (28) 177 (30) 199 (47) 145 (23) 146 (38) 176 (37) 148 (33)

Presp (%) 48 (04) 50 (09) 51 (10) 48 (07) 47 (08) 51 (06) 47 (09)

SSD 131 (42) 190 (83) 194 (83) 153 (90) 156 (95) 195 (85) 150 (87)


123

second hypothesis proposes that response-selection

demands, which do not occur in a simple RT task, can

account for the slower performance in the one-back

choice RT task.

The results provide a clear answer to our research

question. First of all, we replicated the findings of

‘‘Experiment 1’’, regarding differential RT performance on

the various one-back task trials. Second, although the RTs

of one-back trial types S=R=, S=R= and S=R= were

approximately 25 ms slower than a simple reaction, the

SSRTs of these three trial types did not differ from the

SSRT in the simple RT task. This indicates that as far as

the central demands captured by the SSRT are concerned,

three out of the four possible types of one-back choice RT

task trials do not differ from a simple reaction. The SSRT

for trials of the type S=R=, however, was significantly

longer than those observed in the other one-back task trials

and in the simple RT task. Since the increased SSRT is

only apparent in one type of the one-back choice RT task

trials, it cannot be explained by response selection on the

trigger stimulus because in the latter case, SSRTs should be

comparable for all types of one-back trials. Based on this

evidence, the present finding that the trial types S=R=,

S=R= and S=R= yield SSRTs that are comparable to

those observed in a simple RT task, indicates that in the

one-back choice RT task, response selection does not occur

in the interval between stimulus n and response n - 1 (i.e.,

the stimulus–response interval), but probably after the

response to the previous stimulus has been executed.

Therefore, the increased SSRT of our ‘dissident’ trial type

S=R= reflects conflict resolution requirements, rather than

response selection demands.

The current findings are thus taken to indicate that

conflict, not response selection, is responsible for slower

performance in a CRT-1 than in a SRT task, conflict which

occurs selectively on trials of the type S=R=. As we

argued earlier, we believe that it originates from the fact

that S=R= trials involve a response incompatibility, i.e.,

trigger stimulus n requires another response than the one

that is prepared for stimulus n - 1. However, one thing

that remains to be clarified is why S=R= trials, which

also contain this response incompatibility, are not slowed

down to the same extent. In our view, this unpredicted

difference between S=R= and S=R= trials does not

necessarily contradict the conflict hypothesis. Trial types

S=R= and S=R= are the only ones to involve an

incompatibility between the responses required for stimuli

n and n - 1 (i.e., response alternation). One could notice

that S=R= is an incompatible trial that is preceded by a

compatible one (see Fig. 2). By contrast, S=R= is an

incompatible trial that is preceded by an incompatible one.

Conflict Monitoring Theory (CMT; Botvinick et al., 2001)

proposes that a compatibility effect is larger after a

compatible trial than after an incompatible trial (see also

Gratton, Coles, & Donchin, 1992) because in the first case,

the cognitive system is not yet adapted to the incompati-

bility. In the latter case however, experience with the first

incompatibility stimulates top-down adjustments, which

lead to a reduction of subsequent incompatibilities. CMT

would thus predict that conflict in S=R= trials is reduced

because it follows an incompatibility, by which the theory

is able to account for the difference with S=R= trials. In

our view, CMT thus offers a fairly comprehensive account

of the current findings. Interestingly, similar findings were

observed by Verbruggen et al. (2005), who showed that

stop signals interfered with the Simon effect, only when the

previous Simon trial was a compatible one (i.e., analogous

to the present type S=R=).

In conclusion, the purpose of ‘‘Experiment 2’’ was to

understand what distinguishes reaction time performance

on the one-back choice RT task and a simple RT task,

given that in theory both tasks could be performed equally

fast. The results show that the difference between both

tasks is mainly attributed to one type of trial which

involves stimulus alternation with response repetition (i.e.,

type S=R=) and which is an important source of conflict in

the one-back task.

Experiment 3

Beside the occurrence of conflict in the one-back choice

RT task, another finding that emerged from ‘‘Experiment

1’’ is that RTs in the CRT-1 task were influenced by the

length of the preceding ISI whereas no such effect was

observed in the CRT task. This may indicate that part of

the processing that distinguishes the CRT-1 task from

other RT tasks, occurs in the response-stimulus interval

(i.e., the interval between the response to stimulus n - 1

and the presentation of the next stimulus n + 1). But

there are other reasons to assume that part of the one-back

processing is delayed to the response–stimulus interval.

First, whereas response selection is required in the one-

back choice RT task (it remains a choice reaction task

after all), the SSRT data of ‘‘Experiment 2’’ demonstrated

that response selection apparently does not take place in

the stimulus–response interval. Hence, it should logically

take place in the response-stimulus interval. Second, it has

been demonstrated that although the one-back task is

faster, it causes more dual-task impairment than a stan-

dard choice RT task (Szmalec & Vandierendonck, 2007).

This suggests also that an important portion of the one-

back processing—probably the stage of response selec-

tion—occurs in the response–stimulus interval (RSI),

which is not covered by traditional reaction time

measurement.


123

The third experiment is aimed to test the latter

assumption by means of an RSI manipulation. The idea is

that by decreasing the RSI, participants will be forced to

transfer processing that is normally executed in the

response–stimulus interval, towards the stimulus–response

interval (SRI). This way, more processing would occur in

the SRI, and hence be captured by the reaction time mea-

sure. Accordingly, we designed an experiment in which

participants performed the one-back choice RT task and a

standard choice RT task under three different RSI condi-

tions (0, 250 and 1,500 ms). We hypothesized that the RT

advantage of the one-back choice RT task will decrease

with decreasing RSI. The reason why we choose 250 ms

instead of the arithmetical midpoint of 750 ms can be

explained as follows. Initially, we wanted to compare a

long with a short RSI, but since we were not sure that a RSI

of 0 would be feasible to the participants, we decided to

include another rather short RSI.

The most important question is whether or how this

RSI manipulation will affect the RT patterns that were

observed in the previous experiments. We proposed that

trial type S=R= is slower than the other types due to a

conflict that arises between the response that is prepared

for stimulus n - 1 and the processing of signal stimulus

n. It can be assumed that under short RSI conditions,

participants will need to process the signal stimulus even

further, and this will result in a greater amount of over-

lapping processes than under long RSI conditions. In the

extreme condition with RSI 0, the entire processing of the

signal stimulus (which thus also comprises response

selection) should logically take place in the SRI. We

assume that forcing participants to process the entire

signal stimulus prior to execution of the response to the

previous stimulus will increase the competition between

the (incompatible) prepared and newly selected responses

and therefore enhance the degree of conflict in trial type

S=R=. We thus predict that in this experiment, the RT

difference between the conflict and non-conflict trials

(i.e., type S=R= versus the other three types) will

increase with decreasing RSI.

Method


Eighteen adult students (nine females, mean age 20 years,

range 18–24) of the Faculty of Psychology and Educational

Sciences at Ghent University were paid €10 for partici-

pating in the study. None of them had participated in one of

the previous experiments. We used a 2 9 3 within-subject

design with the factors Task (CRT task, CRT-1 task) and

RSI (0, 250, 1,500 ms). The order of the six conditions was

counterbalanced based on a randomized Latin square.


Materials and procedure for the RT tasks were the same as

in ‘‘Experiment 2’’, unless specified otherwise. The choice

stimuli remained on the screen for 200 ms and participants

were allowed two seconds to respond. After their response,

an RSI of 0, 250 or 1,500 ms followed, manipulated in

three different blocks. Prior to the six counterbalanced

conditions, participants practiced each condition until they

met a criterion of 20 consecutively correct (one-back)

choice reactions. For each of the six conditions, partici-

pants performed 16 blocks of 70 trials. The entire

procedure lasted approximately 70 min.

Results

The data from one male participant were not included

because overall accuracy was lower than 60%. The data of

the 17 remaining participants were subjected to the same

trimming procedure as in ‘‘Experiment 1’’. This resulted in

a data reduction of 2.8%. Performance was analyzed on the

reduced data set and RTs were only calculated for the

correct choice reactions. In total, an additional 6.4% of

error and post-error trials were excluded.

Mean speed and accuracy, expressed as a function of

Task (CRT, CRT-1) and RSI (0, 250, 1,500 ms) are pre-

sented in Table 2. A two-way ANOVA with the same

factors shows significant main effects of Task,

F(1,16) = 22.99, np2 = 0.59, p \ 0.001, and RSI,


interaction, F(2,32) = 29.42, np2 = 0.48, p \ 0.001. Plan-

ned comparisons show that both the choice RT task,

F(2,32) = 19.04, np2 = 0.37, p \ 0.0001, and the CRT-1

task, F(2,32) = 39.94, np2 = 0.55, p \ 0.001, were per-

formed faster with increasing RSI and the significant

overall interaction thus indicates that the RT decrease was

more important in the one-back choice RT task. This

implies that whereas the one-back task was performed

faster than the standard choice RT task on the longest RSI,

F(1,16) = 3.82, np2 = 0.19, p = 0.06 (but the effect just

Table 2 Mean speed (upper half) and accuracy (lower half),

expressed as a function of Task (CRT, CRT-1) and RSI (0, 250,

1500 ms) in ‘‘Experiment 3’’

Response–stimulus interval (RSI)

0 ms 250 ms 1,500 ms

CRT task 398 (63) 348 (36) 328 (26)

CRT-1 task 727 (244) 542 (214) 290 (84)

CRT task 0.95 (0.03) 0.95 (0.03) 0.96 (0.03)

CRT-1 task 0.90 (0.05) 0.93 (0.05) 0.97 (0.02)

SDs are in parentheses


123

failed to be statistically reliable), the one-back task was

performed much slower than the CRT task on the shortest

RSI, F(1,16) = 34.35, np2 = 0.68, p \ 0.001 (see Fig. 5).

The accuracy data followed a similar pattern of results.

We further investigated how performance on the four

different trial types was affected by the different RSIs.

Since the trial types are defined by a succession of two

stimuli and two responses, we only included a target trial in

the analysis if the response on both stimuli was correct. We

performed a two-way ANOVA of the RTs in the CRT-1 task

with the factors RSI (0, 250, 1,500 ms) and Trial (S=R=,

S=R=, S=R=, S=R=). Again, Fig. 5 also represents

choice RT task performance on the same stimulus sequen-

ces as those used to define the one-back task trial types, as a

baseline. The results of the ANOVA show that in the one-

back choice RT task, there were significant main effects of

RSI, F(2,32) = 39.94, np2 = 0.55, p \ 0.001, and Trial,


interaction between the two, F(6,96) = 2.55, np2 = 0.03,

p \ 0.05. Planned comparisons show that the difference

between trial type S=R= and the other three types increased

with decreasing RSI, F(2,32) = 3.05, np2 = 0.09, p = 0.06

(for the interaction between the factor RSI and the vector

contrasting S=R= with the other three types).

Discussion

‘‘Experiment 3’’ was designed to test the hypothesis that

during the RSI in the one-back choice RT task, there is

some additional processing that does not occur in the RSI

of the standard choice RT task. This hypothesis is sup-

ported by the data. When we decreased the RSI of the

choice RT task, RTs increased from approximately 330 to

400 ms. However, the same manipulation in the one-back

task shows that RTs increase from about 290 ms up to

730 ms, which is a slowing of over 400 ms. In our view,

this suggests that traditional RT procedures (i.e., the ones

that do include RSIs) do not grasp the cognitive demands

of the one-back choice RT task as well as of other RT tasks

because it misses the CRT-1 task processes that are

occurring in the response–stimulus interval.

In the RSI = 1,500 ms condition, we replicated the

findings of the previous experiments regarding the RT

differences between the four different trial types in the one-

back choice RT task. It is interesting to see that the typical

pattern of results, which shows that trials of the type S=R=

are much slower than the others, varies as a function of the

RSI conditions (see Fig. 5). More precisely, it indicates

that the processing which is transferred from the response–

stimulus interval to the stimulus–response interval fol-

lowing the RSI manipulation (most likely response

selection), seems to augment the conflict that causes the

delay in trial type S=R=. We will come back to this

finding in the ‘‘General discussion’’.

General discussion

The purpose of the present study was to demonstrate that

the one-back choice RT (CRT-1) task, which reflects a

Fig. 5 Mean RTs in

milliseconds, as a function of

Task (CRT, CRT-1), RSI (0,

250, 1,500 ms) and Trial

(S=R=, S=R=, S=R=,

S=R=) in ‘‘Experiment 3’’.

Vertical bars denote standard

errors


123

combination between a choice RT (CRT) task and the 1-

back condition of a traditional n-back task, could be used to

further explore the relation between updating and conflict

resolution. We presented a detailed analysis of the CRT-1

task in which we distinguished four different kinds of one-

back trials, based on the repetition and alternation of

stimulus and response features throughout the task. In

‘‘Experiment 1’’, we observed that one out of the four trial

types was performed much slower than the others (i.e.,

S=R=). This pattern of results proved to be robust since

our series of three experiments indicated that it remains

unaffected under variations of stimulus modality and other

task parameters like RSI. We also addressed two specific

questions regarding the cognitive processes involved in the

one-back choice RT task.

Firstly, we aimed to understand why RTs are slower in

the one-back choice RT task than in the simple RT task,

even though both tasks allow advance preparation of the

motor response. The SSRT data of ‘‘Experiment 2’’ dem-

onstrated that the additional slowing of the CRT-1 task

compared to an SRT task is mainly due to a conflict

between the early processing of trigger stimulus n and the

final processing of one-back stimulus n - 1. Both RT and

SSRT results showed that trial type S=R= is almost

entirely responsible for this difference between the one-

back task and the simple RT task, a finding which was

accounted for by Conflict Monitoring Theory.

Secondly, it was our goal to find out whether part of the

cognitive processing in the one-back choice RT task is

delayed to the response–stimulus interval and is thus not

captured by the RT. The reason for addressing this question

is inspired by the results of the ISI manipulation in

‘‘Experiment 1’’ and by Szmalec and Vandierendonck’s

(2007) observation that in a dual-task setting, the one-back

task yielded faster RTs than a choice RT task whereas the

one-back task obviously caused more dual-task impairment

than the choice RT task. In ‘‘Experiment 3’’, we manipu-

lated the response–stimulus interval in order to coerce

participants into shifting the processing that occurs in the

response–stimulus interval to the stimulus–response inter-

val. We assumed that this manipulation would affect the

stage of response selection in particular because the SSRT

data of ‘‘Experiment 2’’ already suggested that in the one-

back choice RT task, the stage of response selection does

not seem to occur in the stimulus–response interval (thus

probably in the response–stimulus interval). Therefore, we

anticipated that shifting response selection from RSI to SRI

would increase the overlap between the processing of

successive response representations in the SRI of the CRT-

1 task and would hence augment the amount of conflict in

the task. We observed that by decreasing the response–

stimulus interval, the RTs in the one-back choice RT task

increased to a level that greatly exceeded RTs in the

standard choice RT task and most importantly, the differ-

ence between the conflict trial type (S=R=) and the other

three types in the CRT-1 task was amplified. This confirms

our hypothesis that part of the processing involved in the

one-back choice RT task, which is probably response

selection, occurs after the response was executed. All

processing taken into account (i.e., in the SRI and RSI), the

CRT-1 updating task is, compared to a standard choice RT

task, essentially a temporal shift of processes which addi-

tionally calls on processes that can, according to the current

findings, be termed as conflict resolution.

The finding that decreasing the RSI amplifies the dif-

ference between trial type S=R= and the other types

deserves some more attention. In our view, it should be

understood as an indication that in the CRT-1 task, par-

ticipants prefer to delay the processing of the signal

stimulus (i.e., probably the stage of response selection)

until the processing of the previous one is terminated (i.e.,

until the prepared response is executed) to minimize the

conflict between the prepared response and the response

required for the new stimulus. It would be interesting to

know whether a similar RSI manipulation would also affect

the involvement of conflict in other updating tasks (e.g., the

n-back task), in order to find out whether in general, the

tendency to reduce the overlap in the processing of sub-

sequent stimuli or materials (for example by transferring

processing stages that are prone to conflict from the SRI to

the RSI) is a way to limit the degree of conflict in updating.

It is important to underline that all four types of CRT-1

task trials require that response representations are main-

tained and updated, but that conflict occurs only in a subset

of those trials, i.e., the ones that involve incompatible

response representations. We demonstrated that differences

in control requirements between the one-back choice RT

task and a simple RT task (see e.g., stop signal reaction

times in ‘‘Experiment 2’’) are in fact restricted precisely to

those response incompatible trials and thus do not just

apply to all trials that involve updating. This means that

differences in executive control demands between the one-

back choice RT task and other RT tasks (e.g., simple RT

task) that were initially related to the updating require-

ments of the task can on the basis of the current results be

attributed to the occurrence of conflict. One important

future step will be to see whether this conclusion also

applies to other updating tasks like n-back, by testing the

hypothesis that updating in the n-back task demands

executive control resources particularly when competition

occurs between representations in memory (i.e., on n-back

lure trials, as described in the introduction).

In essence, we propose that the requirement to delay a

choice reaction until the next stimulus is quite a demanding

operation in terms of control requirements. First, partici-

pants need to execute a prepared response on presentation


123

of a new target stimulus, while trying to overcome conflict

between the maintained representation of the old response

and the developing representation of the new stimulus (for

a similar idea, see Miller & Cohen, 2001). If conflict

occurs, it must be resolved before the response to the

previous stimulus can be executed. Then, participants

select the response for the current stimulus and refresh the

response representation that is kept active throughout the

task with the response information that will be necessary to

make a correct choice reaction on the subsequent trial.

One may remark that our claim that the one-back choice

RT task is a demanding task stands somewhat at odds with

the observation that the one-back task yields faster reaction

times than less demanding tasks, such as a standard choice

RT task. However, we have shown that an important part of

the one-back processing occurs after the response has been

executed and is by consequence not reflected in the reac-

tion time. But, as demonstrated with the response–stimulus

interval manipulation in ‘‘Experiment 3’’, when all the

CRT-1 task processing can be captured by the reaction time

measurement, it becomes clear that the one-back choice RT

task is performed much slower than a standard choice RT

task. Therefore, we would like to stress that our description

of the one-back choice RT task as a demanding task also

takes into account those stages of processing that occur in

the response–stimulus interval.

At first sight, the conclusion that the one-back choice

RT task requires conflict resolution does not rule out the

possibility that the process of updating is still to some

extent co-responsible for the executive nature of the task.

But then, one has to clearly conceptualize what the process

of updating comprises that the process of resolving conflict

between memory representations does not. This brings us

to the main theoretical conclusion we would like to derive

from the current findings. Our results suggest that, besides

the process of response selection, which appears to play a

role in any executive task (Hegarty et al., 2000), conflict

resolution is an important contributor to the executive

nature of updating tasks. We have demonstrated that here

with the CRT-1 task, and it has recently also proven to be

true for the n-back task (e.g., Jonides & Nee, 2006; Kane

et al., 2007).

This conflict idea is best understood as monitoring

simultaneously developing and extinguishing memory

activations (be it a representation of verbal information in

the n-back task or response representations in the CRT-1

task), and if conflict between those representations occurs,

resolving it to prevent behavioural errors. Therefore, the

act of updating itself could eventually be conceptualized as

a kind of conflict resolution at the level of memory rep-

resentations. In other words, we argue that an updating task

may in essence be carried out by means of a conflict

resolution process, which manages the levels of

simultaneously varying activations. In fact, Jonides et al.

(1997) already pointed in that direction when they argued

that successful performance in the n-back task depends,

among other things, on ‘‘inhibition processes needed to

dampen the trace of the oldest letter in memory so it can be

replaced by the newest letter in the series’’ (p. 471). The

conclusion that the one-back choice RT task involves

conflict resolution forms an important extension of the

previous findings with the n-back task (e.g., Kane et al.,

2007). It suggests that the relation between updating and

conflict resolution is not task-dependent. More precisely,

whereas the n-back task involves competition between

relevant and irrelevant memory contents, the CRT-1

method produces a similar competition between response

representations. This indicates that conflict resolution

requirements are probably not restricted to updating rep-

resentations achieved at the level of working memory slave

systems, but appear rather to be a characteristic of updating

in general.

Another point that deserves some attention has to do with

the ecological validity of the one-back choice RT task, i.e.,

the extent to which the CRT-1 task reflects the kind of

memory updating that is required in daily life. In that

respect, it is important to stress that in daily life, not each

refreshment of a memory content requires an active form of

updating as is described here. One could perfectly argue that

people’s memory contents are continuously updated, while

talking or reading the newspaper for example. According to

our view, these activities reflect effortless, more automated

forms of updating, and are thus not the type of performance

that is predicted by laboratory measures of updating like n-

back or CRT-1 which reflect a more controlled adaptation

of the activation levels in memory. Controlled updating is

required in order to prevent that the presentation of new

task-relevant information affects maintained memory con-

tents (or vice versa) and hence may lead towards suboptimal

performance (e.g., through interference or forgetting). A

real life example of this controlled updating comes from

mental arithmetic, like when a complex addition problem is

resolved by calculating and maintaining intermediate

results, the results of which are in turn systematically

refreshed on the basis of subsequent intermediate calcula-

tions. In such a situation, the actual memory contents need

to be accommodated on the basis of newly incoming

information and this implies a temporary overlap between

several task-relevant activations. It is precisely this tem-

poral overlap of task-relevant activations that is reflected in

the continuous character of the popular laboratory updating

task. The one-back choice RT task or the n-back task for

example, are continuous in the sense that task-relevant

memory contents need to be maintained across successive

trials, and each time a new stimulus is presented, the

maintained information is accommodated based on the new


123

input. In our view, this requires careful and controlled

adaptation of the several task-relevant activations in

memory or in other words, updating in the sense that it is

described in the present study.

In summary, we propose that the view that updating is a

task demand which is fulfilled by means of conflict reso-

lution processes, may be a plausible alternative to the view

that updating is a separable, autonomous executive

function, besides other often postulated functions like

inhibition, shifting or dual-task coordination. Of course, the

relation between updating and conflict needs to be more

extensively investigated before preferring one alternative

above the other, and as we demonstrated in the current

study, the one-back choice RT task offers a potentially

useful paradigm for those future efforts. The question

whether updating is primarily a task demand or an execu-

tive process is of much theoretical relevance because it

leaves open the possibility that the variety of well-docu-

mented executive tasks that are proposed to tap on one

particular executive function, do not all necessarily involve

different control processes. For example, if conflict reso-

lution turns out to be a determinant in the executive nature

of updating tasks, just as it is in a number of inhibition

tasks such as the Stroop task, one will have to consider the

alternative possibility that Stroop and n-back tasks may be

two different tasks that rely on a common control mecha-

nism or control loop. This could also explain why earlier

studies found commonalities underlying the executive

functions of inhibition and updating (e.g., Miyake et al.,

2000). In this context, it is also necessary to refer to the

dual mechanisms of cognitive control (DMC) model,

which was recently introduced by Braver et al. (2007).

Partly inspired on Conflict Monitoring Theory (Botvinick

et al., 2001), these authors advanced a model which pro-

poses to account for the entire notion of executive control

in working memory, by relying on solely two control

mechanisms, one specialized in goal maintenance (proac-

tive control) an the other in conflict resolution (reactive

control). According to us, this model is in line with the

ideas derived from the current findings, in that it also

suggests that, while trying to fractionate or decompose a

single control entity, theorists should be aware that dif-

ferent executive tasks (often defined at face validity) do not

necessarily operationalize autonomous or separable exec-

utive functions.

References

Baddeley, A. (1986). Working memory. Oxford: Oxford University

Press, Clarendon Press.

Baddeley, A. (1996). Exploring the central executive. QuarterlyJournal of Experimental Psychology, 49, 5–28.

Baddeley, A., & Hitch, G. (1974). Working memory, In G. A. Bower

(Ed.), The psychology of learning and motivation: Advances inresearch and theory (Vol. 8, pp. 47–89). New York: Academic

Press.

Botvinick, M., Braver, T., Barch, D., Carter, C., & Cohen, J. (2001).

Conflict monitoring and cognitive control. PsychologicalReview, 108, 625–652.

Braver, T., Gray, J., & Burgess, G. (2007). Explaining the many

varieties of working memory variation: Dual mechanisms of

cognitive control. In Conway, A., Jarrold, C., Kane, M., Miyake,

A., & Towse, J. (Eds.) Variation in working memory. Oxford:

University Press.

Bunting, M., Cowan, N., & Saults, J. S. (2006). How does running

memory span work? The Quarterly Journal of ExperimentalPsychology, 59, 1691–1700.

Chambers, C. D., Bellgrove, M. A., Gould, I. C., English, T.,

Garavan, H., McNaught, E., et al. (2007). Dissociable mecha-

nisms of cognitive control in prefrontal and premotor cortex.

Journal of Neurophysiology, 98, 3638–3647.

Deschuyteneer, M., Vandierendonck, A., & Muyllaert, I. (2006). Does

solution of mental arithmetic problems such as 2 + 6 and 3 9 8

rely on the process of ‘‘memory updating’’? ExperimentalPsychology, 53, 298–208.

Donders, F. C. (1868). La vitesse des actes psychiques [On the speed

of mental processes]. Archives Neerlandaises, 3, 269–317.

Engle, R. W., Kane, M. J., & Tuholski, S. W. (1999). Individual

differences in working memory capacity, what they tell us about

controlled attention, general fluid intelligence and functions of

the prefrontal cortex. In Miyake, A., & Shah, P. (Eds.) Models ofworking memory: Mechanisms of active maintenance andexecutive control. London:Cambridge Press.

Friedman, N., & Miyake, A. (2004). The relations among inhibition

and interference control functions: A latent variable analysis.

Journal of Experimental Psychology: General, 133, 101–135.

Gratton, G., Coles, M., & Donchin, E. (1992). Optimizing the use of

information: Strategic control of activation and responses.

Journal of Experimental Psychology: General, 4, 480–506.

Hegarty, M., Shah, P., & Miyake, A. (2000). Constraints on using the

dual-task methodology to specify the degree of central executive

involvement in cognitive tasks. Memory and Cognition, 28(3),

376–385.

Hommel, B. (1998). Event files: Evidence for automatic integration of

stimulus–response episodes. Visual Cognition, 5, 183–216.

Jonides, J., Schumacher, E., Smith, E., Lauber, E., Awh, E.,

Minoshima, S., & Koeppe, R. (1997). Verbal working memory

load affects regional brain activation as measured by PET.

Journal of Cognitive Neuroscience, 9, 462–475.

Jonides, J., & Nee, E. (2006). Brain mechanisms of proactive

interference in working memory. Neuroscience, 139, 181–193.

Kane, M., Conway, A., Miura, T., & Colflesh, G. (2007). Working

memory, attention control and the n-back task: A question of

construct validity. Journal of Experimental Psychology: Learn-ing, Memory, and Cognition, 33, 615–622.

Kemps, E., Szmalec, A., Vandierendonck, A., & Crevits, L. (2005).

Visuo-spatial processing in Parkinson’s disease: Evidence for

diminished visuo-spatial sketch pad and central executive

resources. Parkinsonism & Related Disorders, 11, 181–186.

Levitt, H. (1970). Transformed up–down methods in psychoacoustics.

The Journal of the Acoustical Society of America, 49, 467–477.

Logan, G. (1994). Spatial attention and the apprehension of spatial

relations. Journal of Experimental Psychology: Human Percep-tion and Performance, 20, 1015–1036.

Logan, G., & Cowan, W. (1984). On the ability to inhibit thought and

action: A theory of an act of control. Psychological Review, 91,

295–327.


123

Logan, G., Cowan, W., & Davis, K. (1984). On the ability to inhibit

responses in simple and choice reaction time tasks: A model and

a method. Journal of Experimental Psychology: Human Per-ception and Performance, 10, 276–291.

Miller, E., & Cohen, J. (2001). An integrative theory of prefrontal

cortex function. Annual Review of Neuroscience, 24, 167–202.

Miyake, A., Friedman, N., Emerson, M., Witzki, A., Howerter, A., &

Wager, T. (2000). The unity and diversity of executive functions

and their contributions to complex ‘‘frontal lobe’’ tasks: A latent

variable analysis. Cognitive Psychology, 41, 49–100.

Morris, N., & Jones, D. (1990). Memory updating in working

memory: The role of the central executive. British Journal ofPsychology, 81, 111–121.

Rabbitt, P. (1968). Three kinds of error-signalling responses in a

serial choice task. Quarterly Journal of Experimental Psychol-ogy, 20, 179–188.

Rabbitt, P. (1997). Methodology of frontal and executive function.

Hove:Psychology Press.

Ridderinkhof, K., Band, G., & Logan, G. (1999). A study of adaptive

behavior: Effects of age and irrelevant information on the ability

to inhibit one’s actions. Acta Psychologica, 101, 315–337.

Rowe, J. B., Toni, I., Josephs, O., Frackowiak, R. S. J., &

Passingham, R. E. (2000). The prefrontal cortex: Response

selection or maintenance within working memory? Science, 288,

1656–1660.

Schubert, T. (1999). Processing differences between simple and

choice reactions affect bottleneck localization in overlapping

tasks. Journal of Experimental Psychology: Human Perceptionand Performance, 25, 408–425.

Smith, E., & Jonides, J. (1997). Working memory: A view from

neuroimaging. Cognitive Psychology, 33, 5–42.

Smith, E., & Jonides, J. (1999). Storage and executive processes in

the frontal lobes. Science, 283, 1657–1661.

Smyth, M., & Scholey, K. (1994). Interference in immediate spatial

memory. Memory & Cognition, 22, 1–13.

Stevens, M., Lammertyn, J., Verbruggen, F., & Vandierendonck, A.

(2006). Tscope: A C library for programming cognitive exper-

iments on the MS Windows platform. Behavior ResearchMethods, 38, 280–286.

Szmalec, A., & Vandierendonck, A. (2007). Estimating the executive

demands of a one-back choice RT task by means of the selective

interference paradigm. Quarterly Journal of Experimental Psy-chology, 60, 1116–1139.

Szmalec, A., Vandierendonck, A., & Kemps, E. (2005). Response

selection involves executive control: Evidence from the selective

interference paradigm. Memory and Cognition, 33, 531–541.

Szmalec, A., Verbruggen, F., Vandierendonck, A., De Baene, W.,

Verguts, T., & Notebaert, W. (2008). Stimulus ambiguity elicits

response conflict. Neuroscience Letters, 435, 158–162.

Van der Linden, M., Collette, F., Salmon, E., Delfiore, G., Degueldre,

C., Luxen, A., et al. (1999). The neural correlates of updating

information in verbal working memory. Memory, 7, 549–560.

Verbruggen, F., Liefooghe, B., & Vandierendonck, A. (2004). The

interaction between stop signal inhibition and distractor inter-

ference in the flanker and Stroop task. Acta Psychologica, 116,

21–37.

Verbruggen, F., Liefooghe, B., Notebaert, W., & Vandierendonck, A.

(2005). The effects of stimulus–stimulus compatibility and

stimulus–response compatibility on response inhibition. ActaPsychologica, 120, 307–326.

Verbruggen, F., Liefooghe, B., & Vandierendonck, A. (2006). The

effect of interference in the early processing stages on response

inhibition in the stop signal task. Quarterly Journal of Exper-imental Psychology, 59, 190–205.


123

Investigating the role of conflict resolution in memory updating by means of the one-back choice RT task

Documents