Acta Psychologica 151 (2014) 184–196

Contents lists available at ScienceDirect

Acta Psychologica
journal homepage: www.elsevier.com/ locate/actpsy

Components of task switching: A closer look at task switching and

cue switching
Florian Schmitz a,⁎, Andreas Voss b
a
Institute of Psychology, Ulm University, Ulm, Germany
b
Institute of Psychology, University of Heidelberg, Heidelberg, Germany

a r t i c l e i n f o a b s t r a c t

Article history: Research using the diffusion model to decompose task-switching effects has contributed to a better understand-
Received 8 November 2013 ing of the processes underlying the observed effect in the explicit task cueing paradigm: Previous findings could
Received in revised form 13 June 2014 be reconciled with multiple component models of task switching or with an account on compound-cue retrieval/
Accepted 16 June 2014
repetition priming. In the present study, we used two cues for each task in order to decompose task-switch and
Available online 5 July 2014
cue-switch effects. Response time data support previous findings that comparable parts of the switching effect
PsychINFO classification:
can be attributed to cue-switching and task-switching. A diffusion model analysis of the data confirmed that
2320 Sensory Perception non-decision time is increased and drift rates are decreased in unpredicted task-switches. Importantly, it was
2330 Motor Processes shown that non-decision time was selectively increased in task-switching trials but not in cue-switching trials.
2340 Cognitive Processes Results of the present study specifically support the notion of additional processes in task-switches and can be
2240 Statistics & Mathematics reconciled with broader multiple component accounts.
© 2014 The Authors. Published by Elsevier B.V. This is an open access article under the CC BY-NC-ND license
Keywords: (http://creativecommons.org/licenses/by-nc-nd/3.0/).
Task switching
Cue switching
Reconfiguration
Inertia
Repetition priming
Diffusion model

1. Introduction must be activated in working memory before stimulus classification

People usually find it more difficult to shift between mental tasks
rather than to repeat the same task. In laboratory experiments, task-
switches are usually accompanied by increased response times (RTs)
and decreased accuracy compared to task-repetitions. This difference
may be accounted for by task-switching costs and/or by benefits of rep-
etition priming. We will briefly address both possibilities below. More
extensive reviews of task switching accounts can be found elsewhere
(e.g., see Kiesel et al., 2010; Vandierendonck, Liefooghe, & Verbruggen,
2010, and Schmitz & Voss, 2012, for a discussion on how theoretical
accounts could be mapped onto parameters of the diffusion model).
1.1. Task-switching costs
According to models of task switching costs, task-set reconfiguration
(TSR) and task-set inertia (TSI) are assumed to contribute to task-
switching effects. The classical task-set reconfiguration account (Rogers
& Monsell, 1995) assumes that in task-switch trials a new task set
⁎ Corresponding author at: Institute of Psychology, Ulm University, 89081 Ulm,
Germany.
E-mail address: florian.schmitz@uni-ulm.de (F. Schmitz).
can proceed. Hence, switching costs are interpreted as the time it
takes to prepare the new task-set, whereas increased error rates
are explained by failures to accomplish the task-set reconfiguration
in time.
Theories postulating an additional processing phase (or multiple
phases) for task-switch trials have been referred to as additional process
accounts (De Jong, 2000; Rogers & Monsell, 1995; Rubinstein, Meyer, &
Evans, 2001). According to this perspective, preparing for the next task
is considered a top-down mechanism exerted by (executive)
control functions (Baddeley, Chincotta, & Adlam, 2001; Miyake,
Friedman, Emerson, Witzki, & Howerter, 2000), and a number of
ideas or metaphors are postulated on how task-set preparation
could proceed. These include re-programming the system (Ruthruff,
Remington, & Johnston, 2001), top-down biasing the relevant task-
demand units (Gilbert & Shallice, 2002), retrieving the new task set
from long-term memory and loading it into working memory
(Goschke, 2000; Mayr & Kliegl, 2000, 2003) or loading it into the region
of direct access of procedural working memory (Risse & Oberauer,
2010). It was argued that the well-known cue–stimulus interval (CSI)
effect denoting reduced switching costs when the CSI is long (e.g.,
Meiran, 1996) supports the view of task-set reconfiguration prior to
stimulus onset.

http://dx.doi.org/10.1016/j.actpsy.2014.06.009
0001-6918/© 2014 The Authors. Published by Elsevier B.V. This is an open access article under the CC BY-NC-ND license (http://creativecommons.org/licenses/by-nc-nd/3.0/).
 F. Schmitz, A. Voss / Acta Psychologica 151 (2014) 184–196
Differently from this view, the classical account on task-set inertia
(TSI; Allport, Styles, & Hsieh, 1994; Allport & Wylie, 2000) assumes
that switching costs result from proactive interference from previous
processing of a competing task. In support of the TSI account, it was
shown that the adverse effects of proactive interference in the explicit
cueing paradigm are largest on trials directly succeeding task switches,
but that they are not confined to these switch trials (e.g., Monsell,
2003). Further support of the TSI account stems from the observation
of switch-cost asymmetries, i.e. larger costs when switching away
from a less dominant to a more dominant task compared with the
other direction (Allport & Wylie, 2000; Allport et al., 1994; Yeung
& Monsell, 2003). Additionally, n-2 repetition costs observed in
paradigms with 3 tasks support the view of task set inertia: Larger
switch costs in the sequence ABA as compared with the sequence
CBA have been accounted for by inertia of “backward inhibition”
(Koch, Gade, Schuch, & Philipp, 2010; Mayr & Keele, 2000). Finally,
neuroimaging data show that switching costs correlate with in-
creases in task-irrelevant neural activity after a switch (Yeung,
Nystrom, Aronson, & Cohen, 2006), supporting the notion of inter-
ference from recently performed tasks as a major determinant of
task-switching costs. To conclude, task-set reconfiguration rather
emphasizes top-down control, whereas task-set inertia points to
bottom-up processes. However, more reactive forms of control
may come into play as well when response conflicts are detected.
Both views have been supported in the literature, and were inte-
grate into multiple-component models of task switching (e.g., Mayr
& Kliegl, 2003; Ruthruff et al., 2001). These models typically specify
several phases of a task switch: In an earlier phase, task-set prepa-
ration is considered to take place, whereas in a later phase, stimulus
information is used to select the response along the constraints
(e.g., S–R links) set in the first phase. As the level of activation of
S–R links is also affected by inertia effects, the latter will also
contribute to the efficiency of response selection in the later phase.
1.2. Compound-cue retrieval/repetition priming
An elegant alternative account of the observed difference in task per-
formance between task-switch and task-repetition trials in the absence
of executive control was offered by Logan and colleagues (Logan &
Bundesen, 2003; Schneider & Logan, 2005, 2007). According to this
approach, participants use a compound-cue retrieval strategy in the ex-
plicit cueing paradigm. Specifically, it is assumed that both task-cue and
target stimulus need to be encoded to form a compound stimulus which
uniquely determines the correct response. Importantly, cue encoding is
predicted to follow a horse-race process which is terminated as soon as
a representation of the cue is found. In all trials there should be a long-
term memory representation, but in cue-repetition trials there should
be as well a short-term memory representation which will easily win
the race, resulting in short encoding time. Hence, cue-encoding benefits
in task-repetition trials are made responsible for the observed dif-
ferences in response times between task-repeat and task-switch trials.
After encoding of cue and target stimulus, the compound cue is
formed and response selection takes place, which is assumed to follow
a random walk process with only a few steps, in line with retrieval the-
ories of binary decision tasks (cf. Nosofsky & Palmeri, 1997). As all pro-
cesses are stimulus driven, this account does not draw on concepts such
as task sets, task switches, and task-set reconfiguration. Hence, it does
not require an endogenous act of control.
Effects of task-switching and cue switching cannot be teased
apart with the classical explicit cueing paradigm, because both
cue-switches sand task-switches are perfectly confounded. Hence,
the repetition priming account was tested using a task-switching
paradigm with a 2:1 cue-to-task mapping (Logan & Bundesen,
2003; Mayr & Kliegl, 2003): Employing two different cues for each
task allows one to dissociate effects of cue switching from those
of task switching. Cue-switching effects denote the difference in
185
task performance between task-repetition trials with either a cue
switch or a cue repeat. Pure task-switch effects denote the differ-
ence between cue-switch trials with either a task-switch or a task
repeat. Evidence obtained with the 2:1 mapping paradigm has sup-
ported the view that cue-switches contribute substantially to the
task-switch effect (Logan & Bundesen, 2003), or that both, cue-
switches and task-switches, contribute to the task-switch effect
(Arrington, Logan, & Schneider, 2007; Logan & Bundesen, 2004;
Mayr & Kliegl, 2003; Schneider & Logan, 2005). Some of the diver-
gent findings concerning the relative magnitudes of the task-
switch and cue-switch effects may be attributed to the conditional
probability of a task switch given a cue switch (Mayr, 2006), as will
be discussed later.
However, some findings challenge the compound-cue retrieval/rep-
etition priming account, and seem to support the existence of task sets
instead. For instance, there are n-2 repetition costs in paradigms with
a 2:1 cue to task mapping irrespective of whether the cue switches or
not (Altmann, 2007; Gade & Koch, 2008), suggesting inhibition on the
level of task-sets. Similarly, task-performance in implicit sequence
learning with a 2:1 cue to task mapping was found not to be affected
by cue switches or cue repetitions (Gotler, Meiran, & Tzelgov, 2003).
Finally, repetition priming cannot easily account for comparably fast
responses in trials with complete switches, i.e. when neither the task,
nor the stimulus, nor the response repeats from one trial to the next.
Conversely, increased response times in trials in which some but not
all features repeat (“partial repetition costs”, e.g., Hommel, 2004) have
been explained by the bottom-up activation of competing task sets or
according responses.
1.3. Further evidence of separable cue-switch and task-switch effects
The above reviewed studies with a 2:1 cue-to-task mapping par-
adigm may help estimate the quantitative contribution of cue-
switch and task-switch processes to the overall switching effect.
Other studies offered more direct evidence that different processes
may contribute to both effects (see Jost, De Baene, Koch, & Brass,
2013, for an excellent review). For instance, it was shown that
only cue-switch costs were affected by CSI and practice, whereas
the task-switch effect was shown to depend on response priming
and task-set inhibition (Mayr & Kliegl, 2003). Similarly, a response
cue interval (RCI) variation was found to affect the cue-switch ef-
fect (suggesting a decay of the cue-representation), but not the
task-switch effect (Horoufchin, Philipp, & Koch, 2011).
Additionally, physiological data support that cue switches and
task switches are governed by distinct processes, separated both
by time and the neural basis. In an EEG study using the 2:1 mapping
paradigm, two dissociable ERP components were found, one associ-
ated with cue-switches at around 300 ms after cue onset, the other
one associated with task switches 400 ms after onset of the target
stimulus with a different topography (Jost, Mayr, & Rösler, 2008).
Additionally, a number of neuro-imaging studies revealed that
cue-switches and task-switches recruit different neural networks
(Brass & Cramon, 2004; De Baene, Kuhn, & Brass, 2012; Wylie,
Javitt, & Foxe, 2006). Hence, cue (-switch) processes and task-
switch processes seem to be separable. In fact, some sort of cue-
processing may be required in all trials (although this may take lon-
ger in case of cue switches), whereas task-switch specific processes
may be confined to true switches of the task set.
Preparation also may comprise several phases. For instance,
Mayr and Kliegl (2003) suggest that in a first phase (corresponding
with the cue-switch effect), the task set is retrieved from long-term
memory. In a second phase (corresponding with the task-switch ef-
fect), it is installed in a task-appropriate attentional configuration
(see also Risse & Oberauer, 2010). This may correspond, e.g., with
the biasing of relevant task-demand units (Gilbert & Shallice,
2002; see also Ruthruff et al., 2001).
186 F. Schmitz, A. Voss / Acta Psychologica 151 (2014) 184–196
Fig. 1. The diffusion model decomposes a decision process into non-decision time and the
actual decision process. Non-decision time captures processes and activity before and after
the actual decision phase. In turn, the actual decision process (displayed right of the
rounded brace) is characterized by a continuous sampling of information. A decision
counter, originating from starting point z, fluctuates over time (horizontal axes) as a func-
tion of systematic stimulus information and random noise (see gray sample path). When it
hits the lower or upper response thresholds (at 0 or a, respectively), the according
response is elicited. The mean slope of the counter across trials denotes the drift rate (v).
1.4. The diffusion model
The diffusion model for response time data (Ratcliff, 1978;
Ratcliff & Rouder, 1998; Voss, Rothermund, & Voss, 2004) can be
used to decompose a binary decision process into a set of psycho-
logically meaningful parameters. A graphical representation of the
model is given in Fig. 1. The basic assumption of the diffusion
model is a continuous accumulation of information over time.
This information accumulation process is assumed to follow a
stochastic diffusion process, which is a process that is driven by
systematic stimulus based information and by random noise. In
Fig. 1, an example of such an information process is displayed as
the gray ragged line between the two thresholds. Outside the
thresholds, predicted response time distributions for both alter-
nate responses are displayed.
In a diffusion model analysis, response times are decomposed
into decision-time and non-decision time. The non-decision pa-
rameter (t0) is conventionally interpreted as capturing basic stimu-
lus encoding and, particularly, motor time (Ratcliff, Thapar, &
McKoon, 2006; Voss et al., 2004). Decision time is determined by
the duration of the continuous sampling process, that is, by the
time the diffusion process takes to reach an upper or lower thresh-
old. This diffusion process is characterized by its starting point (z),
threshold separation (a), and the mean slope (drift rate: v). As a
diffusion process is noisy, it fluctuates randomly around its mean
gradient, resulting in different response-times, and sometimes in
hitting the wrong threshold.
The drift rate (v) is determined by the efficiency with which stimu-
lus information is used to select a response. Large drift rates will corre-
spond with fast and accurate responses. Drift rates can be affected by
task difficulty (Voss et al., 2004) as well as by individual differences in
intelligence and working memory capacity (Schmiedek, Oberauer,
Wilhelm, Suess, & Wittmann, 2007).
The distance of response thresholds (a) is considered to indicate
response caution. A high response criterion (large values of a)
corresponds to slow but accurate responses, while small values of
a denote an impulsive decisional style that is characterized by fast
responses and low accuracy. The a parameter can reflect experi-
mentally induced speed–accuracy settings (Voss et al., 2004). Addi-
tionally, it has been shown that a more cautious setting of the
response criterion contributes notably to slow responding in elder-
ly people (e.g., Ratcliff et al., 2006).
Additional parameters in the full diffusion model (Ratcliff & Rouder,
1998) include inter-trial variability of the starting point, drift rate (sv),
and non-decision parameter (sz, sv, and st0, respectively).
1.5. Task-switching research using the diffusion model
The full diffusion model or simplified variants were used in
previous studies to analyze task-switching data (Karayanidis
et al., 2009, 2010; Klauer, Voss, Schmitz, & Teige-Mocigemba,
2007; Madden et al., 2009; Mansfield, Karayanidis, Jamadar,
Heathcote, & Forstmann, 2011; Schmitz & Voss, 2012). This has
shed light on how its parameters are associated with task-
switching processes.
Consider first the non-decision parameter (t0 ) which captures
processes outside the actual decision phase. Across all studies, it
was found that the non-decision parameter (t 0 ) is increased in
task-switch trials relative to task-repeat trials. In line with
additional process theories of task switching, it was suggested that
this increase corresponds with the requirement to prepare the
new task in these trials (Karayanidis et al., 2009; Schmitz & Voss,
2012). This interpretation was further corroborated by the obser-
vation that the non-decision time was increased only when partic-
ipants could not prepare the new task set prior to stimulus
presentation; otherwise it was reduced to the level as in task-
repetition trials (Karayanidis et al., 2009; Schmitz & Voss, 2012).
These findings can be reconciled with the view that task-preparation
is largely possible prior to stimulus onset. However, when the
time does not suffice to prepare the up-coming task sufficiently,
preparation has to continue after stimulus presentation before
stimulus information can be efficiently used to select the response.
Note that the reported findings of an increase in non-decision
time in task-switch trials relative to task-repeat trials were
interpreted to reflect high-level control processes. This does not
question the conventional interpretation that the baseline level of
this parameter (as indexed by the magnitude of t 0 in task repeat
trials) captures basic perceptual processes and motor execution.
However, the increase in the non-decision parameter (t 0 ) is
not an unequivocal evidence of task-set preparation (see also
Schmitz & Voss, 2012). In fact, increased non-decision time in
task-switch trials might be also predicted by the compound-cue
retrieval/repetition priming account (Logan & Bundesen, 2003;
Schneider & Logan, 2005, 2007). Recall that cue-encoding should
be particularly fast in cue-repetition trials because of a transient
trace of the task-cue in working memory. This should save time
in compound cue formation in these trials relative to cue-switch
trials. Hence, the difference in the non-decision parameter (t 0 )
in task-switch and task-repeat trials can be easily reconciled
with the repetition priming account.
Consider next parameters of the response-selection phase
which may correspond with processes at a later phase when stimu-
lus information is used to select the response in line with settings in
the prepared task set (Mayr & Kliegl, 2003; Ruthruff et al., 2001). It
was argued that drift rates are affected by the state of task readiness
(Schmitz & Voss, 2012) which can be influenced by both controlled
and automatic factors (Koch & Allport, 2006; Rubinstein et al.,
2001). If there is not sufficient time for preparation (at short CSI),
drift rates should primarily reflect inertia effects, which can be ben-
eficial in the case of task-repetition trials or adverse in the case of
task-switches. However, if there is sufficient time to prepare (at
long CSI), preparation can be used additionally to adjust task read-
iness. Accordingly, it was shown that drift rates are influenced by
both task sequence and task predictability (Karayanidis et al.,
2009; Schmitz & Voss, 2012).
Finally, previous research has offered evidence that the re-
sponse criterion is increased in mixed blocks as compared to task-
pure blocks (Schmitz & Voss, 2012). Additionally, some studies
have shown that the response criterion may vary from trial to
trial within one block as a function of the task–cue, or as a function
of the anticipated difficulty or error-proneness of the transition
(Karayanidis et al., 2009; Mansfield et al., 2011; Schmitz & Voss,
 F. Schmitz, A. Voss / Acta Psychologica 151 (2014) 184–196
2012). In the study conducted by Karayanidis et al. (2009), it was
found that participants increased the response criterion only
when a difficult task-switch was cued with certainty, whereas in
task-switch trials cued with non-informative cues, the response-
criterion was comparable to that in task-repetition trials. Similarly,
Schmitz and Voss (2012) report that participants generally in-
creased the response criterion in mixed blocks, but, they slightly re-
duced the response criterion when an easy task-repeat trial could
be predicted with certainty. Additionally, results suggest that
adjusting the response criterion requires time (or cognitive capac-
ity) and is apparently only completed when the nature of the next
task transition is known early enough prior to stimulus onset
(Schmitz & Voss, 2012). Neuropsychological evidence further sug-
gests that setting the response criterion is exerted within a fronto-
striatal network, with the pre-supplementary motor area biasing
the striatum towards lower response thresholds (Mansfield et al.,
2011).
1.6. Aims of the current study and predictions
In previous research using the diffusion model to decompose
task switching costs, task switches and cue switches were con-
founded. Hence, one of the aims of this study was to tease apart ef-
fects of task switching from those of cue switching. Further, the
previously obtained findings in the parameters of the diffusion
model could be reconciled both with multiple component models
of task-switching and the compound-cue retrieval/repetition prim-
ing account. Consequently, we intended to test both theoretical ac-
counts more specifically on the level of parameters. To this end we
employed a task-switching paradigm with a 2:1 cue-to-task map-
ping (Logan & Bundesen, 2003; Mayr & Kliegl, 2003) while keeping
the procedure largely comparable with that in the previous studies
(Schmitz & Voss, 2012). We expected the usually found pure task-
switch and cue-switch effects in the observable performance data
(see Jost et al., 2013, for a review). More importantly, a number of
specific predictions could be derived from the theoretical accounts
that go beyond mean response times and errors that have been
previously investigated. These predictions could be tested with
the diffusion model.
Multiple components models motivated a number of predic-
tions. Given that additional processes are required to prepare the
upcoming task in task-switch trials, there should be an increase in
non-decision time in task-switch trials relative to task-repeat trials
when there is not sufficient time to prepare for the upcoming task
prior to stimulus onset (i.e. at short CSI). One can speculate that
the non-decision time is also slightly increased in task-repeat trials
with cue switches, as the participant needs to realize that the same
task repeats. But, such an orienting reaction would be expected to
be rather small compared with the time required for actual task
preparation. When the CSI is short, drift rates should largely reflect
task set inertia that can be adverse (in the case of task-switches) or
beneficial (in the case of task-repetitions). If the CSI is long, task set
preparation can be used more efficiently to adjust task readiness.
Then, drift rates can be expected to be additionally affected by
preparation.
According to the compound-cue retrieval/repetition-priming
account, the task-switch (or actually cue-switch) effect is predicted
to be largely caused by differences in non-decision time. The latter
should be decreased in cue-repeat trials relative to both other trial
types, as the task cue repeats only in these trials. Consequently, the
beneficial effect of a transient representation from the last trial will
only exceed an effect in cue-repeat trials. Conversely, non-decision
time would be increased in task-switch/cue-switch trials and task-
repeat/cue-switch trials, but no substantial difference between
these trial types is predicted. Given that response selection follows
a random walk process with only a few steps, drift rates can be
187
predicted to be generally steep, but comparable across different
trial types.
Neither theoretical account makes specific predictions concerning
the adjustment of response caution, although classical definitions of
task set have incorporated the “degree of caution to set one's criterion
for response” (Monsell, Yeung, & Azuma, 2000, p. 252). Based on previ-
ous results obtained in some studies using the diffusion model, we pre-
dicted that participants may adjust response caution in a trial-to-trial
fashion given that the nature of the next trial can be predicted and
there is sufficient time prior to stimulus onset.
2. Method
2.1. Sample
Sixty-two participants volunteered in the study (26 female). Most of
them were students, 8 of which with a major in Psychology. On average
they were 23.9 years old (SD = 2.5; range 18–30). All participants re-
ported normal or corrected to normal vision. They received 7 Euros or
partial course credit as compensation.
2.2. Procedure
After signing informed consent, participants were individually
seated in a dimly-lit room. All tasks and instructions were present-
ed on a 19 in., 100 Hz CRT monitor. The session started with the
task-switching paradigm and was completed by a few demographic
questions that were also presented on the screen. After that, partic-
ipants were thanked and debriefed. The total duration of the exper-
iment was about 45 min.
2.3. Paradigm and materials
We used a variant of the classical number–letter task-switching
paradigm (Rogers & Monsell, 1995) with a 2:1 cue-to-task mapping
(Arbuthnott & Woodward, 2002; Logan & Bundesen, 2004). Digit–
letter pairs were used as bivalent stimuli that could be either clas-
sified according to the parity of the digit (odd vs. even) or according
to the type of letter (vowel vs. consonant). All stimuli appeared in a
rectangle in the center of a light gray screen (RGB = 220, 220, 220).
The background color of the rectangle served as a task cue. Four dif-
ferent background colors (red, blue, green, and yellow) were ran-
domly assigned the two tasks so that each task was cued by two
colors. The cue–stimulus interval (CSI) was manipulated in two
groups: in the short CSI group, the background color of the rectan-
gle was changed from neutral gray to the respective task–cue color
150 ms prior to stimulus onset, whereas in the long CSI group, the
new task–cue color appeared 800 ms prior to stimulus onset. The
response–stimulus interval (RSI) was fixed at 800 ms in both
groups.
Stimuli were identical to the ones presented by Rogers and
Monsell: Odd digits were drawn from the set 3, 5, 7, 9, even digits
were drawn from the set 2, 4, 6, 8, consonants were drawn from
the set G, K, M, R and vowels were drawn from the set A, E, I, U.
All stimuli were presented in black Courier font. The digit–letter
pair was approximately 2 cm wide and 1.2 cm high. The rectangle
in which the stimulus pair appeared was ca. 7.5 cm × 7.5 cm and
surrounded by a thin black line. The background colors used as
task cues were red (RGB = 221, 97, 97), blue (RGB = 109, 179,
255), green (RGB = 94, 224, 76), and yellow (RGB = 228, 224,
76). They were randomly assigned to the two tasks (digit task vs.
letter task) in a 2:1 manner for each participant and their assign-
ment did not change throughout the experiment. In case of a classi-
fication error, a gray ‘X’ appeared below the stimulus pair. All
stimuli remained on screen until the correct response was given.
Then the stimulus was removed from the screen and the rectangle
188 F. Schmitz, A. Voss / Acta Psychologica 151 (2014) 184–196
was set to either neutral gray (in the short CSI group) or directly to
the task–cue of the next trial (in the long CSI group).
The experiment comprised 5 runs of blocks, each composed of
two task-pure blocks (one for each task, each block comprised 32
trials plus 4 warm-up trials) and a mixed block comprising both
tasks (128 trials plus 8 warm-up trials). Participants could make a
short pause after completing each block and re-starting the next
block was self-paced. Individual trial lists were generated prior to
the experiment for each participant. Half of all trials were task-
switches and half were task repetitions; half of the latter with
cue-switches and half with cue-repetitions. In the task pure blocks
also, there were cue-repetition and cue-switch trials. Within each
of these trial types, there were equal numbers of trials for the com-
binations of task, response congruency of the relevant and irrele-
vant stimulus features, and response side. As in the Rogers and
Monsell (1995) study, there were no direct stimulus repetitions,
avoiding trial-to-trial stimulus-specific effects of negative priming
or competitor priming. Additionally, there were no sequences of
more than 4 trials affording the classification with the same key.
Trials from the first sequence of blocks were discarded because of
strong training effects, as done in the Schmitz and Voss (2012) study.
The analyzed four sequences of blocks consisted of 128 cue-repeat and
128 cue-switch trials from the task-pure blocks, and of 256 task-
switch/cue-switch trials and 128 task-repeat/cue-switch and 128 task-
repeat/cue-repeat trials from the mixed blocks.
2.4. Analyses
Parameters of the diffusion model were estimated with fast-dm
(Voss & Voss, 2007, 2008) for each person and condition separately
(see Schmitz & Voss, 2012 for details)1. The full set of parameters of
the diffusion model was estimated without constraints using the
information in the joint cumulative distributions of erroneous and
correct responses minimizing the vertical distance of the empirical
and model-implied functions according to the Kolmogorov–
Smirnov statistic. In case of a low number of error trials, parameter
estimation is largely based on the information in the distribution of
correct responses. Parameter estimation is nevertheless possible,
because responses were coded as correct or erroneous and the
starting point was fixed in the middle between both response
thresholds. Conventional RT and error data as well as parameters
of the diffusion model were submitted a number of analyses of var-
iance (ANOVAs) with the trial type as a within participants factor
and the CSI length as a between participants factor. Mixing costs
are addressed first. To this end, an ANOVA was conducted across
task repeat trials from the task pure block and from the mixed
task block. Contrasting trials of the same type across the task pure
and mixed blocks may index block-wise differences in working
memory load, specific processing strategies, global response caution,
and other settings independent of the local task-switching effect.
Next, two separate ANOVAs were conducted for each block. Most im-
portant for the purpose of the current study are the results of the
ANOVA across trials of the mixed block which was complemented by
theoretically informative contrasts capturing pure task-switch and
cue-switch effects for each group (cf. Horoufchin et al., 2011, Exp. 4;
Koch, Lawo, Fels, & Vorlander, 2011, Exp. 3). Specifically, the task-
switch contrast denoted the difference between task-switch/cue-
switch and task-repeat/cue-switch trials, whereas the cue-switch
contrast was composed of task-repeat/cue-switch and task-repeat/
cue-repeat trials.
1
We counterchecked parameter estimates using the EZ program (Wagenmakers, van
der Maas, & Grasman, 2007; Wagenmakers et al., 2008), and findings were highly
comparable.
3. Results
3.1. Data pre-treatment
As done in the Schmitz and Voss (2012) study, trials directly follow-
ing an error (7.6%) were removed for two reasons: First, there may be
confounding error-related cognitive processing unrelated to the de-
mands of the actual task. Second, the nature of the next trial is not un-
equivocal, as task-confusion errors will invert the type of trial
transition (i.e., task-switch trials become task-repetition trials, and
vice versa; see Steinhauser & Hübner, 2006). Additionally, we excluded
outlier latencies according to individual box-plot criteria (Tukey, 1977).
Response times were considered to be outliers when they were longer
than 3 inter-quartile ranges on top of the 75% percentile of the RT distri-
bution, or 3 inter-quartile ranges below the 25% percentile or shorter
than 200 ms (3.3% excluded trials, 0.1% as fast, 3.2% as slow outlier
values). We tested whether the background color of the rectangle had
any effect on response times or errors. There was neither a main effect
nor an interaction effect with trial type (all p N .21). Parameter esti-
mates of the diffusion model were found to closely reproduce empirical
RT distributions. A graphical display of the model fit is displayed
Appendix A. Additionally, model-implied response times and error
rates are displayed in Fig. 2.
3.2. Behavioral data
3.2.1. Latencies
As can be seen in the upper panel of Fig. 2, median response
times (RTs) from the task pure blocks were comparable in both
CSI groups, but the increase in response times in the mixed
block was more pronounced in the short CSI group relative to
the long CSI group. To test for group effects in mixing costs, we
conducted an ANOVA across task repeat trials from the task pure
block and from the mixed task block, collapsing cue-switches
and cue-repetitions. This 2 (task pure vs. mixed task block) by
2 (short vs. long CSI) ANOVA yielded a main effect of trial type
(F[1,60] = 107.53, p b .001, η p 2 = .64), a main effect of the CSI
group (F[1,60] = 10.25, p b .01, ηp2 = .15), and an interaction of
both factors (F[1,60] = 32.12, p b .001, ηp2 = .35). The latter indi-
cated that the increase in response times from the task pure block
to the mixed task block was more pronounced in the short CSI
group (ΔRT = 198 ms) than in the long CSI group (ΔRT =
58 ms). Next, separate ANOVAs were conducted for each block.
For the task-pure blocks, a 2 (trial type) by 2 (CSI) ANOVA con-
firmed that latencies did not differ between trial types or groups
(all p N .26). Differently, in the mixed block, a 3 (trial type) by 2
(CSI) ANOVA revealed main effects of trial type (F[2, 120] =
77.56, p b .001, ηp2 = .56) and of CSI group (F[1, 60] = 24.59, p b .001,
ηp2 = .29), indicating generally slower RTs in the short CSI group
across all trial types (all p b .001). Additionally, there was an interac-
tion of both factors (F[2, 120] = 18.16, p b .001, ηp2 = .23). This pat-
tern was characterized by a significant task-switch contrast (ΔRT =
158 ms) as well as a cue-switch contrast (ΔRT = 98 ms) in the short
CSI group (F[1, 30] = 75.47, p b .001, ηp2 = .72 and F[1, 30] = 23.68,
p b .001, ηp2 = .44, respectively). In the long CSI group, only the task
switch contrast (ΔRT = 73 ms; F[1, 30] = 22.20, p b .001, ηp2 = .43)
reached significance, but not the cue-switch contrast (ΔRT =
14 ms; F[1, 30] = 3.47, p = .07, ηp2 = .10).
3.2.2. Accuracy
As can be seen in the lower panel of Fig. 2, the long CSI group
tended to commit more errors across all trial types. But there was
not much of a difference between the task pure and the mixed
task block (Δerrors = − 0.4% and Δerrors = 0.9%, for the short
and long CSI groups, respectively). This was confirmed in an
ANOVA across task repeat trials with block type and CSI group as factors.
 F. Schmitz, A. Voss / Acta Psychologica 151 (2014) 184–196
Fig. 2. Response times and error rates as a function of trial type (TR/CS = task repetition/cue switch, TR/CR = task repetition/cue repetition, TS/CS = task switch/cue switch, TR/CS = task
repetition/cue switch, TR/CR = task repetition/cue repetition) and cue–stimulus interval (CSI: long vs. short). Columns display the observed median RT and the proportion of errors, re-
spectively. Centered error bars represent the 95% confidence interval (CI) of these statistics. The “+” sign and its corresponding 95% CI shifted to the right of each column represent the
model implied prediction.
Only the main effect of CSI group was significant (F[1, 60] = 5.36, p b .05,
ηp2 =.08), but neither the main effect of block (F[1, 60] = 0.07, p =
.79, ηp2 b .01), nor the interaction of both factors (F[1, 60] =0.58, p =
.45, ηp2 = .01). Testing separately in the task-pure block, a 2 (trial type)
by 2 (CSI) ANOVA confirmed that there was only a main effect of the
CSI group (F[1, 60] = 6.31, p b .05, ηp2 = .10). Neither the main effect
of trial type nor its interaction with the CSI group was statistically signif-
icant (both p N .44). Consider next the mixed block, a 3 (trial type) by 2
(CSI group) ANOVA only yielded a significant main effect of the trial
type (F[2, 120] = 18.46, p b .001, ηp2 =.24), whereas the main effect of
the CSI group (F[1, 60] = 3.73, p =.06, ηp2 = .06), and an interaction ef-
fect of both factors (F[2,120] =3.05, p = .051, ηp2 = .05) missed signif-
icance. In the short CSI group, there was a significant task-switch
contrast (Δerrors = 1.9%, F[1, 30] = 11.22, p b .01, ηp2 = .27) and a
cue-switch contrast (Δerrors = 1.6%, F[1, 30] = 10.18, p b .01, ηp2 =
.25). In the long CSI group, only the task-switch contrast (Δerrors =
3.5%, F[1, 30] =15.74, p b .001, ηp2 = .34) was significant, whereas
the small cue-switch contrast in the non-predicted direction was
not significant (Δerrors = − 1.0%, F[1, 30] = 1.81, p = .19, ηp2 =
.06).
3.3. Diffusion model analyses
3.3.1. Response criterion (a)
Estimates for the a parameter are displayed in the upper panel of
Fig. 3. In the task-pure blocks, a comparable response criterion was
estimated for both CSI groups. In the mixed blocks, the response
criterion was generally increased, but the increase was more pro-
nounced in the short CSI group (Δa = 0.71) as compared to the
long CSI group (Δa = 0.27). The ANOVA testing mixing costs across
trials of the task pure and mixed task block yielded a main effect of
block (F[1,60] = 145.83, p b .001, η p 2 = .71), a main effect of CSI
group (F[1,60] = 11.96, p b .001, η p2 = .17), and an interaction of
both factors (F[1,60] = 28.95, p b .001, η p 2 = .33). An ANOVA
across trials of the task pure blocks with trial type (2) and CSI
group (2) as factors neither indicated a main effect of trial type,
nor of the CSI group, nor an interaction effect of both factors (all
p N .26). An ANOVA across trial types of the mixed block with trial
type (3) and CSI group (2) as factors revealed a main effect of trial
type (F[2, 120] = 3.13, p b .05, η p 2 = .05), a main effect of CSI
189
group (F[1, 60] = 17.03, p b .001, ηp2 = .22), but no interaction ef-
fect (F[2, 120] = 1.42, p = .24, η p2 = .02). In the short CSI group,
there was neither a task-switch contrast (Δa = 0.04, F[1, 30] =
0.78, p = .39, η p 2 = .03) nor a cue-switch contrast (Δa = 0.07,
F[1, 30] = 0.91, p = .35, η p 2 = .03). In the long CSI group, the
task-switch contrast missed significance (Δa = 0.06, F[1, 30] =
3.62, p = .07, ηp2 = .11), but the cue-switch contrast was significant
(Δa = 0.08, F[1, 30] = 4.62, p b .05, ηp2 = .13), and so was the total
switching effect denoting the difference between task-switch/cue-
switch and task-repeat/cue-repeat trials (Δa = 0.14, F[1, 30] = 12.42,
p b .01, ηp2 = .29).
3.3.2. Drift rates (v)
Drift rates were generally higher in trials of the task-pure blocks
compared with those of the mixed blocks, whereas effects of the CSI
group seemed to depend on trial type: Drift rates were higher in the
short CSI group relative to the log CSI group, but they were lower in
the mixed block. This resulted in a more pronounced decrease from
the task pure to the mixed task block in the short (Δν = 1.16) relative
to the long CSI group (Δν = 0.58). This was confirmed in an ANOVA
across task repeat trials of the task pure and of the mixed task block.
There was a main effect of the block type (F[1, 60] = 67.14, p b .001,
ηp2 = .53), no main effect of the CSI group (F[1, 60] = 1.42, p = .24,
ηp2 = .02), but an interaction of both factors (F[1, 60] = 7.52, p b .01,
ηp2 = .11). Testing across trial types of the task pure blocks, the 2
(trial type) by 2 (CSI) ANOVA only revealed a main effect of the CSI
group (F[1, 60] = 10.81, p b .01, ηp2 = .15), but neither a main effect
of trial type (F[1, 60] = 0.56, p = .46, ηp2 b .01) nor an interaction of
both factors (F[1, 60] = 0.46, p = .50, ηp2 b .01). Across trials of the
mixed blocks, the 3 (trial type) by 2 (CSI) ANOVA yielded a main effect
of trial type (F[1, 120] = 62.73, p b .001, ηp2 = .51), no main effect of
the CSI group (F[1, 60] = 0.80, p = .38, ηp2 = .01), but an interaction
of both factors (F[1, 120] = 11.11, p b .001, ηp2 = .15). In the short
CSI group, there was a significant task-switch contrast (Δν = 0.52,
F[1, 30] = 18.64, p = .001, η p2 = .38) and a cue-switch contrast
(Δν = 0.97, F[1, 30] = 49.48, p = .001, η p2 = .62). Differently in
the long CSI group, there was only a significant task-switch contrast
(Δν = 0.71, F[1, 30] = 19.79, p = .001, η p 2 = .40), but no cue-
switch contrast (Δν = 0.07, F[1, 30] = 0.16, p = .69, ηp2 = .01).
190 F. Schmitz, A. Voss / Acta Psychologica 151 (2014) 184–196
Fig. 3. Parameters of the diffusion model as a function of trial type (TR/CS = task repeti-
tion/cue switch, TR/CR = task repetition/cue repetition, TS/CS = task switch/cue switch,
TR/CS = task repetition/cue switch, TR/CR = task repetition/cue repetition) and cue–
stimulus interval (CSI: long vs. short). Error bars represent the 95% confidence interval
of the mean.
3.3.3. The non-decision parameter (t0)
As can be seen in the bottom panel of Fig. 3, estimates of the non-
decision parameter were comparable across all trial types with the
exception of a marked increase in the non-decision parameter in task-
switch trials in the short CSI group. In task-repeat trials, there was not
much of a difference between the task pure and mixed task blocks, nei-
ther in the short (Δt0 = .007) nor in the long CSI group (Δt0 = −.008).
This was confirmed in the ANOVA which yielded neither a main effect of
block type (F[1, 60] b 0.01, p = .96, ηp2 b .01), nor a main effect of CSI
group (F[1, 60] = 1.30, p = .26, ηp2 = .02), nor an interaction of both
factors (F[1, 60] = 0.12, p = .73, ηp2 b .01). In the ANOVA across trials
of the task pure block, none of the main or interaction effects was
significant (all p N .61). The ANOVA across trial types of the mixed
block yielded a main effect of trial type (F[2,120] = 44.73, p b .001,
ηp2 = .43), a main effect of CSI (F[1,60] = 11.85, p b .001, ηp2 = .16),
and an interaction of both factors (F[2,120] = 28.88, p b .001, ηp2 =
.33). The only contrast that indicated a substantial effect was the task-
switch contrast in the short CSI group (Δt0 = .117, F[1, 30] = 56.16,
p b .001, η p 2 = .65). Differently, the non-decision time was
comparable in both conditions of the cue-switch contrast in the short
CSI group (Δt0 = .017, F[1, 30] = 4.00, p = 06., ηp2 = .12). In the
long CSI group, there was virtually no total switching effect in this pa-
rameter (Δt0 =.003, F[1, 30] = 0.23, p = .64, ηp2 b .01), as indicated
by the highly comparable non-decision time in task-switch/cue-switch
and task-repeat/cue-repeat trials. This suggests that the constituents of
the total switching effect, namely the pure task-switching and cue-
switching effects, are also absent. The observation of a small decrease
in non-decision time in task-repeat/cue-switch trials relative to both
other trial types (Δt0 = .017, F[1, 30] = 4.68, p = .04, ηp2 = .14, and
Δt0 = − .014, F[1, 30] = 5.14, p = .03, ηp 2 = .15, for task-switch/
cue-switch and task-repeat/cue-repeat trials, respectively) violates
the logic of the 2:1 mapping paradigm, indicating that the specific
contrasts should not be computed in this case.
4. Additional analyses
Small unpredicted effects such as the latter provoke the question
whether these effects are substantial or whether parameters could
be constrained to be equal across conditions. As formal model com-
parisons are not possible using the KS method, we estimated param-
eters again for each participant using maximum likelihood (ML) on
the basis of individual responses. Additionally to the model without
parameter constraints, a number of alternative models were fitted in
which parameters were systematically constrained to be equal
across conditions. Appendix B summarizes results obtained for 28
models with different combinations of parameter constraints, sepa-
rate for the short and long CSI groups. All models were formally test-
ed against the baseline model (#1) without parameter constraints
using the likelihood-ratio chi-squared (LR-χ2 ) test. In case of a
non-significant difference, the Akaike information criterion (AIC)
and the Bayesian information criterion (BIC) were used to determine
the best model in terms of fit and parsimony.
In terms of the LR-χ2 test alone, the response criterion (a) could be
constrained across conditions in the short and long CSI groups without
a significant increase in model misfit. Differently, constraining drift rates
(ν) across all conditions was not possible, either alone (in the short CSI
group), or after constraining the response criterion (in the long CSI
group). In the short CSI group, the markedly increased non-decision pa-
rameter (t0) in task-switch/cue-switch trials could not be constrained to
be equal with any of the other conditions. Differently, non-decision time
could be constrained across all conditions in the long CSI group, alone
and after constraining the response criterion.
When comparing all models that did not differ significantly in LR-χ2,
the AIC and BIC model comparison indices converged in their recom-
mendation of the unconstrained model in both groups. Accordingly, pa-
rameter estimates of this model were submitted ANOVAs to test for
task-switch and cue-switch effects. Generally, the pattern of parameter
estimates across conditions was very similar to the ones of the KS anal-
yses (see Appendix C for descriptives of the ML parameter estimates).
Only the difference between trial types in the response criterion in the
long CSI group did not reach significance in the ML analyses anymore
(p = .12, for the cue-switch contrast), however, a lack of statistical
power cannot be ruled out either. (Actually, this contrast was also sig-
nificant in the EZ analyses in the long CSI group). Apart from that, the
short CSI group generally displayed increased response caution relative
to the long CSI group (p b .001). In the short CSI group, drift rates
were lowest in task-switch/cue-switch trials and highest in task-
repeat/cue-repeat trials, and all conditions differed from each other
(all p b .05). Differently in the long CSI group, drift rates were reduced
in task-switch/cue-switch trials relative to both other trial types (both
p b .001), but they were comparable in task-repeat/cue-switch and
task-repeat/cue-repeat trials (p = .60). Importantly, the pronounced
increase in the non-decision parameter selectively occurred in the
short CSI group relative to both other trial types (both p b .001). Differ-
ently in the long CSI group, task-switch/cue-switch trials did not differ
 F. Schmitz, A. Voss / Acta Psychologica 151 (2014) 184–196
from task-repeat/cue-repeat trials (p = .62). Again, there was an
unpredicted slight decrease in task-repeat/cue-switch trials relative to
the other conditions in the long CSI group (both p b .05). This small,
unpredicted effect could have been avoided when more constrained
models would have been selected, such as model 20 for the short CSI
group and model 24 for the long CSI group. The pattern in the other pa-
rameters would have been identical in this case.
5. Discussion
The aim of the current study was to dissociate processes of cue
switching and task switching with the diffusion model. To this end,
we employed an explicit cueing paradigm with a 2:1 cue-to-task map-
ping to tease apart both effects and decomposed the performance data
with the diffusion model. By this means, we intended to investigate
predictions derived from multiple component models of task switching
and the compound-cue retrieval/repetition priming account. In the fol-
lowing, we will, first, address effects in the behavioral data. Next, find-
ings for the parameters of the diffusion model will be presented.
5.1. Findings in the behavioral data
Replicating previous studies on mixing costs (Rogers & Monsell,
1995; Ruthruff et al., 2001), we found that the response times were gen-
erally increased in mixed blocks compared to task-pure blocks across all
trial types. More importantly, across trial types of the mixed blocks,
there was evidence of costs related to both cue-switch and task-
switch. Particularly in the short CSI group, RTs were longer in cue-
switch trials relative to cue-repeat trials, indicative of a cue-switch
effect (or repetition priming; Logan & Bundesen, 2003). However, RTs
in the task-switch trials were further increased by about the same mag-
nitude in task-switch relative to cue-switch trials, indicative of a pure
task-switch effect. RTs were generally shorter in the long CSI group rel-
ative to the short CSI group and they were only slightly increased in the
task-switch trials relative to the other trial types in the mixed block.
This pattern was predicted by accounts which posit an extra phase of
task-set preparation in task-switch trials (e.g., Rogers & Monsell,
1995). However, findings also corroborate that preparation can be
largely completed prior to stimulus onset given the predictability of
the next task set and sufficient time to complete preparation. The
absence of a cue-switch contrast in the long CSI group, suggests that
800 ms was sufficient time to arrive at a comparable state of task read-
iness for these participants. Note that reduced task-switch costs in the
classical 1:1 mapping procedure at long CSI have been previously re-
ported in studies using a within-participants design, but not necessarily
in ones using a between-participants design (e.g., Altmann, 2004). It
was suggested that this could reflect strategic factors to engage in
task-set preparation (Koch, 2001). One may speculate that participants
in the current study considered it relevant to prepare for the up-coming
task because of the 2:1 cue-to-task mapping. Additionally, participants
in the long CSI group might have been confident enough to believe
that they will be successful in doing so. In fact, generally fast response
times and increased error rates relative to the short CSI group suggest
a confident processing style in the long CSI group (which was also cor-
roborated by the lower response criterion setting, see below).
Note that the behavioral results can be generally reconciled with the
compound-cue retrieval/repetition priming account. Previous studies
have shown that the effects of CSI on mean response times can be
modeled in the absence of reconfiguration (Logan & Bundesen, 2003;
Schneider & Logan, 2005). Additionally, the absence of a major differ-
ence in RTs in cue-switch and cue-repeat trials in the long CSI group
could indicate that cue-encoding has been finished after 800 ms in
both conditions. However, the further increase in task-switch relative
to cue-switch trials seems to be a challenge to explain in the absence
of task-switch specific processes.
191
For both groups, the pattern of effects in the error scores
across trial types largely paralleled those obtained in the RTs. In
the short CSI group, there was a task-switch as well as a cue-
switch effect, whereas in the long CSI group there was only a
task-switch effect but no cue-switch effect. However, the general
performance level of both groups was inverted: The long CSI
group tended to commit more errors across all trial types com-
pared with the short CSI group. The pattern of shorter RTs and in-
creased errors suggests that the long CSI encouraged participants
to respond less cautiously. This was tested next more directly
with the diffusion model.
5.2. Findings in the parameters of the diffusion model
5.2.1. Non-decision parameter (t0)
According to multiple component models that also specify an
extra phase of task-set preparation, non-decision time was pre-
dicted to be increased in trials affording task-switches, particu-
larly, if the time prior to stimulus onset does not suffice to
finish preparation in advance. The finding of a selective increase
in the non-decision parameter in task-switch trials in the short
CSI group seemed to support this prediction. Additionally, the ab-
sence of a notable task-switch contrast in the long CSI group im-
plies that whatever processes take place during this phase, they
can be largely completed prior to stimulus onset given the infor-
mative task cues and sufficient time, supporting previous findings
(cf. Karayanidis et al., 2010, and Karayanidis et al., 2009, for the
role of cue-informativeness; cf. Schmitz & Voss, 2012, for the
role of preparation time).
The absence of a cue-switch effect implies that there is no pronounce
orienting phase, or, that this phase has been finished in the 150 ms prior
to stimulus onset. Additionally, it constrains the interpretation of the t0
effect: Task-set retrieval from long-term memory (Goschke, 2000; Mayr
& Kliegl, 2003) cannot account for the increase in the non-decision pa-
rameter, as these retrieval processes should also take place on task-
repeat/cue-switch trials. Plausibly, these earlier processes have been
completed in the available CSI, anyway. Most likely, the task-switch spe-
cific increase in the non-decision parameter indexes processes at the
later stage of getting the task-set ready in procedural working memory
(Risse & Oberauer, 2010; see also Goschke, 2000; Mayr & Kliegl, 2000,
2003), which may correspond with the biasing of its task-demand
units (Gilbert & Shallice, 2002; see also Ruthruff et al., 2001).
In any case, the selective effect obtained in the current study implies
that the increase in non-decision time may be used as a relatively specif-
ic indicator of the pure task-switch effect. Hence, the current finding
contributes to existing behavioral and physiological evidence that task
switching requires distinct processes that go beyond those in cue
switching (e.g., Horoufchin et al., 2011; Mayr & Kliegl, 2003; see Jost
et al., 2013, for a review of cue-processing in task switching).
From the perspective of the compound-cue retrieval/repetition
priming account, the pattern of results in the non-decision parameter
is more difficult to explain. The decreased non-decision time in task-
repeat/cue-repeat trials relative to task-switch/cue-switch trials makes
sense. However, non-decision time was not predicted to be decreased
in task-repeat/cue-switch trials, as the task-cue does not repeat in
these trials. Of course, one could argue that cue-encoding is largely com-
pleted by the end of the 150 ms CSI. However, even then, the pro-
nounced increase in task-switch/cue-switch trials seems to be difficult
to explain.
5.2.2. Drift rate (v)
Multiple component models predict that task readiness can be af-
fected by inertia effects and task preparation (Koch & Allport, 2006).
Consequently, drift rates should differ across trial types as a function
of trial transition and the time available for preparation. This was con-
firmed in the present study. To start with mixing effects, drift rates
192 F. Schmitz, A. Voss / Acta Psychologica 151 (2014) 184–196
were generally higher in the task-pure relative to the mixed blocks. This
makes sense given that all previous trials in the task pure block required
the same task set. Consequently, beneficial inertia effects would lead to
a particular high level of task readiness. Additionally, drift rates were
somewhat higher in the short CSI group relative to the long CSI
group, which may reflect the dissipation of cue-driven task-set
priming with increasing CSI. Further, the task-cue may function as
an alerting signal shortly before stimulus presentation. Both factors
would contribute to an optimal state of task-readiness at stimulus
presentation in the short CSI group, accordingly, evidence accumu-
lation should be steep.
To continue with the mixed block, drift rates in the short CSI group
had a pattern that would be predicted if drift rates were largely deter-
mined by inertia effects. Drift rates were lowest in task-switch trials,
as would be expected given the adverse inertia effects or incomplete ac-
tivation or biasing of the relevant task-set relative to the competing one
just switched away from. Drift rates were at an intermediate level in
cue-switch trials, which would be expected given that the relevant
task-demand units are still activated. Finally, drift rates were highest
in cue-repeat trials, possibly reflecting additional cue-driven task-set bi-
asing. Differently in the long CSI group, drift rates were reduced in task-
switch trials only, but comparably high in task-repetitions with cue-
switches and with cue-repetitions. This pattern suggests additional
effects of controlled task-set activation, particularly in the cue-switch
trials. It seems to be reasonable to assume that the long CSI left sufficient
time to realize that the same task will be repeated and, hence, re-
activate it, so that comparable levels of task-readiness can be achieved
as in cue-repeat trials.
Differently, the account on compound-cue retrieval/repetition prim-
ing does not specifically predict substantial differences in drift rates
across trial types. However, if one permits that repetition priming does
not only affect compound cue encoding/formation but also affect re-
sponse retrieval in case the compound-cue repeats, then faster drift
rates would be expected in task-repeat/cue-repeat trials. However, this
would not easily explain the difference obtained between task-switch/
cue-switch and task-repeat/cue-switch trials.
5.2.3. Response criterion (a)
Given that both theoretical perspectives do not make specific as-
sumptions with respect to trial-to-trial adjustment in response caution,
we based our predictions on previous studies that employed the diffu-
sion model. Confirming previous findings, the response criterion was
found to be generally increased in the mixed blocks relative to the
task-pure blocks (Schmitz & Voss, 2012). Consequently, part of the
mixing costs can be attributed to a more cautious response style in the
mixed block compared with the task-pure block. This was the case for
both CSI groups, however, the increase in the response criterion was
more pronounced in the short CSI group compared with the long CSI
group. This finding suggests that the long CSI group perceived the task
as less difficult and chose a more confident processing style.
Compared with the effects of block and of CSI group, the adjustment
of the response criterion in a trial-to-trial fashion was of small magni-
tude only. Results obtained for this effect were less consistent across
methods of parameter estimation, although a lack of statistical power
might have prevented statistical significance of a descriptive effect in
the ML analyses. If anything, participants in the long CSI group tended
to reduce response caution in the case of an easy task-repeat/cue-repeat
trial (as confirmed by the KS and EZ estimates). Differently in the short
CSI group, the response criterion did not differ between trial types of the
mixed blocks. This can be reconciled with previous findings that re-
sponse criteria are only adjusted when there is enough time (Schmitz
& Voss, 2012; see also Karayanidis et al., 2010; Karayanidis et al.,
2009), possibly because fine-grained trial-to-trial adjustment of the re-
sponse criterion consumes time, or is a resources-demanding process
(cf. Mansfield et al., 2011).
5.2.4. Evaluating model predictions
In the present study we used the diffusion model in order to test pre-
dictions derived from two important theoretical frameworks, (a) multi-
ple component models of task switching that comprise task-set
preparation and inertia effects, and (b) the compound-cue retrieval/
repetition priming account. The most important findings in this respect
will be summarized and discussed below.
Generally, it turned out that results in the parameter estimates can
be reconciled with multiple component models, whereas some of the
findings pose a challenge for the compound-cue retrieval account. As
to the latter, the absence of an increase in non-decision time in task-
repeat/cue-switch trials relative to task-repeat/cue-repeat trials is par-
ticularly difficult to explain. As the task cue has changed, there should
not be an advantage in the compound cue retrieval process. Additional-
ly, some of the trial-type dependent effects in the drift rate were not
specifically predicted. Possibly, one could argue that the response re-
trieval process is also facilitated in case the compound-cue repeats.
However, this cannot easily explain the increase in drift rate in task-
repeat/cue-switch trials over task-switch/cue-switch trials that was ob-
served in both groups.
Conversely, predictions from multiple-component models of task
switching (e.g., Mayr & Kliegl, 2003; Ruthruff et al., 2001) were generally
confirmed. The selective increase in the non-decision parameter seems to
correspond with the requirement to prepare for a new task set; and re-
sults of the present study extend previous findings that this effect is in-
deed task-switch (and not cue-switch) specific. However, the increase
in the non-decision parameter does not elucidate how exactly task-set
preparation takes place (but see e.g., Gilbert & Shallice, 2002; Goschke,
2000; Mayr & Kliegl, 2000, 2003; Risse & Oberauer, 2010; Ruthruff et al.,
2001, for theoretical accounts). Nevertheless, one of the core components
should be the activation of relevant S–R links (e.g., Allport et al., 1994;
Monsell et al., 2000) that will allow using stimulus information to select
the appropriate response in the next phase. The stronger the rele-
vant S–R link (relative to competing ones), the higher should be
the task readiness, corresponding with high drift rates. Drift rates
seemed to largely reflect inertia effects when the CSI is short. How-
ever, when there is sufficient time to prepare, drift rates may be
additionally affected by preparation (cf. Koch & Allport, 2006, for
the joint effects on task readiness).
5.3. Scope and limitations
5.3.1. Advantages of modeling task switching data
We conducted this study primarily with the intention to test predic-
tions derived from important theoretical accounts. However, diffusion
modeling may be generally an appealing technique for the analyses of
task-switching data that yields certain advantages over conventional
RT and error analyses. The increase in non-decision time may serve as
a relatively specific indicator of task preparation, as shown in the cur-
rent study. The drift rates in experimental conditions may offer a less bi-
ased performance measure of task-switching costs: They integrate the
information in correct and erroneous responses into a common metric
and simultaneously avoid a bias of response caution. The latter can be
additionally quantified with the criterion parameter.
5.3.2. Suitability of the paradigm
The explicit cueing paradigm with a 2:1 mapping of cues to tasks is
appealing as it promises to tease apart the task-switch and cue-switch
components. However, the paradigm was also criticized for manipulat-
ing conditional task-switch expectancies (Mayr, 2006; Schneider &
Logan, 2006). In fact, some of the divergent results obtained with the
2:1 mapping procedure concerning the relative magnitudes of cue-
switch and task-switch effects may, in part, depend on the conditional
probability of a task switch given a cue switch (see Mayr, 2006, for a
global; and Schneider & Logan, 2006, for a local account). Given a high
conditional probability, participants may engage in task switch activity
 F. Schmitz, A. Voss / Acta Psychologica 151 (2014) 184–196
upon the detection of a cue switch so that the relevant task-set needs to
be prepared again later. Additionally, an orienting reaction may take
place in case the actual task set does not match with expectancies.
Both would lead to increased response times in the task-repeat/cue-
switch condition (cf. Forstmann, Brass, & Koch, 2007), simultaneously
increasing estimates of the cue-switch effect and diminishing estimates
of the task-switch effect. Both effects could have taken place in the
current study, as a task switch occurred in 66% of all cue switches.
Nevertheless, there was still a substantial task-switch effect in the RT
data. Additionally, from a diffusion modeling perspective, one can pre-
dict that postponed task-set preparation and surprise/orienting would
contribute to increased non-decision time in task-repeat/cue-switch
trials. However, this was not observed in the present study.
The problem associated with cue-switch based expectancies could
have been circumvented by using the transition-cueing paradigm
(Forstmann et al., 2007) in which cue switches and task switches can
be combined orthogonally. However, this paradigm is characterized by
particularly large switch costs which are not fully understood
(Vandierendonck et al., 2010). Hence, we chose the 2:1 explicit cueing
paradigm in the current study which is more frequently used in the lit-
erature, too. Additionally, some of the predictions derived from the
compound cue/priming account could be directly tested using the 2:1
mapping paradigm.
5.3.3. Suitability of the diffusion model
The diffusion model applied in the current study is relatively sim-
ple. Hence, its parameters will not suffice to describe all possible
component processes discussed in the rich task switching literature
in a one-to-one fashion. Nevertheless, a model may be considered
useful, when it helps dissociating at least some of the processes of in-
terest or when it helps testing meaningful predictions (Brown &
Heathcote, 2008; Wagenmakers, van der Maas, Dolan, & Grasman,
2008). Following this rationale, we used the diffusion model in
order to test predictions derived from multiple component models
of task switching and the compound-cue retrieval/repetition prim-
ing account. Of course, this makes sense only, if the fitted model ad-
equately captures important aspects of the theoretical account. This
seemed to be relatively evident in the case of the compound-cue re-
trieval/repetition priming account that was developed within a
random-walk framework that closely resembles the diffusion
model. Hence, predictions appeared relatively straightforward.
Differently, multiple component models are generally more complex
and they were not explicitly developed with a continuous sampling
model in mind. However, as previously argued (Schmitz & Voss, 2012),
we believe that diffusion model parameters may help dissociate some of
the core processes, namely the effects of task preparation from response
selection, with the latter possibly more strongly affected by inertia
effects. At least, previous research using the explicit cueing paradigm
has supported this notion (cf., Karayanidis et al., 2009, 2010; Mansfield
et al., 2011; Schmitz & Voss, 2012).
Of course, the question arises as to whether this also holds for the cur-
rently used task-cueing paradigm with 2:1 mapping that was criticized
for introducing additional affordances such as cue-disambiguation
(Altmann, 2006; Kleinsorge, 2012). If the latter corresponds with an ad-
ditional decision process per trial, i.e. cue-based task-set selection, the
question arises as to whether the standard diffusion model is still suited
to describe the data. However, in support of the applied model, it has to
be noted that the pattern of results obtained in this study closely resem-
bles the results in previous studies using the explicit-cueing paradigms
with a 1:1 mapping (Karayanidis et al., 2009, 2010; Schmitz & Voss,
2012). Additionally, the model-implied RT distributions closely matched
empirical RT distributions, supporting the descriptive adequacy of the
model. Finally, the pattern observed in the estimated parameters
seems to correspond well with theoretical predictions, as previously
outlined.
193
5.3.4. Learning effects
Finally, it shall be discussed that task performance improved across
all runs of blocks, although the rate of improvement diminished from
block to block. To give numbers, we contrasted the RT in each block
with the mean of the following blocks for the short CSI (long CSI)
group. The RT in TS/CS trials improved by 319 ms (116 ms) from the
first to the following blocks; the next improvement was only 150 ms
(71 ms). Similarly, in TR/CS trials the first improvement was 272 ms
(159 ms), and the next one was 118 ms (53 ms). In TR/CR trials,
the first improvement was 203 ms (139 ms), and the next was 78 ms
(49 ms). The last improvement from the fourth to the fifth block
reached from 40 ms (TS/CS in the short CSI group) to 13 ms (TR/CR in
the long CS group). Hence, some of the contrasts were still significant
from the fourth to the fifth run. This non-stationary can inflate variance
and may bias parameter estimates, as the model fit tends to be sensitive
to variance. Generally, the effects of learning on parameter estimates
deserve more attention. Future studies should circumvent this limita-
tion by allowing enough trials to arrive at asymptotic performance.
5.4. Conclusion
The previously observed increase in the non-decision parameter (t0)
in task-switch trials was shown to be task-switch specific, whereas it
was not increased in cue-switch trials in which the task was repeated.
The selective increase in the non-decision parameter can be best recon-
ciled with additional process accounts of task switching which assume
that task-switches require additional preparation processes (Rogers &
Monsell, 1995; Rubinstein et al., 2001). The task-switch specific increase
further rules likely that preparatory processes such as loading the
task-set into procedural working memory (Risse & Oberauer, 2010),
transforming it into a task-appropriate attentional configuration
(Mayr & Kliegl, 2000, 2003), for instance by biasing relevant task-
demand units (Gilbert & Shallice, 2002), may contribute to the effect.
In the following phase, the task-set is applied to the stimulus to
retrieve the response, according to multiple component models of task
switching (Mayr & Kliegl, 2003; Ruthruff et al., 2001). The efficiency of
this process, as indicated by the drift rate (v), was shown to depend on
both task-predictability and task-sequence. The first points to the role of
some sort of task-set preparation (e.g., Rogers & Monsell, 1995), whereas
the latter supports the view of task-set inertia (e.g., Allport & Wylie,
2000), which jointly determine the state of task readiness (Koch &
Allport, 2006).
Finally, the response criterion (a) varied largely between task-pure
and mixed blocks and as a function of CSI length. However, trial-to-
trial adjustment within blocks was of comparatively small magnitude.
If there is trial-to-trial adjustment of the response criterion, it seems
to require time or capacity to complete prior to stimulus onset.
Appendix A. Graphical display of model fit
The plots show overlaid predicted (parameter based) and empirical
cumulative distribution functions (CDFs) for all trial types. Predicted
and empirical CDFs were first computed for each participant and then av-
eraged across participants. The plotted functions are joint CDFs of correct
and erroneous responses, with latencies of erroneous responses plotted
on the left side of the horizontal axis (with negative scale values) and la-
tencies of correct trials plotted on the right side (with positive scale
values). The intercept of the cumulative distribution function indicates
the percentage of erroneous responses. The CDFs for the long CSI and
short CSI group are displayed in the upper and lower panels, respectively.
Trial types of the task-pure and mixed blocks are shown on the left and
right sides, respectively. The order of trial types in the legend from top
to bottom corresponds with trial difficulty, as indicated by the order of
the steep increase in the CDF: from left to right task-repeat/cue-repeat
(TR/CR), task-repeat/cue-switch (TR/CS), and task-switch/cue-switch
(TS/CS) trials.
194 F. Schmitz, A. Voss / Acta Psychologica 151 (2014) 184–196

Appendix B. Model comparison

# Parameter constraints Short CSI Long CSI

LR-χ 2 (df) AIC BIC LR-χ 2 (df) AIC BIC

1 No constraints btw. conditions – 165.80 240.94 – −166.90 −92.54

2 a: TS = CS 2.09 (1) 167.90 243.04 0.51 (1) −166.39 −92.03
3 a: TS = CR 2.89 (1) 168.69 243.83 0.15 (1) −166.76 −92.39
4 a: CS = CR 0.66 (1) 166.46 241.60 0.06 (1) −166.85 −92.48
5 a: ALL 4.02 (2) 169.82 244.96 1.25 (2) −165.65 −91.28
6 ν: TS = CS 2.52 (1) 168.32 243.47 2.71 (1) −164.19 −89.82
7 ν: TS = CR 6.40 (1)* 172.20 247.34 2.97 (1) −163.93 −89.56
8 ν: CS = CR 2.30 (1) 168.10 243.24 0.89 (1) −166.01 −91.64
9 ν: ALL 7.41 (2)* 173.21 248.35 5.04 (2) −161.86 −87.50
10 t0: TS = CS 7.08 (1)** 172.88 248.03 1.25 (1) −165.66 −91.29
11 t0: TS = CR 7.19 (1)** 172.99 248.14 1.21 (1) −165.70 −91.33
12 t0: CS = CR 0.89 (1) 166.69 241.83 1.12 (1) −165.78 −91.41
13 t0: ALL 10.69 (2)** 176.49 251.63 2.44 (2) −164.47 −90.10
14 a: ALL; ν: TS = CS 4.95 (3) 170.75 245.90 5.84 (3) −161.07 −86.70
15 a: ALL; ν: TS = CR 17.17 (3)*** 182.97 258.12 9.99 (3)* −156.92 −82.55
16 a: ALL; ν: CS = CR 8.99 (3)* 174.79 249.93 2.61 (3) −164.29 −89.92
17 a: ALL; ν: ALL 18.87 (4)*** 184.67 259.81 11.21 (4)* −155.70 −81.33
18 a: ALL; t0: TS = CS 7.99 (3)* 173.79 248.93 2.38 (3) −164.53 −90.16
19 a: ALL; t0: TS = CR 11.22 (3)* 177.02 252.16 3.47 (3) −163.43 −89.06
20 a: ALL; t0: CS = CR 3.41 (3) 169.21 244.35 2.22 (3) −164.68 −90.32
21 a: ALL; t0: ALL 13.84 (4)** 179.64 254.79 4.14 (4) −162.76 −88.39
22 a: ALL; t0: ALL; ν: TS = CS 28.25 (5)*** 194.05 269.19 15.48 (5)** −151.43 −77.06
23 a: ALL; t0: ALL; ν: TS = CR 71.92 (5)*** 237.72 312.86 21.72 (5)*** −145.19 −70.82
24 a: ALL; t0: ALL; ν: CS = CR 27.18 (5)*** 192.98 268.12 6.23 (5) −160.67 −86.30
25 a: ALL; ν: ALL; t0: TS = CS 46.55 (5)*** 212.35 287.49 22.16 (5)*** −144.75 −70.38
26 a: ALL; ν: ALL; t0: TS = CR 70.97 (5)*** 236.78 311.92 23.06 (5)*** −143.85 −69.48
27 a: ALL; ν: ALL; t0: CS = CR 26.97 (5)*** 192.77 267.92 12.39 (5)*** −154.52 −80.15
28 a: ALL; ν: ALL; t0: ALL 81.54 (6)*** 247.34 322.48 28.55 (6)*** −138.35 −63.98

Note. a = response criterion, ν = drift rate, t0 = non-decision parameter; TS = task switch/cue switch, CS = task repeat/cue switch, CR = task repeat/cue repeat, ALL = parameters of
all conditions constrained to be equal; LR-χ 2 (df) = likelihood-ratio chi-squared test with df degrees of freedom, AIC = Akaike information criterion, BIC = Bayesian information crite-
rion, the value of the model recommended by each information criterion is printed in boldface; *p ≤ .05, **p ≤ .01, ***p ≤ .001.
 F. Schmitz, A. Voss / Acta Psychologica 151 (2014) 184–196 195

Appendix C. Means (standard deviations) of the maximum likelihood Kiesel, A., Steinhauser, M., Wendt, M., Falkenstein, M., Jost, K., Philipp, A. M., et al. (2010).
Control and interference in task switching — A review. Psychological Bulletin, 136(5),
parameter estimates 849–874.
Short CSI Long CSI Klauer, K. C., Voss, A., Schmitz, F., & Teige-Mocigemba, S. (2007). Process components of
the Implicit Association Test: A diffusion-model analysis. Journal of Personality and
TS/CS TR/CS TR/CR TS/CS TR/CS TR/CR Social Psychology, 93(3), 353–368.
Kleinsorge, T. (2012). Task switching with a 2:1 cue-to-task mapping: Separating cue dis-
a 1.87 (0.60) 1.83 (0.50) 1.83 (0.58) 1.36 (0.47) 1.41 (0.50) 1.33 (0.33)
ambiguation from task-rule retrieval. Psychological Research, 76(3), 329–335.
ν 1.89 (0.54) 2.31 (0.88) 3.81 (2.23) 2.21 (1.09) 3.28 (1.45) 3.41 (1.33) Koch, I. (2001). Automatic and intentional activation of task sets. Journal of Experimental
t0 .437 (.065) .318 (.071) .326 (.057) .334 (.058) .311 (.053) .330 (.045) Psychology: Learning, Memory, and Cognition, 27(6), 1474–1486.
sz 0.50 (0.25) 0.18 (0.25) 0.17 (0.23) 0.22 (0.27) 0.25 (0.32) 0.51 (0.33) Koch, I., & Allport, A. (2006). Cue-based preparation and stimulus-based priming of tasks
sν 0.13 (0.23) 0.77 (0.62) 1.63 (1.71) 0.55 (0.70) 1.28 (0.91) 1.03 (0.84) in task switching. Memory & Cognition, 34(2), 433–444.
st0 .164 (.111) .075 (.068) .091 (.070) .125 (.104) .116 (.069) .133 (.071) Koch, I., Gade, M., Schuch, S., & Philipp, A. M. (2010). The role of inhibition in task
switching: A review. Psychonomic Bulletin & Review, 17(1), 1–14, http://dx.doi.org/
Note. Parameters of the diffusion model: a = response criterion, ν = drift rate, t0 = non-
10.3758/PBR.17.1.1.
decision parameter, sz = inter-trial variability of the starting point, sν = inter-trial vari- Koch, I., Lawo, V., Fels, J., & Vorlander, M. (2011). Switching in the cocktail party: Explor-
ability of the drift rate, st0 = inter-trial variability of the non-decision parameter. Trial ing intentional control of auditory selective attention. Journal of Experimental
types: TS/CS = task switch/cue switch, TR/CS = task repeat/cue switch, TR/CR = task Psychology: Human Perception and Performance, 37(4), 1140–1147.
repeat/cue repeat. Logan, G. D., & Bundesen, C. (2003). Clever homunculus: Is there an endogenous act of
control in the explicit task-cuing procedure? Journal of Experimental Psychology:
Human Perception and Performance, 29(3), 575–599.
References Logan, G. D., & Bundesen, C. (2004). Very clever homunculus: Compound stimulus strat-
egies for the explicit task-cuing procedure. Psychonomic Bulletin & Review, 11(5),
Allport, A., Styles, E. A., & Hsieh, S. (1994). Shifting intentional set: Exploring the dynamic 832–840.
control of tasks. Attention and performance 15: Conscious and nonconscious informa- Madden, D. J., Spaniol, J., Costello, M. C., Bucur, B., White, L. E., Cabeza, R., et al. (2009). Ce-
tion processing. Cambridge, MA, US: The MIT Press. rebral white matter integrity mediates adult age differences in cognitive perfor-
Allport, A., & Wylie, G. R. (2000). Task switching, stimulus–response bindings, and negative mance. Journal of Cognitive Neuroscience, 21(2), 289–302, http://dx.doi.org/10.1162/
priming. Cambridge, MA: MIT Press. jocn.2009.21047.
Altmann, E. M. (2004). Advance preparation in task switching: What work is being done? Mansfield, E. L., Karayanidis, F., Jamadar, S., Heathcote, A., & Forstmann, B. U. (2011).
Psychological Science, 15(9), 616–622. Adjustments of response threshold during task switching: A model-based func-
Altmann, E. M. (2006). Task switching is not cue switching. Psychonomic Bulletin & tional magnetic resonance imaging study. The Journal of Neuroscience, 31(41),
Review, 13(6), 1016–1022. 14688–14692, http://dx.doi.org/10.1523/jneurosci.2390-11.2011.
Altmann, E. M. (2007). Comparing switch costs: Alternating runs and explicit Mayr, U. (2006). What matters in the cued task-switching paradigm: Tasks or cues?
cuing. Journal of Experimental Psychology: Learning, Memory, and Cognition, Psychonomic Bulletin & Review, 13(5), 794–799.
33(3), 475–483. Mayr, U., & Keele, S. W. (2000). Changing internal constraints on action: The role of back-
Arbuthnott, K. D., & Woodward, T. S. (2002). The influence of cue-task association and lo- ward inhibition. Journal of Experimental Psychology: General, 129(1), 4–26.
cation on switch cost and alternating-switch cost. Canadian Journal of Experimental Mayr, U., & Kliegl, R. (2000). Task-set switching and long-term memory retrieval.
Psychology/Revue canadienne de psychologie expérimentale, 56(1), 18–29, http://dx. Journal of Experimental Psychology: Learning, Memory, and Cognition, 26(5),
doi.org/10.1037/h0087382. 1124–1140.
Arrington, C. M., Logan, G. D., & Schneider, D. W. (2007). Separating cue encoding from tar- Mayr, U., & Kliegl, R. (2003). Differential effects of cue changes and task changes on task-
get processing in the explicit task-cuing procedure: Are there ‘true’ task switch effects? set selection costs. Journal of Experimental Psychology: Learning, Memory, and
Journal of Experimental Psychology: Learning, Memory, and Cognition, 33(3), 484–502. Cognition, 29(3), 362–372.
Baddeley, A. D., Chincotta, D., & Adlam, A. (2001). Working memory and the control of Meiran, N. (1996). Reconfiguration of processing mode prior to task performance. Journal
action: Evidence from task switching. Journal of Experimental Psychology: General, of Experimental Psychology/ Learning, Memory & Cognition, 22(6), 1423–1442.
130(4), 641–657. Miyake, A., Friedman, N. P., Emerson, M. J., Witzki, A. H., & Howerter, A. (2000). The unity
Brass, M., & Cramon, D. (2004). Decomposing components of task preparation and diversity of executive functions and their contributions to complex ‘frontal lobe’
with functional magnetic resonance imaging. Journal of Cognitive Neuroscience, task: A latent variable analysis. Cognitive Psychology, 41(1), 49–100.
16(4), 609–620. Monsell, S. (2003). Task switching. Trends in Cognitive Sciences, 7(3), 134–140.
Brown, S. D., & Heathcote, A. (2008). The simplest complete model of choice response Monsell, S., Yeung, N., & Azuma, R. (2000). Reconfiguration of task-set: Is it easier to
time: Linear ballistic accumulation. Cognitive Psychology, 57(3), 153–178. switch to the weaker task? Psychological Research/Psychologische Forschung, 63(3),
De Baene, W., Kuhn, S., & Brass, M. (2012). Challenging a decade of brain research on task 250–264.
switching: Brain activation in the task-switching paradigm reflects adaptation rather Nosofsky, R. M., & Palmeri, T. J. (1997). An exemplar-based random walk model of speed-
than reconfiguration of task sets. Human Brain Mapping, 33(3), 639–651. ed classification. Psychological Review, 104(2), 266–300.
De Jong, R. (2000). An intentional-activation account of residual switch costs. Control of Ratcliff, R. (1978). A theory of memory retrieval. Psychological Review, 85(2), 59–108.
cognitive processes: Attention and performance XVIII (pp. 357–376). The MIT Press. Ratcliff, R., & Rouder, J. N. (1998). Modeling response times for two-choice decisions.
Forstmann, B. U., Brass, M., & Koch, I. (2007). Methodological and empirical issues when Psychological Science, 9(5), 347–356.
dissociating cue-related from task-related processes in the explicit task-cuing proce- Ratcliff, R., Thapar, A., & McKoon, G. (2006). Aging and individual differences in rapid two-
dure. Psychological Research, 71(4), 393–400. choice decisions. Psychonomic Bulletin & Review, 13(4), 626–635.
Gade, M., & Koch, I. (2008). Dissociating cue-related and task-related processes in Risse, S., & Oberauer, K. (2010). Selection of objects and tasks in working memory. The
task inhibition: Evidence from using a 2:1 cue-to-task mapping. Canadian Quarterly Journal of Experimental Psychology, 63(4), 784–804.
Journal of Experimental Psychology/Revue canadienne de psychologie expérimentale, Rogers, R. D., & Monsell, S. (1995). Costs of a predictable switch between simple cognitive
62(1), 51–55. tasks. [Peer Reviewed Journal (270); Print]. Journal of Experimental Psychology:
Gilbert, S. J., & Shallice, T. (2002). Task switching: A PDP model. Cognitive Psychology, General, 124(2), 207–231.
44(3), 297–337. Rubinstein, J. S., Meyer, D. E., & Evans, J. E. (2001). Executive control of cognitive processes
Goschke, T. (2000). Intentional reconfiguration and involuntary persistence in task set in task switching. Journal of Experimental Psychology: Human Perception &
switching. In S. Monsell, & J. Driver (Eds.), Control of cognitive processes: Attention Performance, 27(4), 763–797.
and performance XVIII (pp. 331–355). Cambridge, MA, US: The MIT Press. Ruthruff, E., Remington, R. W., & Johnston, J. C. (2001). Switching between simple cogni-
Gotler, A., Meiran, N., & Tzelgov, J. (2003). Nonintentional task set activation: Evidence tive tasks: The interaction of top-down and bottom-up factors. Journal of Experimental
from implicit task sequence learning. Psychonomic Bulletin & Review, 10(4), 890–896. Psychology: Human Perception and Performance, 27(6), 1404–1419.
Hommel, B. (2004). Event files: Feature binding in and across perception and action. Schmiedek, F., Oberauer, K., Wilhelm, O., Suess, H. -M., & Wittmann, W. W. (2007). Indi-
Trends in Cognitive Sciences, 8(11), 494–500. vidual differences in components of reaction time distributions and their relations to
Horoufchin, H., Philipp, A. M., & Koch, I. (2011). The dissipating task-repetition benefit in working memory and intelligence. Journal of Experimental Psychology: General,
cued task switching: Task-set decay or temporal distinctiveness? Journal of 136(3), 414–429.
Experimental Psychology: Human Perception and Performance, 37(2), 455–472. Schmitz, F., & Voss, A. (2012). Decomposing task-switching costs with the diffusion
Jost, K., De Baene, W., Koch, I., & Brass, M. (2013). A review of the role of cue processing in model. Journal of Experimental Psychology: Human Perception and Performance,
task switching. Zeitschrift für Psychologie, 221(1), 5–14. 38(1), 222–250, http://dx.doi.org/10.1037/a0026003.
Jost, K., Mayr, U., & Rösler, F. (2008). Is task switching nothing but cue priming? Evidence Schneider, D. W., & Logan, G. D. (2005). Modeling task switching without switching tasks:
from ERPs. Cognitive, Affective, & Behavioral Neuroscience, 8(1), 74–84. A short-term priming account of explicitly cued performance. Journal of Experimental
Karayanidis, F., Jamadar, S., Ruge, H., Phillips, N., Heathcote, A., & Forstmann, B. U. (2010). Psychology: General, 134(3), 343–367.
Advance preparation in task-switching: Converging evidence from behavioral, brain Schneider, D. W., & Logan, G. D. (2006). Priming cue encoding by manipulating
activation, and model-based approaches. Frontiers in Psychology, 1, http://dx.doi. transition frequency in explicitly cued task switching. Psychonomic Bulletin &
org/10.3389/fpsyg.2010.00025. Review, 13(1), 145–151.
Karayanidis, F., Mansfield, E. L., Galloway, K. L., Smith, J. L., Provost, A., & Heathcote, A. Schneider, D. W., & Logan, G. D. (2007). Task switching versus cue switching: Using
(2009). Anticipatory reconfiguration elicited by fully and partially informative transition cuing to disentangle sequential effects in task-switching performance.
cues that validly predict a switch in task. Cognitive, Affective, & Behavioral Journal of Experimental Psychology: Learning, Memory, and Cognition, 33(2),
Neuroscience, 9(2), 202–215. 370–378.
196 F. Schmitz, A. Voss / Acta Psychologica 151 (2014) 184–196
Steinhauser, M., & Hübner, R. (2006). Response-based strengthening in task shifting: Ev-
idence from shift effects produced by errors. Journal of Experimental Psychology:
Human Perception and Performance, 32(3), 517–534, http://dx.doi.org/10.1037/
0096-1523.32.3.517.
Tukey, J. W. (1977). Exploratory data analysis. Reading, MA: Addison-Wesley.
Vandierendonck, A., Liefooghe, B., & Verbruggen, F. (2010). Task switching: Interplay of
reconfiguration and interference control. Psychological Bulletin, 136(4), 601–626.
Voss, A., Rothermund, K., & Voss, J. (2004). Interpreting the parameters of the diffusion
model: An empirical validation. Memory & Cognition, 32(7), 1206–1220.
Voss, A., & Voss, J. (2007). Fast-dm: A free program for efficient diffusion model analysis.
Behavior Research Methods, 39(4), 767–775.
Voss, A., & Voss, J. (2008). A fast numerical algorithm for the estimation of diffusion model
parameters. Journal of Mathematical Psychology, 52(1), 1–9.
Wagenmakers, E. -J., van der Maas, H. L. J., Dolan, C. V., & Grasman, R. P. P. P. (2008). EZ
does it! Extensions of the EZ-diffusion model. Psychonomic Bulletin & Review, 15(6),
1229–1235.
Wagenmakers, E. -J., van der Maas, H. L. J., & Grasman, R. P. P. P. (2007). An EZ-diffusion
model for response time and accuracy. Psychonomic Bulletin & Review, 14(1), 3–22,
http://dx.doi.org/10.3758/bf03194023.
Wylie, G. R., Javitt, D., & Foxe, J. (2006). Jumping the gun: Is effective preparation contin-
gent upon anticipatory activation in task-relevant neural circuitry? Cerebral Cortex,
16(3), 394–404.
Yeung, N., & Monsell, S. (2003). Switching between tasks of unequal familiarity: The role of
stimulus-attribute and response-set selection. Journal of Experimental Psychology:
Human Perception and Performance, Vol. 29. (pp. 455–469). American Psychological
Association, 455–469.
Yeung, N., Nystrom, L. E., Aronson, J. A., & Cohen, J. D. (2006). Between task compe-
tition and cognitive control in task switching. The Journal of Neuroscience, 26(5),
1429–1438.