Full text of DOI:10.3758/BF03194955

Memory & Cognition
2002, 30 (4), 540-550
Limitations in advance task preparation:
Switching the relevant stimulus dimension
in speeded same–different comparisons
NACHSHON MEIRAN and HADAS MARCIANO
Ben-Gurion University of the Negev, Beer-Sheva, Israel
When participantsswitchbetweenrelevantstimulus dimensions in speededclassificationtasks, task-
switching cost is reducedby advancepreparation. Previous studies in which speededclassificationtasks
were used have suggestedthat this effectresultsfrom attending to the relevantstimulus dimension. Be-
cause selective attention to the relevant stimulus dimension in same–different judgments is relatively
poor (e.g., Santee & Egeth, 1980), it was predicted that advance task preparation for a shift in the rele-
vant stimulus dimension would be compromised. This prediction was borne out in two experiments
comparing dimension shifts (
vs.
no
) with task rule shifts ( vs.
) and shifts in the
shape
and responses ( – vs.
yes no
fill
same?
different?
mapping of right–left keys to
–
). The resultsindicatethatadvance
yes
no yes
attentional selection of the relevant dimension is an optional preparatory strategy in task switching,
employed only in conditions enabling flexible refocusing of attention.
Typically, task switchingis associatedwith adecrementin
Despite thefact that preparationeffects on switchingcost
switchingcost
performance, called
(e.g., Allport, Styles, & have clear importancefor cognitivecontrol, relativelylittle
Hsieh, 1994;Fagot, 1994;Mayr & Keele, 2000;Monsell& is known abouttheirunderlyingprocessesor boundarycon-
Driver, 2000;Rogers& Monsell,1995;butsee Jersild, 1927, ditions, as compared with other effects in the task-switching
and Spector& Biederman, 1976,as well as Meiran, 2000b, literature. There are few related pieces of evidence.Rogers
forexamplesof conditionswithoutswitchingcost). Formally, and Monsell(1995)found thatadvancepreparationreduced
switching cost is the difference in performance (especially switchingcost only when the preparatoryintervalwas con-
switch
in reaction time [RT] between two types of trials:
trials, in which the task is differentfrom thatinthepreceding According to these authors, the blockingof thepreparatory
nonswitch
stant for a block of trials, and not when it varied randomly.
trial,and
trials,inwhichthetaskis thesame as that interval enabled a flexible, but riskier, strategy of advance
task preparation.GotlerandMeiran(2001)used a paradigm
in the preceding trial.
In this literature, special attention has been given to the in which tasks were ordered randomlyand each trial began
abilityto prepare for a switch. The reason is that several theo- with a task cue. They found that removing the task cue
ries assume thattask informationmust be activatedor spec- upon the presentation of the imperative stimulus increased
ified before task executionbegins (e.g., Logan & Gordon, the effects of preparation on switching cost by increasing
2001; Norman & Shallice, 1986). This ability has tradition- switching cost in the short preparatory interval. Presum-
ally been linked to an empirical marker: the reduction in ably, removing the task cue forced participants to process
switching cost as a result of advance task preparation. This the cue more thoroughlyupon its presentation,rather than
preparation effect is considered to be an important marker delayingpreparation until after the presentation of the im-
of flexiblecognitivetaskcontrol.For example, Allport et al. perative stimulus (e.g., de Jong, 2000; Rogers & Monsell,
(1994), who found negligible reduction in switching cost 1995). This resulted in a greater degree of preparation in
n2
owing to advance preparation, concluded that switching all trials, including trial
1. Therefore, performance in
n
cost doesnotreflect a cognitivecontroloperation.In contrast, the following trial required a greater degree of advance
Rogers and Monsell (1995) and Meiran (1996) found that preparation. Meiran, Chorev, and Sapir (2000; see also
advancepreparationreduced switchingcost and concluded Meiran, 1996, Experiment4) found that a singlesession of
that switching cost reflects the increased demand for con- practice reduced the effects of advancetask preparation on
trol processing in the switch condition.
switchingcostbyreducingswitchingcostin the short prepa-
ration interval. An additional session of practice did not
modulate advance preparation, because it reduced switch-
ing cost to the same degree regardless of whether the
preparatory interval was short or long.
The final example is most relevant to the present study
becauseit concernsa testofMeiran’s(2000a)model.Before
This research was supported by a research grant from the Israeli Sci-
ence Foundationto the first author.We thank Alex Gotler and Roy Luria
for help in programming and data analysis. Correspondenceconcerning
this article should be addressed to N. Meiran, Department of Behavioral
Sciences and Zlotowski Center for Neuroscience, Ben-GurionUniversity
of the Negev, Beer-Sheva, Israel (e-mail: nmeiran@bgumail.bgu.ac.il). describingthe experiment,we will briefly present the model.
Copyright 2002 Psychonomic Society, Inc.
540

LIMITATIONS IN ADVANCE TASK PREPARATION
541
This model describes a particular control strategy used in gruity effects (see Meiran, 2000a, 2000b, for a review of
those task-switching experiments that involve speeded the relevant literature and a list of replicable effects that
classification tasks. It is argued that participants control the model explains).
their response selection by directing their selective atten-
To return to Meiran (2000b), this experiment tested the
tion to the relevant dimension in the target stimulus (see following prediction. Given Meiran’s (2000b) model, ef-
Shalev & Algom, 2000, for recent evidence regarding the ficient response selection can be achieved without the re-
special statusof attentionto dimensions).Specifically, it is focusing of attentional selection, provided that the target
argued that response selection and response activationare stimuli do not contain irrelevant information, so that their
based on an interactionbetween abstract mental represen- single attribute can directly map to the correct keypress.
tations of the target stimuli, as well as between the possi- The experiment involved switching between two tasks.
ble responses (cf. Hommel, 1997). This representation is The tasks involvedthe location of a target stimuluswithin
3
up down
right left
and –
so abstract that the codes being used for stimuli and re- a 2 2 grid. The two tasks were
–
sponses are similar to one another and constitute a com- (Meiran, 1996). In this particular experiment, there were
monrepresentationaldomain(see Hommel, Müssler, Asch- two types of target stimuli, randomly intermixed. One
ersleben, & Prinz, in press, for a review of supporting type was bivalent,containinginformation relevantto both
evidence). There is one notable asymmetry in this strat- tasks (e.g., a target positioned in the upper-left corner of
3
up left
egy: Selective attention does not filter out irrelevant re- the 2 2 grid, which therefore contained both and
sponse information, althoughit filters out irrelevant stim- information).Anothertypeoftargetstimuluswas univalent,
ulus information. Accordingly, the responses become containinginformationrelatedtothe current task only(e.g.,
roughly equally associated with their two possible inter- a target positioned in the upper part of the grid, yet cen-
pretations. Thus, for example, if the experiment involves tered horizontally—thus, one that couldonlybe classified
up
switching between size classification (small–large) and as ). As was predicted,the task-switchingcost was large
shape classification (circle–square), the keypress indicat- and was reduced by advance preparation when the target
circle small
or , depending on which task is cur- stimuli were bivalent, as in previous studies. Presumably,
ing either
rently beingexecuted,is represented by bothof these attrib- the redirection of selective attention was required to filter
utes, with more or less equal weights being given to them. out the irrelevant information in these stimuli. In contrast,
square
The same is true for the key used to indicate either
theswitchingcostwas much smaller and was barely affected
. In contrast to this unbiased/unselected represen- by advance preparation when the target stimuli were uni-
large
or
tation of the responses, the representation of the target valentand attentionalrefocusing was not required for suc-
stimuli is such that the irrelevantinformationis (nearly per- cessful response selection. Presumably, even when the
fectly) filtered out. Accordingly, a small square, for exam- preparatory interval was short, the participantscould pro-
square
ple, is mentally represented as (mostly)
in the con- ceed directly to response selection withoutneeding to first
shape
task.
small
text of the
task and as (mostly)
in the context redirect their attention to the relevant dimension.
Note that Meiran’s (2000a) model describes a particu-
size
of the
The filtering of irrelevant stimulus information is suffi- lar control strategy. This strategy is probablychosen in re-
cientto ensure correct response selection.The reason is that sponse to the constraints afforded in the particular exper-
oncethetargetstimulusis representedas having(nearly) only imental setting. In this setting, selectivelyattending to the
one feature, this singlefeature is sufficient to map the stim- relevant target stimulus dimension is probably the easiest
ulus to the correct response. For example, the feature and least taxing mode of control. However, the strategy is
small
small
would map to the key representing both
and likelyto changein conditionsthat make selectiveattention
circle
large
square
, andnottotheotherkey, representing
and
.
difficult and taxing. This possibility was examined in the
A further assumption in the model is that the redirection present experiments. Accordingly,we explored the ability
of selective attention to the relevant stimulus dimension to prepare for a task switch in same–different judgments,
can be performed before the presentationof the target stim- in which the dimension along which sameness was deter-
ulus, provided that participants are given sufficient time mined was changed between trials. Specifically, the partic-
for advance preparation. However, if this preparatory in- ipants switched between classifying sameness according
shape fill
or
terval is too short, the duration required to redirect selec- to either
.
tive attention is added to the RT. Because attention re- Previous studieshaveshownthat,relativetospeededclas-
focusing is required only in switch trials, redirecting sification,same–different judgments are characterized by
attention is reflected in the task-switching cost. Thus, ac- inefficiency in filtering irrelevant perceptual dimensions
cording to the model, (most of) the reduction in switching (e.g., Santee & Egeth, 1980; but see also Miller, 1978,
cost owing to advance preparation reflects the fact that at- Miller & Bauer, 1981, the references in Santee & Egeth’s
tentional refocusing took place during the preparatory in- study, and also Watanabe, 1988, who found evidence for
terval. Somewhat counterintuitively, the model argues that efficient filtering of irrelevantinformation,but only under
switching between speeded classification tasks does not relatively restricted circumstances). In Santee and Egeth’s
involve the retrieval of relevant stimulus–response map- study, participantswere asked to classifyfigures orto judge
pings. This aspect of the model is required in order to ex- the sameness of pairs of figures according to their shape.
plain the fact that preparation does not reduce task con- Critically, theseresearchers compared a conditionin which

542
MEIRAN AND MARCIANO
the stimuli varied only in shape with conditions in which The Present Experiments
they also varied orthogonallyon the irrelevant dimensions
In the present experiments, we compared three types of
of size and shading, thus requiring filtering of irrelevant operation shifts within the same–different paradigm in
information.The resultsindicatedno interference from the which two stimuli were presentedsimultaneouslyand par-
irrelevant dimensions in the classification task. Specifi- ticipants were required to indicate whether they were the
same different
or . The most importantconditionfor ourpre-
cally, RT was statisticallythe same for the filtering condi-
dimensionshift fill shape
vs.
tion and the conditioninvolvingunidimensionalvariation. dictioninvolved
). This con-
(
In contrast, there was interference in the same–different dition was compared with two control conditions includ-
task rule shift same? different? response
task. Given these differences between tasks, it was pre- ing
dictedthattheparticipants’abilitytopreparefor dimensional
selection in same–different tasks would be compromised.
(
vs.
), and
(assignment of the right and left keys as
no yes
-
mapping shift
yes no
–
, respectively, vs.
–
, respectively).In all cases,
Operationally, thiswould be reflectedina reductionor even shifts were randomly ordered, with a shifting probability
an elimination of preparation effects on switching cost. of .5. It should be noted that, formally, task rule shift and
We alsowished to rule outthepossibilitythatthisreduced response-mapping shift are equivalent, because both re-
preparation effect does not reflect a general property of sult in the same changein the correct overt response.How-
the same–different task or a lack of motivation to prepare ever, we will show evidencethat despitethis formal equiv-
(de Jong, 2000). Therefore, we compared the dimension alence, the participants did not treat the two conditions
shift conditionwith conditionsin which task aspects other equally. Most important,thedifference or lackof difference
than the relevant stimulus dimension were switched. For between response-mapping shift and task rule shift is not
this reason, we will first provide a short review of the lit- crucial for our main predictionconcerningdimensionshift.
erature on preparation for a task switch as a functionof the
switched operation.
It should be further noted that, to be able to conductour
examination, we needed to ensure that there would be a
switchingcostinthesame–differentjudgments.Fortunately,
two studies have shown this already. Proctor and Fisicaro
(1977) presented participants with two stimuli, which
Advance Task Preparation as a Function of the
Switched Operation
In almost all of the articles we reviewed, we found evi- were two stripes varying in color, two circles varying in
dence that the preparation for a switch reduced switching size, or two lettersvaryingin letteridentity. RTswere longer
cost. Thiswas true for a changeinstimulusmodality(Quin- when the three classes of stimuli were intermixed in the
lan & Hill, 1999; but see also Spence & Driver, 1997), a same blockof trials, as compared with when onlyone class
change in stimulus–response mapping between compati- of stimuli was presented in a given block of trials. More
®
®
ble mapping (left left, right right) and incompatible recently, Garcia-Ogueta (1993) presented participants
®
®
mapping(left right, right left; de Jong, 1995), and task with two shapes, presentedeither simultaneouslyor sequen-
switching(e.g., Allport et al., 1994;de Jong,2000;de Jong, tially, and found that mixing trials with simultaneous and
Berendsen, & Cools, 1999; Fagot, 1994; Goschke, 2000; sequential presentation modes resulted in a performance
Mayr & Keele, 2000;Meiran, 1996, 2000a,2000b;Meiran cost. Nonetheless, in neither of these studies were the ef-
et al., 2000).
fects of advance preparation on switching cost examined
We found only two exceptionsfor which preparationdid (althoughProctor & Fisicaro, 1977, studiedthetime course
not reduce switching cost. Both referred to switching per- ofresourcedemand,whichmay be relatedtoadvanceprepa-
ceptual operations. First, Los (1999a, 1999b) asked par- ration).
ticipantsto name digitsthat were visuallydegraded in two
The otheraspect of our studyinvolvescomparingswitch
different ways: added visual noise or segment deletion. types within an experiment. Such comparisons have been
Presumably, these two forms of degradation involve dif- reported by previous investigators (e.g., Allport et al.,
ferent perceptual operations. Los found that shifting be- 1994; Hübner, Futterer, & Steinhauser, 2001; Kleinsorge
tween the two kinds of degraded digits (hence, perceptual & Heuer, 1999;see also Van-Duren & Sanders, 1988), but
operations) resulted in a switching cost. Nonetheless, in- none of these experimentshad findings that were relevant
forming participantsat the beginningof the trial regarding to our predictionsconcerningdifferentialeffects of prepa-
the type of degradation that would appear in the upcom- ration. With respect to preparation,we employed the com-
ing digit did not reduce this switching cost (Los, 1999b). mon procedure, which is based on randomly ordering the
Second, Lamb and his colleagues (Lamb, London, Pond, trials involving the two tasks, providing a task cue at the
& Whitt, 1998; Lamb, Pond, & Zahir, 2000) studied shift- beginningof a trial, and manipulatingthe interval between
cue target
ing between levels of hierarchicalstructure—for example, the task cue and the imperative stimulus, or the
–
A
S
interval
a large composed of small s (Navon, 1977). Partici-
(CTI). In this procedure, the effect of advance
H
S
pants were asked to detect an or an that appearedat ei- preparation on switching cost is indexed by the two-way
ther the local level or the globallevel of the structure. They interactionbetween task switch and CTI. Accordingly, we
found a cost for levelswitching(cf. Ward, 1982). Nonethe- predicted a triple interaction whereby this two-way inter-
less, this cost was affected neither by block-wide ex- action would be modulated by switch-type.
pectancy (Lamb et al., 1998) nor by expectancy resulting
In our experiments, we presented participantswith two
from a cue at the beginningof the trial (Lamb et al., 2000). figures, one on the right side and one on the left side of a

LIMITATIONS IN ADVANCE TASK PREPARATION
543
Accordingly, although the two figures varied on four di-
shape fill size
tilt
shape
fill
and could
mensions,
,
,
, and , only
size
tilt
be relevant;
and
were neverrelevant. Thus, the two
same
figures were to be judged as
with respect to the task-relevant stimulusdimension (e.g.,
same fill
if they were the same
they were judged as
if the relevantdimensionwas
and both figures were empty; see Figure 1) although they
differed along all of the other three irrelevant dimensions
size shape different
tilt
(e.g.,
,
, and ). In other words,
figures
figures
same
differed along all four dimensions, whereas
differed on all but one dimension.
same
The participants performed either the
task (“are
task (“are the fig-
group, the partic-
different
the figures the same?”) or the
dimension shift
ures different?”). In the
ipants shifted between relevant perceptual dimensions.
The relevant dimensions changed randomly from trial to
trial, and each trial began with a verbal cue, instructing
shape
the participantsas to which dimensionwas relevant,
Figure 1. A schematic representation of an exemplar trial be-
ginning with a task cue and followed by a target display. In this
example, the correct answer was yes, because both figures were
empty.
fill
same
or . Forany givenparticipantinthisgroup,thetask (
,
different
) and response mapping were constant, and the
only element that varied randomlybetween trials involved
decision rule shift
the relevant dimension. In the
group,
dif-
task and the
same
the participantsshifted between the
computerscreen. The figures differed alongfourperceptual
dimensions. They were circles or squares (shape), which
were small or large (size) and full or empty (fill), and were
crossed by a vertical line or a horizontal line (tilt; see Fig-
ure 1 for an example). Meiran, Hommel, Bibi, and Lev
(2002, Experiment 3), who employed speeded classifica-
tion tasks, used the same target stimuli as those used in
the present experiments. Each of the four classification
tasks studied by Meiran, Hommel, et al. (2002) involved
ferent
same?
task. The instructional cues were the words
different?
and
. For any given participant,the relevantstim-
ulus dimension and response mapping were constant. Fi-
response-mapping shift
nally, in the
pants shifted between two possibleresponse mappings. In
yes
group, the partici-
one mapping, the key on the right indicated
no
. For any given participant in this
, and the
key on the left indicated . The instructionalcue for this
no yes
mapping was
–
group, the relevant stimulus dimension and the task re-
mained constant.
full empty
vs. ) of a cen-
unidimensionalclassification(e.g.,
trally presented figure, which could vary along four per-
ceptual dimensions. They found that switching tasks was
associated with a substantialswitching cost (140 msec, on
average, when minimal time was allowed for advance
preparation—i.e., the shortest CTI), which was reduced to
about a sixth of its original size (23 msec, on average) by
Method
Participants. Forty-eight students from Ben-Gurion University
of the Negev, Israel, participated as a part of an introductory course
requirement. The 16 participants within each switch type group were
assigned to one of four subgroups, so that within each subgroup, the
allowing a long CTI for advance preparation. Meiran, nonswitched aspects were constant. For example, when response
mappings shifted, there were four subgroups of participants. For one
Hommel, et al.’s (2002) findings were relevant for the
subgroup, the task was same, and the relevant dimension was fill; for
another subgroup, the task was different, and the relevant dimension
present experimentsin that they showed that there is noth-
ing special about the target stimuli used in the present ex-
was fill. Two additional subgroups performed the shape task. The par-
periments that would prevent advance task preparation
ticipants were assigned to the various conditions according to order
from taking place (it should be noted that Meiran, Hom-
mel, et al.’s [2002] results were replicated by Meiran,
Neulinger,& Mayr[2002],using thesame type of task cues
as those used here).
of entry into the experiment.
Stimuli. We used IBM–PCcompatible microcomputers controlled
by software written in MEL language (Schneider, 1988). Target stim-
uli were presented in white on a black background and varied along
four dimensions. The stimulus was either a small/large circle (with
a diameter subtending a visual angle of approximately 1.4º or 3.0º)
or a small/large square (each side subtending 1.4º or 3.0º) that was
either empty (only the circumference depicted in white on black) or
filled (the entire figure filled with a light gray color). A line that sub-
tended 4.5º crossed the figure in its middle and was either horizon-
tal or vertical. One stimulus was presented on the right side of the
screen, and the other was presented on the left, their centers 9.8º
from each other. The instructional cues were the Hebrew equivalents
of fill, shape, same?, different?, yes–no, and no–yes. They were po-
sitioned in the center, between the figures. The height of the words
EXPERIMENT 1
Because we were interested in judgments made on the
basis of a singleperceptual dimension,we tried to lead the
participants to adopt an analytic strategy and to prevent
same
same
re-
their making holistic
comparisons (fast
sponses, if the stimuliare identicalto one anotherin every
respect). For this reason, we used the disjunctive version
of the same–different task (see Farell, 1985, for a review). was 0.5º, and the width ranged between 1.9º and 3.3º. The two fig-

544
MEIRAN AND MARCIANO
ures either were completely different along all four dimensions or less constant rate with increasing CTI. In agreement with
were identical along the task-relevant dimension only and different
on all other dimensions.
F
5
visual inspection, this test was significant [ (2,45)
MS 5
5.49,
2,539], whereas the deviation from linearity
e
Procedure. The participants were tested in two identical sessions,
separated by 1–3 days. Each session comprised 25 warm-up trials
followed by four identical blocks of 100 trials. In each trial, the CTI
(170, 470, 1,470, or 2,970 msec), target stimuli, and relevant rule
changed randomly with equal probability. The relevant rule depended
on switch type. Each trial consisted of presenting the instructional
cue for a variable CTI, followed by the presentation of the figures
along with the instructional cue until the response. After the response
was given, the screen went blank for 1,430 msec. Previous studies
had indicated that switching cost was reduced by increasing the
response–cue interval (e.g., Meiran et al., 2000). Although we did
not perform a preliminary experiment in which the response–cue in-
terval varied, experience with the up–down/right–left paradigm in-
dicated that a response–cue interval of less than 1 sec was sufficient
for most of the reduction in switching cost. Thus, with an interval of
1,430 msec, we assumed that nearly all of the CTI-related reduction
in switching cost reflected advance reconfiguration, rather than the
passive dissipation of the task set that was adopted in trial n21.
F 5
was nonsignificant (
1.23). Notably, the linear com-
ponent of the interaction was also significant when the
nonswitch condition was represented by the first task rep-
F
5
3,168], with a nonsignificant residual inter-
1). These analyses show that the three switch
etition, excluding higher order repetitions [ (2,45)
MS 5
action (
3.37,
Fe,
types differed with respect to the linear rate of cost reduc-
tion, with two groups showing such a reduction, whereas
one group did not.
We also conducteda three-way ANOVA on proportions
of errors (bottomof Figure 2) accordingto group, CTI, and
switch and found that the triple interaction approached
F
5
MS 5
p 5
0.0002, .068],
significance [ (6,135) 2.01,
indicating a similar, albeit noisier,eversion of the pattern
found for RT. Thus, the findings argue against a speed–
accuracy account.
Results and Discussion
Nonswitch RT. Althoughthe present experimentscon-
The first trial in a block and trials preceded by errors or centrated on switchingcost, some conclusionscan also be
by exceedinglylong RTs (3,000 msec) were omitted from drawn with respect to mixing cost. Mixing cost is defined
all analyses. Trials in which the RT was exceedinglylong as the difference in performance (e.g., RT) between the
(3,000msec) were analyzedfor accuracy, butnotfor RT. We nonswitch condition and a condition involving a single
computed the mean RT per condition and submitted these task (Fagot, 1994; Logan & Gordon’s, 2001, concurrence
means to various analyses of variance (ANOVAs).
cost; Meiran et al., 2000; see also Kray & Lindenberger,
Preparation effects on switching cost. We conducted 2000; Los, 1996). The reason one may draw conclusions
3
3
4 mixed model ANOVA with the between- regarding mixing cost is that, if the single-task baseline
a 3
2
subjects independent variable switch type (dimension, de- had been included, it would have been the same for all of
cision rule, or response mapping) and the within-subjects the switch types we studied. Specifically, this single-task
independent variables switch (switch vs. nonswitch) and baseline would have involved a constant dimension, a
CTI (170, 470, 1,470, or 2,970 msec). To save space, we constanttask rule, and a constantresponsemapping.Given
will not report effects that were qualified by higher order the definition of mixing cost, differences in nonswitch
interactions. Alpha level was set at .05 in all analyses.
The triple interaction was statistically significant ences in mixing cost, despitethe fact that the absolutesize
MS 5
RTs among the three switch types would indicate differ-
F
5
[ (6,135) 3.12,
1,909], indicating differences of mixing cost was not determined.Anotherrelevant piece
e
between conditions with respect to the effects of prepara- of evidenceconcerningmixing cost, to be discussed in the
tion on switching cost (Figure 2). To show the source of Task Repetition Effects section below, is whether perfor-
thistriple interaction,we examinedthe simple interactions mance improves as a consequence of repeating the same
between CTI and switch within each switch type. Thissim- task over and over. If task repetition continuesto be effec-
ple interactionwas statistically significant in the decision tive beyond the first repetition, this suggests that mixing
F
5
MS 5
rule group [ (3,135) 11.86,
e
1,909] and in the cost is larger than zero, because the single-task baseline
F
5 MS 5
response-mappinggroup [ (3,135) 4.45,
(indicating that switching cost was reduced by prepara-
F ,
1,909] essentially comprises consistent task repetitions.
The overallcomparison between groupswas significant
MS 5
e
F
5
tion), but not in the dimension group (
1). Because the [ (2,45) 5.16,
228,395]. Follow-up pairwise
e
lack of significancecould reflect low statistical power, we comparisons indicated that dimension shift RTs were
F
5
MS 5
e
increased the power by examining the one degree-of- faster than rule shift RTs [ (1,45) 4.16,
228,395]
228,395),
F
5
MS 5
freedom linear componentof the simple interactionin the and key shift RTs ( (1.45) 10.06,
dimension switch group, and it was nonsignificant as well which did not differ significantly from one another (
e
F 5
F ,
(
1).
1.28). These resultsindicatedthat the smallest mixingcost
Given the fact that the effect of CTI on switching cost, was found in the dimension shift condition.
when found, indicated cost reduction, we examined the Rule shift versus response-mapping shift. The rule
triple interactionbetween switch, switch type, and the lin- shift condition and the key-mapping shift condition were
ear componentofCTI. AlthoughthefunctionrelatingCTIto included mainly for the purpose of comparison, and as
switching cost is generally asymptotic, rather than linear such, any differences between them were not critical to the
(e.g., Meiran et al., 2000), visual inspection suggested present study. Formally, these conditionswere equivalent,
that, in the present experiment, the asymptote had not because the consequences on overt responses were the
been reached and switching cost was reduced at a more or same for thetwo typesof switch. Specifically, changingthe

LIMITATIONS IN ADVANCE TASK PREPARATION
545
Figure 2. Reaction time (RT, in milliseconds) and proportion of errors (PE) as a function of switch
type and cue–target interval (CTI, in milliseconds) in Experiment 1. The uniform 95% confidence
interval for all means is presented once to prevent visual noise. It is useful for evaluating reductions
in switching cost owing to increasing CTI, and based on the MSe of the triple interaction, 613 msec
and 6.004 PE.
same? different?
to
5 different
) or were the same
rule from
reversed the yes–no decision. fered on all dimensions (
5 same
Moreover, each of the participants in that group received along a singledimension(
y one mapping, either yes no no yes
onl or . Therefore, a dicted to result in zero s
). Such a strategy is pre-
ferent
witching cost in dif
-
re
–
–
rule change was equivalent(for overt responses) to chang- sponses, in which the relevant dimension does not matter.
yes no no yes
ing the mapping from, for example,
Nonetheless,this formal equivalencedoes not necessarily a switching cost in
–
to
–
.
However, it is conceivablethat the strategy would produce
same
responses, for which a given di-
imply that the conditionswere psychologicallyequivalent. mension must be attended. To explore this possibility, we
In fact, the results indicated a significant difference in conducted an ANOVA in the dimension shift group only,
witching cost between them, as can be seen in Figure 2
ariables decision (same vs.
including the independent v
s
F
5
MS 5
[ (1,45) 5.26,
the participants did not treat the two conditions equally. msec switching cost for
Was there a shift in the dimension shift condition? 854 msec), which did not differ reliably from the signifi-
same
10,041]. This result suggests that different), CTI, and switch. There was a significant 57-
e
different
judgments (911 vs.
Importantly, in the dimension shift group only, the partic- cant 72-msec cost for
judgments (826 vs. 754 msec;
F 5
ity that the par
ipants could, in principle, ignore the relevant dimension
gether and simply judge whether the two targets dif-
1.49). Thus, the results did not support the possibil-
ticipantssimply looked for a singlesame -
di
alto

546
MEIRAN AND MARCIANO
mension. Additional evidence against this possibility is linear component of this main effect was highly signifi-
MS 5
F
5
the fact that nonswitch RTs were shortest in the dimen- cant [ (1,45) 15.67,
9,389], indicatingan aver-
e
sion shift group. However, if the participantshad checked age reduction of 12 msec per each task repetition,whereas
all the stimulus dimensions, as this strategy suggests, this the deviation from linearity was clearly nonsignificant
F ,
should have slowed them, relative to the other groups for
which only one dimension was to be checked (we thank did not differ significantlybetween groups (
Sander Los for pointing out this to us). results lead to two conclusions. First, because the single-
(
1; see Figure 3). Importantly, the linear component
F ,
1). These
Another variant of the single-task argument just de- task baseline was based on more task repetitionsthan just
scribed is that the source of switching cost in the dimen- four, it would be reasonable to conclude that mixing cost
sion shift group reflects the fact that the participants was positive in all groups. Second, the difference between
searched all of the dimensionson every trial. According to groups in mixing cost was unlikely to be due to differen-
this explanation,switching affected the order in which the tial task repetition effects, because these effects were
dimensions were searched. The results rule out this possi- comparable in size in the present analysis, as was indi-
bility as well, for the following reason. If a serial self- cated by a nonsignificant two-way interaction.
terminatingsearch is assumed, such a strategy is predicted
same
to result in switching cost in
judgments, because the
EXPERIMENT 2
hypothesisis that the dimensionof sameness would be the
first to be searched in nonswitch trials. However, because
The goalof Experiment2 was to replicateand extendthe
findingconcerningreducedeffectsofpreparationonswitch-
ing cost in the dimension shift condition. Such a replica-
tion seems essential,given the null result in this condition.
To increase the generality of the conclusions, two condi-
tions were compared between groups in this experiment.
different
responses require searching all the dimensions,
there shouldnot be a switching cost in such responses. For
this very reason, exhaustive search is not predicted to re-
sult in a switching cost in either response. Because we ob-
same
different
re-
served switching costs in both
and
sponses, this alternative explanation is ruled out as well.
simultaneous presentation condition
The
was a straight-
different
same
responses were slower than
The fact that
forward replicationof the dimension shift conditionin Ex-
F
5
MS 5
responses [ (1,45) 20.75,
26,302] is the com-
e
sequential presenta-
periment 1. It was compared with a
mon finding in disjunctive same–different tasks (e.g.,
Farell, 1985), which provides some validation that strate-
gies were invoked in dealing with the same–different task
that were similar to those found in previous studies in
which there was no switching. Taken together, the present
analysis indicates that the participants treated the dimen-
sion switch as a task switch.
tion
condition, in which, after the task cue was provided,
one of the two randomly chosen targets was presented for
300 msec before the entire display was presented, includ-
ing both targets and the task cue. The fact that Garcia-
Ogueta (1993)found that switchingbetween simultaneous
and sequentialpresentationresulted in a cost suggeststhat
thesepresentationmodesinvokedifferentprocessingstrate-
gies. Moreover, Santee and Egeth (1980, Experiment 3)
also examined selection in sequentially presented stimuli
Task repetition effects. As was mentionedbefore in the
Nonswitch RT section, comparing the groups with respect
to their mean nonswitchRTs indicateda differencein mix-
ing cost. However, we wished to demonstrate that there
was a mixingcost in all three groups.We thereforedecided
to examine whether consecutive task repetitions would
produceperformance gain (Meiran et al., 2000). Given the
fact that a single-taskconditionwould have involvedmore
task repetitions than would mixed-task conditions, such a
result would indicate that performance in the nonswitch
conditiondid not reach that in the single-taskcondition.In
other words, this would imply that mixing cost was posi-
tive. We therefore analyzed the results according to group,
CTI, and consecutivetaskrepetition,whileexcludingswitch
RT to avoidrepeatingthe preceding analyses.Thus, in con-
secutivetaskrepetition,we comparedthefirst, second,third,
and fourth consecutive task repetitions (which, in Rogers
& Monsell’s, 1995, terms are the second through fifth po-
sitionsin the run, because in these terms, the first position
in the run indicates switch trials). To avoid repeating pre-
vious results, we concentrated only on effects involving
size
and found less interference from the
shape
dimension to the
dimension as the interstimulus interval increased.
On the basis of the reasoning so far, this may suggest an
improvementin participants’abilityto efficiently directat-
tentiontotherelevantstimulusdimension.Accordingly, our
aim was to explore the possibilitythat preparation-related
reduction in switching cost would be found with sequen-
tial presentation, but not with simultaneous presentation.
Method
Participants. The 24 participants came from the same population
as did those who took part in Experiment 1. Half of them were as-
signed to each of the two presentation mode groups.
Stimuli and Procedure. The only difference, relative to Exper-
iment 1, was that, for the sequential presentation group, one of the
two targets (half of the time the right target, and half the left target)
was presented first, and after 300 msec the second target was added
to the display. In that group, RT was measured from the presentation
of the second target until the response.
consecutive task repetition. This variable was associated Results and Discussion
F
5
MS 5
with a significant main effect [ (3,135) 7.22,
The results were treated as in Experiment 1, and cell
e
7,235] and did not interact significantly with group. The means were analyzed according to presentation mode (si-

LIMITATIONS IN ADVANCE TASK PREPARATION
547
Figure 3. Reaction time (RT, in milliseconds) and proportion of errors (PE) as a function of switch type and consecu-
tive task repetition in Experiment 1. The uniform 95% confidence interval is useful for evaluating reductions in RT
owing to consecutive task repetitions, and based on the MSe of the (nonsignificant) two-way interaction between con-
secutive task repetition and switch type, 625 msec and 6.01 PE.
multaneousvs. sequential),CTI, and switch (see Figure 4). approachedsignificance—most noteworthy, includingin-
F ,
Only two main effects were significant.The main effect of teractions involving CTI and task switch (all
s
1).
F
5
MS 5
presentation mode [ (1,22) 9.88,
1,600,797]
The results were not as predicted, since the trend of
e
indicated faster responses in sequential presentation means indicated some reduction in switching cost in the
(706 msec) than in simultaneouspresentation(993 msec). simultaneousgroup. However, this trend was quite differ-
This result indicates that the participants took advantage ent from the trend that indexes preparation effects on
of the sequential presentation and processed the first tar- switchingcost. Specifically, these effects are characterized
get before the second target appeared(see Santee& Egeth, by an overadditive interaction between task switch and
1980, for a similar result). In fact, the benefit (287 msec) CTI, whereby the effects of CTI are larger in switch trials
was almost as large as the time allowed for inspecting the than in nonswitch trials. The nonsignificant interaction
first target, before the entire display came up (300 msec). that we observedhere was underadditive:Whereas CTI did
F
5
The secondsignificanteffect was thatof switch [ (1,22)
not have an effect in switch trials, increasing CTI in non-
7,901], indicatinglonger RTs in switch tri- switch trials resulted in slowing. Similar underadditivein-
MS 5
108.72,
e
als (883 msec) than in nonswitch trials (816 msec). Al- teractions usually index the dissipation of the task set
though the main effect of CTI approached statistical sig- adopted in the preceding trial, rather than set preparation.
F
5
MS 5
p 5
19,485, .08; 841, The reasoning behind this interpretationis that nonswitch
nificance [ (3,66) 2.35,
e
832, 863, and 863 msec, in the shortest through longest trials require the same task set as that which had been
CTIs, respectively], none of the other sources of variance adoptedin the preceding trial. If this set dissipates,this re-

548
MEIRAN AND MARCIANO
Figure 4. Reaction time (RT, in milliseconds) and proportion of errors (PE) as a function
of presentation mode and cue–target interval (CTI, in milliseconds) in Experiment 2. In the
sequential presentation, RT was measured from the onset of the second target. The uniform
95% confidence interval is useful for evaluating reductions in switching cost owing to in-
creasing CTI, and based on the MSe of the (nonsignificant) triple interaction, 55 msec and
6.01 PE.
sults in a lesser degree of readiness in nonswitchtrials(see plained by Meiran’s (2000a) model. This model was sup-
Meiran et al., 2000). Accordingly, the fact that this under- ported bothby model fittingand, most importantly, by test-
additivetrend was restricted to the simultaneouscondition ing some of its novel predictions(Meiran, 2000b). One of
may be attributed to the fact that, in the sequential condi- its crucial aspects is the suggestion that the reduction in
tion, RT measurement began300 msec later (after the pre- switching cost owing to advance preparation reflects the
sentationofthesecond target). Thisallowedmore complete redirection of selective attention to the relevant stimulus
dissipation of the task set adopted in the preceding trial. dimension.Thischoiceof controlstrategy probablyresults
In a parallelanalysisonthe proportionof errors (Figure 4, from the relative ease in drawing selective attention to the
bottom), we found only a significant main effect of task relevantstimulusdimensionin speededclassificationtasks.
F
5
MS 5
0.0004], in the same di-
e
switch [ (1,22) 5.41,
Conditionsin which such drawing of attentionbecomes
rection as that found for RT. Thus, the results argue against difficult and taxing were therefore predicted to result in a
a speed–accuracy account.
GENERAL DISCUSSION
changein controlstrategy. In thepresentstudy, we examined
such a condition by capitalizingon the known differences
between speeded classification tasks, in which selective
attention to the relevant stimulus dimension is relatively
Previous research on task switching in which speeded efficient, andspeededsame–different judgments,for which
classification tasks were used led to a series of empirical such selection is far less efficient (e.g., Santee & Egeth,
regularities, which have been reasonably successfully ex- 1980). Accordingly, we predicted that the advance prepa-

LIMITATIONS IN ADVANCE TASK PREPARATION
549
ration for a dimension switch would be compromised in ment). These results argue against the notion of a unitary
same–different judgments. The results of the two experi- task set that is activatedas a singlepackage.They are more
ments supported this prediction.
consistent with a distributed notion of a task set, accord-
We can rule out several alternativeexplanationsof these ing to which the various task aspects may be activatedsep-
results. Specifically,there is nothingspecial about switch- arately or as a unit, depending on the structure of the ex-
ing dimensions in the present set of stimuli that prevents periment (e.g., Hübner et al., 2001; but see also Logan &
advance task preparation. The reason is that other studies Gordon, 2001). If this reasoning is correct, the issue con-
task switch
—that is,
in which dimension switching in speeded classification cerning which instances qualify as a
was explored and the same stimuli were used found large which elements were switched and how many of them—
reductionsin switchingcost owing to advancepreparation becomes rather arbitrary and a matter of theoretical taste.
(Meiran, Hommel, et al., 2002; Meiran, Neulinger, &
The present study (also Meiran, 2000b)indicatesan ad-
Mayr, 2002). In addition, the lack of advance preparation ditional qualification. It seems the various shift task ele-
cannot be attributed to lack of motivation to prepare (de ments might be shifted at different points in time. For ex-
Jong, 2000) or to a general propertywhereby advancetask ample, in Experiment 1, switching was associated with a
preparation is not possible in same– different judgments. shift of a singletask element in all three groups, yet the re-
Thisis becausepreparationresultedin a reductioninswitch- sults dependedon which elementhadbeen switched.When
ing cost in the conditionsinvolvingrule shift or response- the switch was on the relevant dimension, the participants
mapping shift.
delayed focusing of attentionuntilafter the presentationof
An interesting question is how switching was possible the target stimuli (stimulus-cued reconfiguration). When
at all, given the limitation on selective attention, and why the task rule or response mapping was involved, the ele-
switching dimensions produced cost. With respect to the ments were at least partly switched in preparation for the
first part of the question, although the literature has sug- target stimulus during the empty CTI.
gested that there are limitations on selective attention in
The second broad implication concerns the attempt to
same–differentjudgments,these limitationswere not com- identifya generalmodelfor cognitivecontrolin task switch-
plete, because the participantscouldperform theinstructed ing. This approach has characterized most of the previous
task. We suggestthat, given the relativedifficulty of draw- studies on task switching, in which a control strategy that
ing attention to the relevant dimension in same–different had been identified in one experimental setup was tested
judgments,the participantschose to postponethe direction in anothersetup. For example, Allport et al. (1994)studied
ofattentionuntilafter the targetstimulihad been presented. switchingbetweencolornamingand word readingand used
Rogers and Monsell (1995) called this control strategy Stroop stimuli. On the basis of their results, they argued
stimulus-cued reconfiguration
. It implies that the empty that advance preparation does not reduce task-switching
CTI was not used for advancepreparationand, thus, its in- cost. This conclusion had been debated by Rogers and
crease did not lead to performance gain. Hence, according Monsell (1995)and by Meiran (1996), but neitherof these
to this reasoning, part of the switching cost we observed re- studiesusedStrooptaskswitching.Thecomparisonbetween
flectedthetimeit tooktodirectattentiontotherelevantstim- Meiran, Hommel, et al.’s (2002) results and those of the
ulusdimensionafter the target stimulushad been presented. present studysuggeststhatan attempttocharacterizegeneral
Stimulus-cuedreconfigurationis unlikelytofully explain switching processes is unlikely to succeed. A more plau-
the switchingcost, because task repetitionsled to faster re- sible route of inquiry is to first identify the various control
sponses. This, in itself, could contributeto switching cost strategies that participants use in task-switching experi-
because, by definition, switching cost reflects the differ- ments. At the next stage, the research should concentrate
ence between the first and second times in which the task on the situationalconstraintsand their effects on the choice
has been performed. Nonetheless, the observed slope was of control strategies. One control strategy, which we have
12 msec per task repetition, whereas the switching cost in identified in previous research (Meiran, 2000a, 2000b),
2
5
thedimensionshiftgroup(Experiment1) was 868 809
was based on the direction of attentionto the relevantstim-
59msec evenafter excludinghigh-orderrepetitions.In other ulus dimension. The present results suggestthat this strat-
words, we suggest that the switching cost we observed has egy is unlikelytobe chosenin conditionsinwhich attention
at least two components: one owing to task repetition to dimensions is relatively limited and, probably, its redi-
(micro-practice, Rogers & Monsell, 1995; retroactive ad- rection is difficult and taxing. In such conditions, partici-
justment, Meiran, 1996) and another owing to stimulus- pants choose a relatively less taxing and more bottom-up
cued reconfiguration.
strategy of stimulus-cued reconfiguration.
The broaderimplicationsofthepresentresultsare mainly
task switching
two. First, it seems that the notion of
is rel-
REFERENCES
ativelycrudeand shouldthusbe replacedby finer and more
precise terms. For example, the literature has suggested
that switching cost increases with an increasingnumberof
switched task elements (e.g., Hübner et al., 2001; Mayr,
2001; but see also Allport et al., 1994, for a null differ-
ence, and Logan & Gordon, 2001, for a theoretical treat-
Allport, D. A., Styles, E. A., & Hsieh, S. (1994). Shiftingintentional
set: Exploringthe dynamiccontroloftasks. In C. Umiltà & M.Moscov-
itch (Eds.), Attention and performance XV: Conscious and noncon-
scious informationprocessing (pp. 421-452).Hillsdale, NJ: Erlbaum.
de Jong, R. (1995).Strategical determinants of compatibility effects with
task uncertainty. Acta Psychologica, 88, 187-207.

550
MEIRAN AND MARCIANO
de Jong, R. (2000). An intention-activation account of residual switch
costs. In S. Monsell& J. Driver (Eds.), Controlof cognitive processes:
Attention and performance XVIII (pp. 357-376). Cambridge, MA:
MIT Press.
de Jong, R., Berendsen, E., & Cools, R. (1999). Goal neglect and in-
hibitorylimitations: Dissociable causes of interference effects in con-
flict situations. Acta Psychologica, 101, 379-394.
Meiran, N. (2000a).Modelingcognitivecontrolin task-switching. Psy-
chological Research, 63, 234-249.
Meiran, N. (2000b).Reconfigurationof stimulus task-sets and response
task-sets duringtask-switching. In S. Monsell & J. Driver (Eds.), Con-
trol of cognitive processes: Attention and performance XVIII (pp. 377-
400). Cambridge, MA: MIT Press.
Meiran, N., Chorev, Z., & Sapir, A. (2000). Component processes in
task switching. Cognitive Psychology, 41, 211-253.
Fagot, C. (1994). Chronometric investigations of task switching. Un-
published doctoral dissertation, University of California, San Diego. Meiran, N., Hommel,B., Bibi, U., & Lev, I. (2002).Consciousnessand
Farell, B. (1985). “Same”–“different” judgments: A review of current
controversies in perceptual comparisons. Psychological Bulletin, 98,
419-456.
control in task switching. Consciousness & Cognition, 11, 10-23.
Meiran, N., Neulinger,Y., & Mayr, U. (2002). Cue-based differences
in control strategies during task switching: Backward inhibition in
perceptual classification. Manuscript in preparation.
Garcia-Ogueta, M. I. (1993). Internal attentionalswitching:Effects of
predictability, complexity and practice. Acta Psychologica, 83, 13-32. Miller,J. (1978).Multidimensionalsame–differentjudgments:Evidence
Goschke, T. (2000). Intentional reconfigurationand involuntarypersis-
tence in task-set switching. In S. Monsell & J. Driver (Eds.), Control
against independent comparisons of dimensions. Journal of Experi-
mental Psychology: Human Perception & Performance, 4, 411-422.
of cognitiveprocesses: Attentionand performance XVIII (pp. 331-356). Miller,J., & Bauer,D. W. (1981).Irrelevant differences in the “same”–
Cambridge, MA: MIT Press.
”different” task. Journal of Experimental Psychology: Human Per-
ception & Performance, 7, 196-207.
Gotler, A., & Meiran, N. (2001). Cognitive processes underlying a
frontally-mediated component of task-switching. Brain & Cognition,
47, 142-146.
Hommel,B. (1997).Towardanaction-conceptmodel ofstimulus–response
compatibility. In B. Hommel & W. Prinz (Eds.), Theoretical issues in
stimulus–response compatibility(pp. 281-320).Amsterdam: Elsevier.
Hommel, B., Müssler, J., Aschersleben, G., & Prinz, W. (in press).
The theory of event coding (TEC): A framework for perception and ac-
tion planning. Behavioral & Brain Sciences.
Monsell,S., & DriverJ. (Eds.) (2000). Controlof cognitive processes:
Attention and performance XVIII. Cambridge, MA: MIT Press.
Navon, D. (1977). Forest before trees: The precedence of global fea-
tures in visual perception. Cognitive Psychology, 9, 353-383.
Norman, D. A., & Shallice,T. (1986). Attention to action: Willed and
automatic control of behavior. In R. J. Davidson, G. E. Schwartz, &
D. Shapiro (Eds.), Consciousnessand self regulation(Vol 4, pp. 1-18).
New York: Plenum.
Hübner, R., Futterer, T., & Steinhauser, M. (2001). On attentional Proctor, R. W., & Fisicaro,S. A. (1977). Time, capacity, and selection
controlas a sourceofresidualshiftcosts:Evidencefromtwo-component
task shifts. Journal of Experimental Psychology: Learning, Memory,
& Cognition, 27, 640-653.
Jersild, A. T. (1927). Mental set and shift. Archives of Psychology, 14
(Whole No. 89).
Kleinsorge,T., & Heuer, H. (1999).Hierarchical switching in a multi-
dimensional task space. Psychological Research, 62, 300-312.
Kray, J., & Lindenberger, U. (2000). Adult age differences in task
switching. Psychology & Aging, 15, 126-147.
between perceptual attributes. Journal of Experimental Psychology:
Human Perception & Performance, 2, 337-348.
Quinlan, P. T., & Hill, N. I. (1999). Sequential effects in rudimentary
auditory and visual tasks. Perception & Psychophysics, 61, 375-384.
Rogers, R. D., & Monsell, S. (1995). The cost of a predictable switch
between simple cognitivetasks. Journal of Experimental Psychology:
General, 124, 207-231.
Santee, J. L., & Egeth, H. E. (1980).Selective attention in the speeded
classification and comparison of multidimensional stimuli. Percep-
tion & Psychophysics, 28, 191-204.
Lamb, M. R., London, B., Pond, H. M., & Whitt, K. A. (1998). Auto-
matic and controlled processes in the analysis of hierarchical structure. Schneider, W. (1988). Micro Experimental Laboratory: An integrated
Psychological Science, 9, 14-19.
system for IBM PC compatibles. Behavior Research Methods, In-
struments, & Computers, 20, 206-217.
Shalev, L., & Algom, D. (2000). Stroop and Garner effects in and out
of Posner’s beam: Reconciling two conceptions of selective attention.
Journal of Experimental Psychology: Human Perception & Perfor-
mance, 26, 997-1017.
Lamb, M. R., Pond, H. M., & Zahir, G. (2000). Contributionsof auto-
matic and controlled processes to the analysis of hierarchical struc-
ture. Journal of Experimental Psychology: Human Perception & Per-
formance, 26, 234-245.
Logan, G. D., & Gordon, R. D. (2001). Executive control of visual at-
tention in dual-task situations. Psychological Review, 108, 393-434. Spector, A., & Biederman, I. (1976). Mental set and mental shift re-
Los, S. A. (1996). On the origin of mixing costs: Exploring information
processing in pure and mixed blocks of trials. Acta Psychologica, 94,
145-188.
Los, S. A. (1999a). Identifyingstimuli of different perceptual categories
in mixed blocksof trials: Evidence for cost in switching between com-
putationalprocesses. Journalof Experimental Psychology:HumanPer-
ception & Performance, 25, 3-23.
Los, S. A. (1999b).Identifyingstimuli of different perceptual categories
in pure and mixed blocksof trials: Evidencefor stimulus-drivenswitch
costs. Acta Psychologica, 103, 173-205.
Mayr, U. (2001).Age differences in the selection of mental sets: Therole
of inhibition, stimulus ambiguity, and response-set overlap. Psychol-
ogy & Aging, 16, 96-109.
visited. American Journal of Psychology, 89, 669-679.
Spence, C., & Driver, J. (1997).On measuring selective attention to an
expected sensory modality. Perception & Psychophysics, 59, 389-403.
Van Duren, L. L., & Sanders, A. F. (1988). On the robustness of the
additive factors stage structure in blocked and mixed choice reaction
designs. Acta Psychologica, 69, 83-94.
War d, L. M. (1982). Determinants of attention to local and global fea-
tures of visual forms. Journal of Experimental Psychology: Human
Perception & Performance, 8, 562-581.
Watanabe, T. (1988). Effect of irrelevant differences as a function of
the relations between relevant and irrelevant dimensions in the same–
different task. Journal of Experimental Psychology: Human Percep-
tion & Performance, 14, 132-142.
Mayr, U., & Keele, S. W. (2000). Changing internal constraints on ac-
tion:Theroleof backward inhibition.Journalof Experimental Psychol-
ogy: General, 129, 4-26.
Meiran, N. (1996). Reconfiguration of processing mode prior to task
performance. Journalof Experimental Psychology:Learning, Memory,
& Cognition, 22, 1423-1442.
(Manuscript received September 25, 2001;
revision accepted for publication March 26, 2002.)