PS YC HOLOGIC AL S CIE NCE
Research Article
Advance Preparation in Task
Switching
What Work Is Being Done?
Erik M. Altmann
Michigan State University
ABSTRACT—The
preparation effect in task switching is usually
interpreted to mean that a switching process makes use of the
interval between task-cue onset and trial-stimulus onset (the
cue-stimulus interval, or CSI) to accomplish some of its work
ahead of time. This study undermines the empirical basis for this
interpretation and suggests that task activation, not task
switching, is the functional process in cognitive control. Experiments 1 and 2 used an explicit cuing paradigm, and Experiments 3 and 4 used a variation in which the trial after a task
cue was followed by several cueless trials, requiring retention of
the cue in memory. Experiments 1 and 3 replicated the preparation effect on switch cost, and Experiments 2 and 4 showed
that this effect vanishes when CSI is manipulated between subjects, leaving only a main effect of CSI when the task cue is a
memory load.
Given foreknowledge of an upcoming task, people often prepare ahead
of time so that their performance on the task proceeds more smoothly.
For example, one might gather the tools necessary to perform a manual
task, or collect one’s thoughts before giving a lecture. In the latter
case, preparation can be interpreted as activating a set of mental
structures in anticipation of needing them again soon. The more time
invested in such advance preparation (to a point), the faster and less
error prone the resulting performance. This preparation benefit seems
to extend to simpler tasks used to study cognitive control in the laboratory. For example, consider the tasks of classifying a rectangle by
its height (tall or short) or its width (thick or thin). Simple as they are,
these tasks engage several information processing stages, including
perception (e.g., Logan & Bundesen, 2003) and response selection
(e.g., Schuch & Koch, 2003). In task-switching studies, by randomizing which task cue is presented on a given trial (height or width), one
can measure how the system establishes task representations specific
to the current trial. The argument has been made that some of this
processing can take place in a preparatory cue-stimulus interval (CSI)
Address correspondence to Erik M. Altmann, Department of Psychology, Michigan State University, East Lansing, MI 48824; e-mail:
[email protected].
616
separating onset of the task cue (e.g., ‘‘height’’) and the trial stimulus
(e.g., a rectangle).
The question usually asked about this preparatory processing is
how it affects switch cost, the amount by which performance degrades
when the system has to perform a different task on the current trial
than it did on the previous trial. Switch cost is the focus of most taskswitching studies in the literature, reflecting the assumption that it
measures a functional process, intuitively thought of as ‘‘a sort of
mental ‘gear changing’ ’’ (Monsell, 2003, p. 135). From this perspective, it makes perfect sense to study switch cost, and cousins like
‘‘residual’’ switch cost (e.g., De Jong, 2000; Monsell, 2003), and to
manipulate factors like CSI, to probe the nature of the gear-changing
process. In particular, it makes sense that longer preparation intervals
should reduce switch cost, because the gear changing can begin ahead
of time. Thus, it makes sense that the CSI effect on switch cost is
ranked as one of the basic phenomena of task switching (Logan, 2003;
Monsell, 2003).
Despite its intuitive appeal, the gear-changing perspective may not
be the most fruitful way to think about control processes (Altmann,
2003). An alternative approach is to view task switching as a memory
problem (Allport, Styles, & Hsieh, 1994; Allport & Wylie, 2000;
Koch, 2003; Logan, 2003; Mayr & Kliegl, 2003; Sohn & Anderson,
2001; Waszak, Hommel, & Allport, 2003), and in particular to start
with the assumption that the most active task representation in
memory is the one that governs behavior (Altmann, 2002; Altmann &
Gray, 2002; Altmann & Trafton, 2002). From this activation perspective, the function of control processes is to activate the current
task, whatever that might be, rather than to activate a different task
only when a switch is indicated. Thus, in this activation model, the
same basic processes would operate on switch trials and repeat trials
alike, and switch cost, in the end, might reflect an emergent property
of some kind rather than the operation of a functional process. The
possibility that switch cost might represent a side effect, and not a
functional process, leads one to ask whether the CSI effect on switch
cost is really as basic as it seems.
The present study examined preparatory processes in task switching from an activation perspective, using experimental manipulations
of CSI that addressed a methodological bias in existing research
that may have confounded current theoretical interpretations. As it
Copyright r 2004 American Psychological Society
Volume 15—Number 9
Erik M. Altmann
happens, the CSI effect on switch cost is a product largely (exclusively, as far as the author is aware) of within-subjects manipulations
of CSI. This leaves open the possibility that carryover effects, for
example, of the kind examined by Poulton (1982), are somehow affecting performance at any given level of CSI, as a function of the
system’s exposure to other levels of CSI. Indeed, although few studies
have manipulated CSI between subjects, those that have generally
have failed to find a CSI effect on switch cost. For example, neither
Arrington (2002, Experiment 2) nor Koch (2001, Experiments 1 vs. 3)
found an effect of CSI on switch cost, and although Sohn and Anderson (2003) did find an effect, they manipulated task foreknowledge
within subjects, effectively introducing a within-subjects CSI manipulation with a CSI of 0 as one of the levels. In theoretical terms, the
activation view suggests that CSI should in fact affect repeat trials as
well as switch trials, because the work being done involves activation,
not task switching. The current experiments tested this possibility
without methodological bias by manipulating CSI both within and
between subjects.
The specific hypotheses were as follows. First, longer CSIs should
reduce switch cost when CSI is manipulated within participants
(replicating existing results), but not when CSI is manipulated between participants. Second, when CSI is manipulated between subjects, the manipulation should affect switch and repeat trials in
qualitatively the same way, either changing response time (RT) on
both kinds of trials or affecting neither kind of trial. The implications
of this second hypothesis for an activation model of task switching are
drawn out in the Discussion.
EXPERIMENTS
Four experiments were designed by crossing two variables. The first
variable, as already noted, was experimental design. CSI was manipulated within participants in Experiments 1 and 3, in which an
effect on switch cost was expected, and between participants in Experiments 2 and 4, in which no effect on switch cost was expected.
Because the hypothesis was for a null effect in Experiments 2 and 4, a
large number of participants were included, as the basis for statistical
power analyses.
The second variable was whether all or only some trials were cued.
In the two one-trial experiments (1 and 2), a single trial followed each
task cue, and the cue remained perceptually available for the full
duration of the trial. In contrast, in the two multiple-trial experiments
(3 and 4), a run of ‘‘cueless’’ trials followed the initial cued trial,
requiring participants to retain the cue in memory. In the multipletrial experiments, the expectation was that longer CSIs should have a
large effect on RT on the first, cued trial. When a task representation
has to be retained in memory, it seems critical that it be properly
activated in the first place, whereas the system might assign the activating process a lower priority when a cue is always available perceptually. Another goal for the multiple-trial experiments was to
replicate a set of within-run phenomena (within-run slowing, withinrun error increase, and full-run switch cost; Altmann, 2002). Other
researchers have suggested that RTs may decrease across a run
(Gilbert & Shallice, 2002; Salthouse, Fristoe, McGuthry, & Hambrick,
1998), and the hope was to accumulate new evidence on the matter.
The relevance of within-run effects to the activation hypothesis is
addressed in the Discussion.
Volume 15—Number 9
All four experiments used an explicit cuing procedure in which a
task cue appeared first and a trial stimulus second, with an intervening CSI. After trial-stimulus onset, cue and stimulus remained
visible until the participant’s response. In the one-trial experiments,
this response triggered the response-cue interval (RCI). In the multiple-trial experiments, this response triggered a run of contiguous
cueless trials, after which came the RCI.
Method
Participants
A total of 198 Michigan State University undergraduates participated
in the four experiments in exchange for partial course credit. Experiment 1 involved 30 participants and Experiment 3 involved 20.
Experiments 2 and 4 each involved 70, with participants randomly
assigned to the two CSI conditions. Eight additional participants were
excluded because their accuracy was below 90%, the level at which
feedback asked them to be more accurate.
Materials
Each task cue was either ‘‘h’’ (meaning ‘‘height’’) or ‘‘w’’ (meaning
‘‘width’’), presented in white in the center of a dark computer display
in 40-point Monaco font. Each trial stimulus was one of four rectangles formed by combining vertical and horizontal dimensions of 80
and 160 pixels, drawn in gray with a 5-pixel pen, also in the center of
the display. The height task was to judge the vertical height of the
rectangle (tall or short), and the width task was to judge the horizontal
width of the rectangle (thick or thin). QWERTY keys ‘‘Z’’ and ‘‘/’’ were
used for responding in both tasks, with the response-to-key mapping
randomized between subjects.
Design and Procedure
Each session began with an on-line introduction to the tasks and
stimuli, followed by a practice block, with the experimenter present to
answer any questions. Participants were told the total number of
blocks and advised to take breaks if necessary between blocks.
Task cues were presented for a CSI of either 100 ms or 900 ms,
followed by a stimulus, after which cue and stimulus both remained
visible until the response. In Experiments 1 and 2, this response
triggered an RCI equal to 1 s minus the CSI. (The response-stimulus
interval, or the sum of RCI and CSI, was kept constant to control for
priming effects from the previous trial; Meiran, 1996.) In Experiments
3 and 4, this first-trial response was followed by a run of additional
trials presented without task cues; each response triggered onset of the
next trial stimulus. The last trial of the run was followed by an RCI
again equal to 1 s minus the CSI. In Experiments 1 and 3, CSI was
randomized from trial to trial. In Experiments 2 and 4, CSI was manipulated between participants.
After each block, participants received feedback; if they scored
below 90%, they were asked to be more accurate, and if they scored
100%, they were asked to try to go faster. A session contained 25
blocks (including the practice block). In Experiments 1 and 2, a block
consisted of 35 trials. In Experiments 3 and 4, a block consisted of 17
runs of trials, each containing 7 trials on average, with a minimum of 5
and the maximum unbounded. (Run length was computed by adding 5
to a sample from an exponential distribution with l 5 .407, producing
617
Preparation in Task Switching
a nonaging hazard function with no objective information about run
length; see Luce, 1986.)
The first four blocks were excluded from analysis, as were the first
five trials of each remaining block in Experiments 1 and 2 and the first
two runs of each remaining block in Experiments 3 and 4. Included
trials were examined with analyses of variance (ANOVAs), with the
RT data being medians of correct trials. In all four experiments,
ANOVAs were applied to trials immediately following task cues, with
independent variables of CSI (short, long), continuity (switch, repeat),
and task (height, width). In Experiments 3 and 4, separate ANOVAs
evaluated within-run effects, using run position (2 through 9) as an
additional independent variable. (Beyond Position 9, the average
number of observations per subject per cell dropped below five.)
Results
Experiment 1 (One Trial, CSI Manipulated Within Participants)
RTs for Experiment 1 appear in Figure 1 (top left). There were main
effects of CSI, F(1, 29) 5 52.6, p < .001, Z2 5 .65, with short-CSI
trials (686 ms) slower than long-CSI trials (626 ms); of continuity, F(1,
29) 5 71.3, p < .001, Z2 5 .71, with switch trials (711 ms) slower
than repeat trials (601 ms); and of task, F(1, 29) 5 12.3, p < .002,
Z2 5 .30, with the height task (676 ms) slower than the width task
(636 ms). Of possible interactions, only CSI Continuity was reliable, F(1, 29) 5 32.2, p < .001, Z2 5 .53.
Error rates (see Table 1) showed a main effect of continuity, F(1,
29) 5 27.4, p < .001, Z2 5 .49, but no other main effects or interactions were reliable.
Experiment 2 (One Trial, CSI Manipulated Between Participants)
RTs for Experiment 2 appear in Figure 1 (top right). There was no
main effect of CSI, F(1, 68) 5 2.5, p 5 .12, Z2 5 .04, but there were
effects of continuity, F(1, 68) 5 136.8, p < .001, Z2 5 .67, with
switch trials (700 ms) slower than repeat trials (610 ms), and of task,
F(1, 68) 5 29.7, p < .001, Z2 5 .30, with the height task (662 ms)
slower than the width task (587 ms). No interactions were reliable; in
particular, the CSI Continuity interaction was not significant, F(1,
56) 5 1.2, p 5 .27, Z2 5 .02.
A power analysis was conducted to test whether a CSI Continuity
interaction could have been detected. This analysis used the means
from Experiment 1 (to estimate effect size) and variance and intercorrelations from Experiment 2 (the procedure was adapted from
D’Amico, Neilands, & Zambarano, 2001). The power estimate was .97,
well above the usual criterion of .8 (Cohen, 1988; Murphy & Myors,
Fig. 1. Response times on switch and repeat trials as a function of cue-stimulus interval
(CSI). The graphs at the top show results for Experiments 1 and 2, and the graphs at the
bottom show results for Run Position 1 in Experiments 3 and 4. CSI was manipulated
within participants in Experiments 1 and 3 and between participants in Experiments 2
and 4. Error bars are 95% confidence intervals computed with the error term of the CSI
Continuity interaction (Loftus & Masson, 1994, Equation 2).
618
Volume 15—Number 9
Erik M. Altmann
TABLE 1
Error Rates (Percentages) for Experiments 1 and 2 and for Run
Position 1 in Experiments 3 and 4
Experiment
Task and continuity
Height
Switch
Repeat
Width
Switch
Repeat
1
2
3
4
100-ms cue-stimulus interval
4.25
2.13
4.16
2.73
3.70
0.29
2.81
1.28
3.76
2.26
3.49
2.11
3.45
0.63
2.38
1.53
900-ms cue-stimulus interval
Height
Switch
Repeat
Width
Switch
Repeat
3.87
2.67
3.94
2.46
2.64
0.63
3.51
0.98
3.09
2.15
2.87
2.23
2.49
0.74
3.97
1.51
1998). The CSI Continuity Experiment interaction, tested with
Erlebacher’s (1977) procedure, was also reliable, F(1, 97) 5 5.2,
p < .03, Z2 5 .08.
Error rates (see Table 1) showed main effects of continuity, F(1,
68) 5 33.3, p < .001, Z2 5 .33, and task, F(1, 68) 5 7.4, p < .009,
Z2 5 .10. No other main effects or interactions were reliable, Fs < 1.
Experiment 3 (Multiple Trials, CSI Manipulated Within Participants)
RTs for Run Position 1 (the first trial after a task cue) in Experiment 3
appear in Figure 1 (bottom left). There were main effects of CSI, F(1,
19) 5 252.7, p < .001, Z2 5 .93, with short-CSI trials (1,126 ms)
slower than long-CSI trials (864 ms); of continuity, F(1, 19) 5 27.3,
p < .001, Z2 5 .59, with switch trials (1,139 ms) slower than repeat
trials (852 ms); and of task, F(1, 19) 5 8.2, p < .02, Z2 5 .30, with
the height task (1,042 ms) slower than the width task (948 ms). Of
possible interactions, only Continuity CSI was reliable, F(1, 19) 5
12.0, p < .004, Z2 5 .39.
The main effect of CSI was greater than in Experiment 1, F(1,
48) 5 142.7, p < .001, Z2 5 .75.
Error rates for Run Position 1 (see Table 1) showed a main effect of
continuity, F(1, 19) 5 33.8, p < .001, Z2 5 .64, and a CSI Continuity interaction, F(1, 19) 5 4.7, p < .05, Z2 5 .20. No other main
effects or interactions were reliable.
RTs for Run Positions 2 through 9 appear in Figure 2 (top left).
There was a main effect of position, F(7, 133) 5 14.3, p < .001,
Z2 5 .43, with the linear trend accounting for 91% of the variance in
the main effect of position. There was also a main effect of task, F(1,
19) 5 25.7, p < .001, Z2 5 .58, as well as a Continuity Task interaction, F(1, 19) 5 5.6, p < .03, Z2 5 .23, and a Position CSI Task interaction, F(7, 133) 5 2.1, p < .05, Z2 5 .10. No other main
effects or interactions were reliable.
Error rates for Run Positions 2 through 9 also appear in Figure 2
(bottom left). There were interactions of Position Task, F(7, 133) 5
2.1, p 5 .50, Z2 5 .10, and CSI Continuity Task, F(1, 19) 5 4.7,
p < .05, Z2 5 .20. No other main effects or interactions were reliable.
Fig. 2. Response times (top) and error rates (bottom) on switch and repeat trials for Run
Positions 2 through 9 in Experiments 3 and 4. Cue-stimulus interval (CSI) was manipulated
within participants in Experiment 3 and between participants in Experiment 4. Error bars are
95% confidence intervals based on the error term for the continuity effect (up bars) and the
position effect (down bars).
Volume 15—Number 9
619
Preparation in Task Switching
Experiment 4 (Multiple Trials, CSI Manipulated Between Participants)
RTs for Run Position 1 in Experiment 4 appear in Figure 1 (bottom
right). There were main effects of CSI, F(1, 68) 5 12.9, p < .002,
Z2 5 .16, with short-CSI trials (986 ms) slower than long-CSI trials
(801 ms); of continuity, F(1, 68) 5 83.8, p < .001, Z2 5 .55, with
switch trials (995 ms) slower than repeat trials (793 ms); and of task,
F(1, 68) 5 27.6, p < .001, Z2 5 .29, with the height task (937 ms)
slower than the width task (850 ms). No interactions were reliable; in
particular, the CSI Continuity interaction was not significant, F(1,
68) 5 .03, p 5 .86, Z2 5 .00.
The statistical power to detect the CSI Continuity interaction in
Experiment 4 was .85, which again was above the usual criterion of .8,
and the CSI Continuity Experiment interaction was again reliable, F(1, 84) 5 4.1, p < .05, Z2 5 .07.
The main effect of CSI was greater than in Experiment 2, F(1,
136) 5 6.0, p < .02, Z2 5 .04.
Error rates for Run Position 1 (see Table 1) showed a main effect of
continuity, F(1, 68) 5 47.0, p < .001, Z2 5 .41, with switch errors
more frequent than repeat errors. No other main effects or interactions
were reliable.
RTs for Run Positions 2 through 9 appear in Figure 2 (top right).
The main effect of position was reliable, F(7, 476) 5 33.7, p < .001,
Z2 5 .33, with the linear trend accounting for 89% of the variance in
the main effect of position. There was a main effect of task, F(1,
68) 5 131.4, p < .001, Z2 5 .66. No other main effects or interactions
were reliable.
Error rates for Positions 2 through 9 also appear in Figure 2 (bottom
right). The main effect of position was reliable, F(7, 476) 5 2.6,
p < .02, Z2 5 .04, with the linear trend accounting for 46% of the
variance in the main effect of position, and the quadratic trend accounting for 28%. There were main effects of continuity, F(1,
68) 5 7.8, p < .008, Z2 5 .10, and task, F(1, 68) 5 19.9, p < .001,
Z2 5 .23. No other main effects or interactions were reliable.
DISCUSSION
The experimental hypotheses were confirmed. First, the CSI Continuity interaction—the effect of preparation interval on switch cost—
was evident when CSI was manipulated within subjects (Fig. 1, left
panels), but not when CSI was manipulated between subjects (Fig. 1,
right panels). This was true for multiple-trial and one-trial experiments alike, generalizing the finding. Second, when CSI was manipulated between subjects, the manipulation affected switch and repeat
trials the same way: In Experiment 2, the main effect of CSI was small
(42 ms) and not reliable, whereas in Experiment 4, the main effect of
CSI was large (185 ms) and significant. The difference between Experiments 2 and 4—one trial per cue versus multiple trials per cue—
links the presence of a main effect of CSI to the need to retain the task
cue in memory.
These results have two important theoretical consequences. First,
they show that the CSI Continuity interaction depends on the
cognitive system being exposed to multiple CSIs. This dependence on
experimental design indicates that the effect is less general than recent reviews of the task-switching literature have indicated (Logan,
2003; Monsell, 2003), and also provides a new constraint on formal
models of task switching. Switching-process models (De Jong, 2000;
Gilbert & Shallice, 2002; Meiran, 2000; Rubinstein, Meyer, & Evans,
2001; Ruthruff, Remington, & Johnston, 2001; Sohn & Anderson,
620
2001) now have an opportunity to explain why exposure to multiple
CSIs should affect the scheduling priority of the switching process.
Any model that cannot do this through functional considerations can
probably be discounted at a fairly basic level.
Second, these results dissociate the switching process from a preparatory process linked to memory: As Figure 1 shows, when the task
cue had to be retained across a run of cueless trials (Experiments 3
and 4, bottom panels), Position 1 RT was substantially higher than in
the one-trial experiments (Experiments 1 and 2, top panels), and was
reliably affected by CSI. Moreover, this main effect of CSI, unlike the
CSI Continuity interaction, was present when CSI was manipulated
between subjects (Experiment 4). It seems, then, that whereas the
priority of the switching process is variable, the priority of the memory
process is consistently high, at least in the face of varying CSIs. In a
similar dissociation, probability cues affect RT but not switch cost
(Dreisbach, Haider, & Kluwe, 2002). Thus, a question for future work
is whether the preparatory effects found here are related to those
reported by Dreisbach et al., and whether the underlying processes
could ultimately subsume the construct of a switching process, explaining switch cost as an emergent property rather than as an index of
functional activity.
Methodologically, these results show (again) that experimental design can influence even low-level cognitive processing, potentially
leading to a confounding of theory and experimental method. Poulton
(1982; see also Erlebacher, 1977) examined a number of examples in
which theoretical inferences were made from carryover effects, and
the danger is present in the case of task switching, too, if we assume
that the switching process automatically takes advantage of available preparation time. More generally, one might ask whether other
results in the task-switching literature have been the basis of
theoretical inferences when in fact they are closely linked to experimental design.
These results may also indicate a construct-validity problem with
switch cost measured using the popular alternating-runs paradigm
(e.g., Koch, 2003; Monsell, 2003; Rogers & Monsell, 1995; Yeung &
Monsell, 2003). In that paradigm, trials are grouped in runs, as in the
current multiple-trial experiments, but run length is constant, and the
task alternates predictably between runs (e.g., AABBAABB. . .). The
task cue is implicit in the run position of a trial, and although an
external cue is also often provided (see Koch, 2003), storing the cue in
memory may be more efficient than encoding it perceptually on every
trial (e.g., Gray & Fu, 2004). Thus, one model of performance in this
paradigm is that the system activates a task representation on the first
trial of a run, refers to this representation on subsequent trials of the
run, and then activates another representation on the first trial of the
next run. The problem is that, because the task alternates every run,
there are no repeat trials in Position 1. Thus, although switch cost can
be estimated by comparing Position 1 RT with Position 2 RT, this
difference may include the substantial cost of the first trial of any
run—compare, for example, the scale of the ordinate for Position 1 RT
(Fig. 1, bottom panels) with the scale for later positions (Fig. 2, top
panels). Without a Position 1 repeat trial, these first-trial costs cannot
be canceled through subtraction to find switch cost, so estimates of
switch cost may be substantially inflated—though, more to the point,
the alternating-runs paradigm affords no way to know.
Finally, these studies replicate within-run effects (Altmann, 2002;
Altmann & Gray, 2002) with new tasks and materials. Within-run
slowing was apparent in Experiments 3 and 4 (Fig. 2, top panels), and
Volume 15—Number 9
Erik M. Altmann
within-run error increase and full-run switch cost were apparent in
Experiment 4 (Fig. 2, bottom right). (In Experiment 3, with error data
being noisy anyway, and with much lower power than Experiment 4,
no interpretable error patterns were reliable.) Full-run switch cost is
qualitatively distinct from conventional switch cost and its cousins (as
surveyed by Monsell, 2003), because it is manifest only in errors and
spans the full run of trials following a task cue. In the activation
model, it reflects the previous task cue intruding on memory retrieval
during the current run (Altmann, 2002). Also, within-run slowing and
error increase are tied to the substantial Position 1 cost observed here
(and elsewhere; Allport & Wylie, 2000; Altmann & Gray, 2002; Gopher, Armony, & Greenshpan, 2000; Kramer, Hahn, & Gopher, 1999).
These within-run effects reflect gradual decay of the task cue in
memory, which is detrimental locally (in causing performance to degrade) but functional globally in that it prevents buildup of catastrophic proactive interference from old task cues (Altmann, 2002).
For a task cue to decay, it must have activation in the first place,
implying a cost to activating a task cue in memory—not a cue for the
other task, but a cue for any task. A theory-based simulation of these
mechanisms is described in Altmann and Gray (2004).
What work is being done, then, during advance preparation in task
switching? This study has identified a high-priority, task-neutral
process linked to memory retention, which makes good use of preparation intervals regardless of how they are manipulated. This study
has also identified a constraint on the switching process specified in
most task-switching models, which is that its priority is variable; when
the system is exposed to only one level of preparation interval, the
priority of this process drops to the point that, even with 900 ms of
preparation time, all switch cost is ‘‘residual’’ (e.g., De Jong, 2000;
Monsell, 2003; Sohn & Anderson, 2001). Comparing the two processes, one might rank the high-priority process, with the larger effect
on RT, as the one that unambiguously reflects functional activity. The
present results say little about whether the switching process is
functional or not, but one possibility is that it simply reflects a priming
effect (Dreisbach et al., 2002; Ruthruff et al., 2001; Sohn & Anderson,
2001; Sohn & Carlson, 2000) modulating the activating process.
Certainly this reduction is more plausible than the opposite, that the
switching process somehow explains the large main effect of CSI in
the multiple-trial experiments.
In the end, then, this study suggests that a general account of advance preparation in task switching will implicate either (a) a functional switching process plus a task-neutral, memory-linked process
with much larger effects or (b) the latter process by itself, with switch
cost emerging as a side effect, for example, of priming. The latter
model may be more parsimonious, but future theoretical work will
have to demonstrate that an activation model can in fact subsume gear
changing as an explanatory construct.
Acknowledgments—This work was supported in part by Office of
Naval Research Grant N00014-03-1-0063. The author thanks John
Anderson and an anonymous reviewer for suggestions for revision, Rick DeShon for statistical consultations, and Bruce Burns for
comments.
REFERENCES
Allport, A., Styles, E.A., & Hsieh, S. (1994). Shifting intentional set: Exploring
the dynamic control of tasks. In C. Umilta & M. Moscovitch (Eds.), At-
Volume 15—Number 9
tention and performance XV: Conscious and nonconscious information
processing (pp. 421–452). Cambridge, MA: MIT Press.
Allport, A., & Wylie, G. (2000). Task-switching, stimulus-response bindings,
and negative priming. In S. Monsell & J. Driver (Eds.), Attention and
performance XVIII: Control of cognitive processes (pp. 35–70). Cambridge,
MA: MIT Press.
Altmann, E.M. (2002). Functional decay of memory for tasks. Psychological
Research, 66, 287–297.
Altmann, E.M. (2003). Task switching and the pied homunculus: Where are we
being led? Trends in Cognitive Sciences, 7, 340–341.
Altmann, E.M., & Gray, W.D. (2002). Forgetting to remember: The functional
relationship of decay and interference. Psychological Science, 13, 27–33.
Altmann, E.M., & Gray, W.D. (2004). Beyond switch cost: Modeling memory
processes in task switching. Manuscript submitted for publication.
Altmann, E.M., & Trafton, J.G. (2002). Memory for goals: An activation-based
model. Cognitive Science, 26, 39–83.
Arrington, C.M. (2002). Explorations in task space: Similarity effects on task
switching. Unpublished doctoral dissertation, Michigan State University,
East Lansing.
Cohen, J.D. (1988). Statistical power analysis for the behavioral sciences (2nd
ed.). Hillsdale, NJ: Erlbaum.
D’Amico, E.J., Neilands, T.B., & Zambarano, R. (2001). Power analysis for
multivariate and repeated measures analysis: A flexible approach using
the SPSS MANOVA procedure. Behavior Research Methods, Instruments,
& Computers, 33, 479–484.
De Jong, R. (2000). An intention-activation account of residual switch costs. In
S. Monsell & J. Driver (Eds.), Attention and performance XVIII: Control
of cognitive processes (pp. 357–376). Cambridge, MA: MIT Press.
Dreisbach, G., Haider, H., & Kluwe, R.H. (2002). Preparatory processes in the
task-switching paradigm: Evidence from the use of probability cues.
Journal of Experimental Psychology: Learning, Memory, and Cognition,
28, 468–483.
Erlebacher, A. (1977). Design and analysis of experiments contrasting the
within- and between-subjects manipulation of the independent variable.
Psychological Bulletin, 84, 212–219.
Gilbert, S.J., & Shallice, T. (2002). Task switching: A PDP model. Cognitive
Psychology, 44, 297–337.
Gopher, D., Armony, L., & Greenshpan, Y. (2000). Switching tasks and attention
policies. Journal of Experimental Psychology: General, 129, 308–339.
Gray, W.D., & Fu, W.-T. (2004). Soft constraints in interactive behavior: The
case of ignoring perfect knowledge in-the-world for imperfect knowledge
in-the-head. Cognitive Science, 28, 359–382.
Koch, I. (2001). Automatic and intentional activation of task sets. Journal
of Experimental Psychology: Learning, Memory, and Cognition, 27,
1474–1486.
Koch, I. (2003). The role of external cues for endogenous advance reconfiguration in task switching. Psychonomic Bulletin & Review, 10, 488–492.
Kramer, A.F., Hahn, S., & Gopher, D. (1999). Task coordination and aging:
Explorations of executive control processes in the task switching paradigm. Acta Psychologica, 101, 339–378.
Loftus, G.R., & Masson, M.E.J. (1994). Using confidence intervals in withinsubject designs. Psychonomic Bulletin & Review, 1, 476–490.
Logan, G.D. (2003). Executive control of thought and action: In search of the
wild homunculus. Current Directions in Psychological Science, 12, 45–48.
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,
575–599.
Luce, R.D. (1986). Response times: Their role in inferring elementary mental
organization. New York: Oxford.
Mayr, U., & Kliegl, R. (2003). Differential effects of cue changes and task
changes on task-set selection costs. Journal of Experimental Psychology:
Learning, Memory, and Cognition, 29, 362–372.
Meiran, N. (1996). Reconfiguration of processing mode prior to task performance. Journal of Experimental Psychology: Learning, Memory, and
Cognition, 22, 1423–1442.
Meiran, N. (2000). Modeling cognitive control in task-switching. Psychological
Research, 63, 234–249.
621
Preparation in Task Switching
Monsell, S. (2003). Task switching. Trends in Cognitive Sciences, 7, 134–140.
Murphy, K.R., & Myors, B. (1998). Statistical power analysis: A simple and
general model for traditional and modern hypothesis tests. Mahwah, NJ:
Erlbaum.
Poulton, E.C. (1982). Influential companions: Effects of one strategy on another
in the within-subjects designs of cognitive psychology. Psychological
Bulletin, 91, 673–690.
Rogers, R.D., & Monsell, S. (1995). Costs of a predictable switch between
simple cognitive tasks. Journal of Experimental Psychology: General,
124, 207–231.
Rubinstein, J.S., Meyer, D.E., & Evans, J.E. (2001). Executive control of
cognitive processes in task switching. Journal of Experimental Psychology: Human Perception and Performance, 27, 763–797.
Ruthruff, E., Remington, R.W., & Johnston, J.C. (2001). Switching between
simple cognitive tasks: The interaction of top-down and bottom-up
factors. Journal of Experimental Psychology: Human Perception and
Performance, 27, 1404–1419.
Salthouse, T.A., Fristoe, N., McGuthry, K.E., & Hambrick, D.Z. (1998). Relation of task switching to speed, age, and fluid intelligence. Psychology
and Aging, 13, 445–461.
622
Schuch, S., & Koch, I. (2003). The role of response selection for inhibition of
task sets in task shifting. Journal of Experimental Psychology: Human
Perception and Performance, 29, 92–105.
Sohn, M.-H., & Anderson, J.R. (2001). Task preparation and task repetition:
Two-component model of task switching. Journal of Experimental Psychology: General, 130, 764–778.
Sohn, M.-H., & Anderson, J.R. (2003). Stimulus related priming during task
switching. Memory & Cognition, 31, 775–780.
Sohn, M.-H., & Carlson, R. (2000). Effects of repetition and foreknowledge in
task-set reconfiguration. Journal of Experimental Psychology: Learning,
Memory, and Cognition, 26, 1445–1460.
Waszak, F., Hommel, B., & Allport, A. (2003). Task-switching and long-term
priming: Role of episodic stimulus-task bindings in task-shift costs.
Cognitive Psychology, 46, 361–413.
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, 29,
455–469.
(RECEIVED 6/12/03; REVISION ACCEPTED 8/26/03)
Volume 15—Number 9