x... “$33
itﬂcwpﬂumﬂuWXhWWu 5:

7:.
t I I
. Orr»! . . .
u *‘gi 1 |.' . ,
5......aflsax‘...‘ H’ﬁdﬁhhlvixq
. 9.5.:41.IJ«~:3:!I‘ . 3.3-1
‘ .1 .II
, . 3.. . hunch?
Justina
Ifxﬁtl. I. ,
nil-l; ‘
. .

4
. x:

.. .‘
1‘... .. .
“an“... .Qﬁ...i...5s ..
at: :thr

I... v.13
7 ’-

 

. (ﬂ. . .
.mMauw..u.frvwi :

4
.l x. z ,
$15.: {ii}:
.1 15...... .3.
1 .- .I..l.:

9. 3.x. . tr! .q,.;v~.w.m«.nﬂ......f

; 2...: .3.

 

 

THESIS

‘.
r '1 /‘\ n
U/ w’Ivd‘

' LIBRARY
Michigan State
University

 

 

 

This is to certify that the

dissertation entitled

EXPLORATIONS IN TASK SPACE:
SIMILARITY EFFECTS ON TASK SWITCHING

presented by

Catherine M. Arrington

 

has been accepted towards fulﬁllment
of the requirements for

Ph.D. degree in Psychology

 

iﬂh“ C}n»

 

 

Major professor

7/95/01

Date

 

MS U is an Afﬁrmative Action/Equal Opportunity Institution 0-12771

 

 

PLACE IN RETURN BOX to remove this checkout from your record.
To AVOID FINES return on or before date due.
MAY BE RECALLED with earlier due date if requested.

 

DATE DUE DATE DUE DATE DUE

 

 

 

 

 

 

 

 

 

 

 

 

 

 

6/01 c:/CIRC/DateDue.p65vp. 15

 

-h—na _—.-.._. - nu...—

 

EXPLORATIONS IN TASK SPACE:
SIMILARITY EFFECTS ON TASK SWITCHING

By

Catherine M. Arrington

A DISSERTATION

Submitted to
Michigan State University
in partial fulﬁllment of the requirements
for the degree of

DOCTOR OF PHILOSOPHY
Department of Psychology

2002

ABSTRACT

EXPLORATIONS IN TASK SPACE:
SIMILARITY EFFECTS ON TASK SWITCHING

By

Catherine M. Arrington

Task switching paradigms involving rapid transitions between simple but
potentially conﬂicting tasks have been widely used to study executive control processes.
The current research explores the importance of considering the representational task
space within which these executive control processes are acting. In four experiments,
task space is manipulated by varying the similarity among tasks.

Experiment 1 introduces a manipulation of task similarity based on shared
attentional components of the task. Four two-choice discrimination tasks form two pairs
of tasks: height judgments and width judgments; color judgments and brightness
judgments. Within each pair the dimensions of the stimulus that must be attended in
order to respond are similar and previous research shows that attending to one dimension
actually entails attending to the other when the stimulus is encoded. Across pairs, the
dimensions are dissimilar and can be attended independently. Task similarity facilitates
switching between tasks as seen in reduced switch costs when switching between similar
tasks in comparison to switching between dissimilar tasks. Experiment 2 further
examines the task similarity effect demonstrated in Experiment 1 through manipulations
of the timing of trial elements designed to isolate automatic and controlled processes

involved in task switching. The similarity effect disappears when participants are

provided with a longer preparation interval, the time between the task cue telling what
judgment will be performed and the target stimulus, suggesting a controlled process that
establishes attentional control settings.

Experiments 3 and 4 are parallel to the first two experiments, but define task
similarity based on shared response modality—manual or vocal—rather than judged
stimulus dimension. Switching between similar tasks again shows facilitation, indicating
the robustness of the similarity effect across different manipulations of similarity.
Decreases in the similarity effect occur both with increases in delay interval, the time
between the response to the previous task and the onset of the stimulus for the current
task, and preparation interval. These results indicate that both automatic and controlled
processes are acting on the response component of the task set during task switching.

These task similarity effects demonstrate the importance of considering switches
between tasks in the context of the relationships among tasks in task space. Findings
from the current research provide a starting point for building a model of how tasks are
represented relative to one another and how speciﬁc executive control and automatic
processes act on individual cognitive components of task sets when moving from one

point in task space to another.

Copyright 2002 by

Catherine M. Arrington

To my grandparents who bequeathed to me through their children my inquisitiveness and

wonder in the world.

Francis C. Higgins
Catherine B. Higgins
Charles A. Arrington, Sr.

Ottie W. Arrington

ACKNOWLEDGEMENTS

Numerous individuals have contributed to this dissertation either directly or
through their unwavering support of me and I thank them all. Foremost, my special
thanks go to my advisor Torn Carr. His enthusiasm for our research is matched only in
the depth of his caring. I have been blessed with a skilled mentor, outstanding colleague,
and true friend. As members of my doctoral guidance committee, Rose Zacks, Erik
Altmann, and Rick DeShon not only gave guidance in the preparation of the research
presented in this dissertation, but have advised, supported, and challenged me to become
a better researcher. Daniel Bouk and Emily Lauher assisted in data collection and
analysis.

In ways that cannot be counted, the following people have contributed to this
work and to the pleasure of my soul. They have my deepest gratitude. My fellow
graduate students from various areas have provided constant companionship: Karin and
Chris Butler, Doug Davidson, Carrick Williams, Megan Mahoney, Richard Falk, and
Sian Beilock. A number of special women have touched my life during my time at
Michigan State: Denise Carr has cared for, fed, and cheered me, Shari Stockmeyer has
given me daily encouragement, Laura Symonds has inspired me, and the women of the
Monday Morning Group have given me perspective at the start of each week. Finally, I
thank my family who are the source of my strength and my joy: my parents Tony and

Bonnie Arrington and my siblings Elaine, Caleb, and Jane Arrington.

vi

List of Tables ........

TABLE OF CONTENTS

oooooooooooooooooooooooooooooooooooooooooo

List of Figures .................................................

1 Introduction .....
1.1 Task Switching

1.1.1 Methodology .......................................

1.1.2 Controlled

and Automatic Processes in Task Switching .....

1.1.3 Task Characteristics that Inﬂuence Task Switching .........
1.2 Representational Task Space ................................
1.2.1 Task Switching within the Context of a Task Space ........
1.2.2 Analogies from Other Areas of Cognition ................
1.2.3 The Structure of Task Space ...........................

1.3 Task Similarity .

1.4 Overview of the Current Research ...........................

2 Task Similarity as Shared Attentional Control Settings ............

2.1 Experiment] ..
2.1.1 Method .

2.1.2 Results ............................................

2.1.3 Discussion
2.2 Experiment 2 . .
2.2.] Method .

oooooooooooooooooooooooooooooooooooooooooo

oooooooooooooooooooooooooooooooooooooooooo

OOOOOOOOOOOOOOOOOOOOOOOOOOOOOOOOOOOOOOOOO

000000000000000000000000000000000000000000

ooooooooooooooooooooooooooooooooooooooooo

3 Task Similarity as Shared Attentional Control Settings ............

3.1 Experiment3 ..
3.1.1 Method .

3.2 Experiment4 ..
3.2.1 Method .

3.2.2 Results ............................................

3.2.3 Discussion

IIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIII

oooooooooooooooooooooooooooooooooooooooooo
OOOOOOOOOOOOOOOOOOOOOOOOOOOOOOOOOOOOOOOOO
OOOOOOOOOOOOOOOOOOOOOOOOOOOOOOOOOOOOOOOOOO

OOOOOOOOOOOOOOOOOOOOOOOOOOOOOOOOOOOOOOOOO

vii

ix

bNNi-A

11
11
I4
21
25
27

29
33
34
37
44
47
50
52
62
65

67
71
71
74
78
79
80
81
92

4 General Discussion ..........................................
4.1 Task Switching as Movement through Task Space ..............
4.2 Linking Processes to Task Components .......................
4.3 Task Sequence Effects ....................................
4.4 Similarity in Task Switching and Other Dual Task Paradigms .....
4.5 Future Directions .........................................
4.6 Conclusion ..............................................

References

viii

95
95
97
99
101
105
107

108

LIST OF TABLES

Table 1. Mean :t Standard Error for Response Times (A) and Accuracies (B) for

Trial Conditions Based on Task on Trial N and Trial N-l for Experiment 1 ......

Table 2. Mean at Standard Error for Response Times (A) and Accuracies (B) for
Trial Conditions Based on Task on Trial N and Trial N-l Separated by Timing

Condition for Experiment 2 ............................................. 53

Table 3. Mean 3: Standard Error for Response Times (A) and Accuracies (B) for
Trial Conditions Based on Task on Trial N, Trial N-l, and Trial N-2 Separated

by Timing Condition for Experiment 2 .................................... 60

Table 4. Mean :t Standard Error for Response Times (A) and Accuracies (B) for

Trial Conditions Based on Task on Trial N and Trial N-l for Experiment 3 ......

Table 5. Mean at Standard Error for Response Times (A) and Accuracies (B)for
Trial Conditions Based on Task on Trial N and Trial N-l Separated by Timing

Condition for Experiment 4 ............................................. 83

Table 6. Mean d: Standard Error for Response Times (A) and Accuracies (B) for
Trial Conditions Based on Task on Trial N, Trial N-l, and Trial N-2 Separated

by Timing Condition for Experiment 4 .................................... 89

ix

LIST OF FIGURES

Figure 1. Schematic diagram of four tasks from Experiment 1 presented in

a representational task space. Ovals represent the task set for each task.

Overlap between tasks sets indicates a shared task component—attentional

control setting. ....................................................... 32

Figure 2. Example trial line for cued task paradigm used in Experiment 1.
Elements are not presented to scale. ...................................... 36

Figure 3. Mean switch costs when switching from similar and dissimilar tasks

for each task type in Experiment 1. Error bars represent 95% conﬁdence intervals
calculated based on the error term for the interaction of similarity by task as

described in Loﬁus and Masson (1994). ................................... 39

Figure 4. Timing of elements in the trial line for the three groups in Experiments

2 and 4. Group US and Group 7/5 have the same preparation interval, but different

delay intervals. Comparing these conditions equates for the amount of controlled
processing that can take place once the identity of the upcoming task is known, but
varies the amount of time over which decay of activity associated with the task set

from Trial N-l occurs. Group 7/5 and Group 1/11 have the same delay interval but
different preparation intervals. Comparing these conditions varies the amont of
controlled processing while equating for the decay of activity associated with the

task set from Trial N-l. ................................................ 51

Figure 5. Mean switch costs when switching from similar and dissimilar tasks

for each timing condition in Experiment 2. Error bars represent 95% conﬁdence
intervals calculated based on the error term for the similarity effect for AN OVAs
calculated for each between subjects condition separately. ..................... 56

Figure 6. Mean switch costs when switching from similar and dissimilar tasks
for each task type in Experiment 3. Error bars represent 95% conﬁdence intervals
calculated as in Experiment 1. ........................................... 76

Figure 7. Mean switch costs when switching from similar and dissimilar tasks
for each timing condition in Experiment 4. Error bars represent 95% conﬁdence
intervals calculated as in Experiment 2. ................................... 85

1 Introduction

In stimulus rich environments that allow for varied behavioral responses, the
human cognitive system is noteworthy for its ﬂexibility. The coordination of perceptual
input, storage and retrieval of information in memory, and performance of goal directed
behavior occurs through a combination of controlled and automatic processes.
Controlled processes are voluntary, effortful, and goal directed; while automatic
processes are involuntary, effortless, and stimulus driven. The interplay of controlled and
automatic processes has been at the core of several inﬂuential models of cognitive
function (Norman & Shallice, 1986; Posner & Snyder, 1975; Schneider & Shiffrin,
1977). Understanding the complex relationships between these two modes of processing
information is critical to explaining how individuals deal with the demands of an ever
changing environment.

Several models have been put forth specifying when executive control is
necessary for or involved in appropriate performance of goal directed behavior. Norman
and Shallice’s (1986) attention to action model outlined ﬁve types of task situations that
require executive control: 1) planning and decision making, 2) troubleshooting or error
correction, 3) novel or unfamiliar action sequences, 4) dangerous or difﬁcult tasks, and 5)
overcoming habitual or prepotent responses. Under these conditions when tasks cannot
be performed automatically, the necessary executive control is carried out by the
supervisory attention system (SAS). Within a more focused analysis of situations
involving coordination of cognitive functions, Logan (1985) presented a model of
executive control emphasizing the importance of strategy in guiding behavior. He

proposed that executive functions choose among competing strategies, develop and

maintain strategies during task performance, and disable strategies once the behavior has
been completed or is no longer relevant. In the experimental paradigms developed for
studying the interaction between executive control processes and automatic processes in
the laboratory setting, researchers have attempted to capture these conditions—generally
through the use of dual task environments. Recently, there has been increasing interest in

using task switching paradigms to study executive control.

1.1 Task Switching

1.1.1 Methodology

The task switching paradigm involves simple tasks such as identifying a digit or
naming a color that are interleaved in environments that ask for fast reaction.
Performance of these simple tasks is disrupted when a switch from one task to another is
required. This disruption is evident when contrasting the performance of a given task,
say task A, on a trial that follows the performance of a different task B, to the
performance of task A when it follows another trial of task A. The task switch condition
(B-A) results in slower and less accurate performance than the task repetition or no
switch condition (A-A). The difference between these two conditions is referred to as
switch cost. Switch costs are thought to reﬂect the time required for executive control
processes that must be engaged when transitioning between tasks and/or the interference
in performing a task due to recent performance of a different task.

The task switching methodology was ﬁrst introduced by J ersild (1927), but
remained largely unused until the early 19905 when the revival of the methodology led to
an explosion of empirical studies. The basic task switching paradigm involves the

performance of two or more tasks presented in sequential order at a rapid pace. These

tasks are most oﬁen simple categorization tasks such as deciding whether a digit is even
or odd or less than or greater than ﬁve (Sudevan & Taylor, 1987), or deciding whether a
letter is a consonant or vowel (Rogers & Monsell, 1995); identification tasks such as
color naming or word reading using Stroop (1935) stimuli (Allport, Styles, & Hsieh,
1994); or simple mental operations such as addition or subtraction (J ersild, 1927; Spector
& Biederman, 1976). Typically stimuli in the task are multivalent, meaning that the same
set of stimuli can afford several tasks. Such multivalent stimuli produce larger and more
consistent switch cost than univalent stimuli where the stimuli are unique for each task in
the experiment (Meiran, 2000a). Additionally the responses, which typically involve
simple button presses or vocal responses, are often multivalent as well, such that the same
two responses are used to indicate the stimulus values for each task. In these task
environments where stimuli and responses for multiple tasks can be multivalent, the
overlapping nature of the stimulus-response (S-R) mappings results in a degree of
interference or cross talk among tasks, which may be a critical boundary condition for
ﬁnding switch costs (Monsell & Rogers, 1995; Spector & Biederman, 1976).

Since the tasks are interleaved and the multivalent nature of the stimuli does not
uniquely specify the task to be performed, the appropriate task on any given trial must be
speciﬁed in some manner. Researchers have developed a number of methods for
indicating the current task including presenting tasks in a pre-set or alternating order (i.e.
AABBAABB; Rogers & Monsell, 1995), linking a feature of the target display to the
identity of the task such as location or color of the stimulus (Sohn & Anderson, 2001 ), or
providing instructional cues that indicate the appropriate task. Cues can either occur on

each trial (Meiran, 1996; Sudevan & Taylor, 1987) or prior to a run of trials on which the

same task will be performed until another cue appears (Altmann & Gray, 2002). Clearly,
the permutation of these various task elements presents an almost endless variety of
potential task switching experiments, many of which will be discussed in more detail
later.

The task switching paradigm provides an excellent experimental environment for
studying how executive control and automatic processes interact in the coordination of
task performance when an individual is presented with multiple, potentially competing
tasks. Task switching implements a number of conditions theorized to require executive
control such as planning, performing novel behaviors, and switching and maintaining
strategies. Furthermore, the critical focus of task switching is not on the performance of
the individual tasks, but rather on the processes that are engaged when transitions are
made between tasks. In particular, the switch cost measure is calculated by subtracting
task repetitions from task switches. Thus, switch costs are thought to capture processes
associated with transitioning between tasks while removing or subtracting out the time
associated with the processes involved speciﬁcally in task performance. If the controlled
and automatic processes that coordinate task performance are separate from the cognitive
operation executed in the performance of individual tasks, then the switch cost measure
provides an opportunity to assess these processes.

1.1.2 Controlled and Automatic Processes in Task Switching

Out of an increasingly large body of research using the task switching
methodology, a number of proposals have been made to account for the experimental
ﬁndings in terms of the processes thought to underlie task switching. Generally

speaking, these proposed mechanisms may be divided into processes that are either

controlled or automatic. Additionally a second factor that can be used to categorize these
proposals is whether the process results in facilitation or inhibition of a given task set].
Several theories link the time associated with switching tasks to controlled
processes of preparing for the upcoming task. Rogers and Monsell’s (1995) task set
reconﬁguration model was an early and inﬂuential model of task switching that focused
on this preparatory component. Using the alternating tasks procedure and manipulating
the amount of time that elapsed between the response to the previous trial and the onset
of the stimulus for the current trial, they demonstrated a marked decrease in the
magnitude of the switch cost when the time between trials increased. They proposed that
the increase in time between the two tasks allows for reconﬁguration of the task set,
which requires time to carry out. However, they also found that even at relatively long
intervals between trials there remained a cost of switching, which they called residual
switch cost. They accounted for the residual switch cost by positing an additional control
component that is initiated by the onset of the new target stimulus, the so-called stimulus-
cued completion hypothesis. DeJong (2000; DeJong, Berendsen, & Cools, 1999) also
proposed an active preparation account of task switching. However, be attributed the
Footnote 1 In the task switching literature, the term task set has been used in two
distinct, but closely related and oﬁen interchangeable forms tied to two different
deﬁnitions of the word set. The ﬁrst deﬁnition of set is “a state of psychological
preparedness”. Thus task set refers to the state of being prepared to perform a particular
task. The second deﬁnition of set is a “ a number of things. . .that belong or are used
together”. Based on this deﬁnition, task set refers to the group of cognitive operations
that are necessary for performance of a particular task. This second meaning of the
phrase emphasizes the separable nature of task components and the fact that preparing for
a task is multifaceted. Further, task sets have associated levels of activation that
determine which task set will control current behavior. Thus the notion of preparedness
for the upcoming task can be described in terms of the activation level of the task set in

comparison to other available task sets. I will use of the term task set in this second
sense—as a group of cognitive operations.

residual switch cost to a failure to engage the appropriate task set that occurs more often
on switch trials than on no switch trials. Thus, he concluded that residual switch costs
result not from a need to complete task set reconﬁguration on every trial, but from a
higher proportion of trials where task preparation does not begin until the target stimulus
appears.

While providing an important starting point for considering switch costs in terms
of the controlled processes that are involved in preparing for an upcoming task, Rogers
and Monsell gave very little consideration to what the nature of the processes involved in
task set reconﬁguration might be. Several more recent models have attempted to more
speciﬁcally describe these processes. Rubinstein, Meyer, and Evans (2001) proposed two
separate and additive control processes associated with switching between task sets: goal-
shiﬁing and rule-activation. Goal-shifting involves updating working memory, while
rule-activation involves either retrieving from long term memory or activating the
production rules associated with a particular task. Another account of task switching that
focuses on updating of memory processes is Altrnann and Gray’s (2002) model. They
describe task switching within a larger functional decay theory of cognitive control.
Basic to the functional decay model is the concept that task performance is guided by the
memory for the current task and that this memory decays over time. Thus at any point
when an instructional cue is presented—typically but not necessarily associated with
switch trials in the task switching paradigm—a time consuming encoding of the one and
related task set must occur.

Along with Rogers and Monsell, another early and inﬂuential account of task

switching was Allport, Styles, and Hsieh’s (1994) hypothesis of task set inertia.

According to the task set inertia hypothesis, switch costs are the result of proactive
interference of S-R mappings between the previous task and the current task. When tasks
are performed in rapid succession, the activation associated with a recently performed
task decays slowly over time, such that the relevant S—R mappings for the previous task
may interfere with the performance of the current task. In support of the task set inertia
hypothesis, Allport et a1. presented evidence that the nature of the task being switched
from is critical for determining the size of the switch costs. They argued that if switch
costs result from the time required to prepare for the upcoming task then the
characteristics of the upcoming task should inﬂuence the magnitude of the switch costs
rather than the characteristics of the task that has just been completed. Unlike the task set
reconﬁguration hypothesis that focuses on active preparation for the upcoming task, the
dissipation of task set inertia occurs passively indicating an automatic process that
determines costs associated with switching among tasks.

Mayr and Keele (2000) have proposed an additional mechanism that may be
involved in task switching that focuses on active inhibition of the previous task set in
order to overcome the type of interference such as that proposed in the task set inertia
model. Using a paradigm in which three different tasks were presented in a
pseudorandom order, Mayr and Keele examined the effects of task order beyond simple
task switch versus task repetition, which is deﬁned in terms of the task on Trial N and
Trial N-l, by looking at the effect of the task identity on Trial N-2 (lag-2 effects in their
terminology). The key ﬁnding in their experiments was that trials in which the
participant must return to a task that has been recently switched from led to poorer

performance than trials that did not involve a return to a recently abandoned task. Thus,

performance on the second task C in sequences such as C-B-C was slower than
performance of task C in sequences such as A-B-C where all tasks were unique. They
labeled this ﬁnding backward inhibition and suggested that the process of rapidly
switching between tasks is carried out in part by inhibiting the task set for the task that is
being switched from as a way of increasing the difference in activation between the
current task and the previous task. This inhibition of the previous task set then results in
slower reactivation of that task set when the sequence of trials requires a return to a task
soon after the participant has switched from the task.

Recently, a number of researchers have been focused on developing integrated
accounts of the processes involved in task switching. Several theories have proposed
two-component models that include both controlled and automatic processes. A model of
cognitive control in task switching put forth by Meiran (2000b; Meiran, Chorev, & Sapir,
2000) includes mechanisms for both active preparation for the upcoming task and passive
dissipation of interference from the previous task set. This model attempts to reconcile
the task set reconﬁguration and task set inertia theories. In these experiments Meiran
used a cueing paradigm to independently manipulate the preparation interval during
which controlled task set reconﬁguration can be carried out and the overall delay interval
between trials over which automatic interference effects would decay. Using this method
he found evidence supporting both mechanisms.

Sohn and colleagues (Sohn & Carlson, 2000; Sohn & Anderson, 2001) have
proposed another two component model of task switching that includes active task
preparation and passive task repetition effects. They separated the controlled preparation

effects from automatic effects associated with repeating a task by varying whether the

participant had foreknowledge of the identity of the upcoming task. Even when the
upcoming task was not known and therefore controlled preparation for the task could not
be implemented, performance was better if the task repeated across trials. This ﬁnding
suggests a priming from the previous trial to the current trial that beneﬁts performance of
a task when the task is repeated. This model differs from that of Meiran et al. (2000) in
that while both models include a controlled task preparation component, the automatic
component in Sohn’s model proposes a mechanisms for facilitating the upcoming task
through repetition priming, while Meiran et al.’s model suggests that the automatic
processes are associated with interference from previous task sets that inhibit
performance of the current task. In sum, the theoretical work presented thus far suggests
that task switching may be achieved through a combination of controlled or active
preparation for the upcoming task and automatic or passive interactions between the
previously active task set and the new task set.
1.1.3 Task Characteristics that Inﬂuence Task Switching

With the primary focus of task switching experiments on the executive control
processes that are engaged during the time between performance of each task, instead of
the processes engaged in order to carry out the performance of a task, little attention has
been paid to the characteristics of the actual tasks being performed. Indeed as noted
above, a wide range of tasks has been used in previous task switching experiments.
While the particular cognitive demands of speciﬁc tasks and the relationships between
tasks have been largely ignored, evidence does exist that suggests that task characteristics
may inﬂuence the size and nature of switch costs. For example, Mayr and Kliegl (2000)

examined the effect of task difﬁculty on task switching. They found that some

manipulations of task difﬁculty, such as the extent of the retrieval demands, do affect
switch costs. However, other manipulations of task difﬁculty do not affect switch costs,
such as inverting the stimuli thus making the perceptual aspect of task performance more
difﬁcult.

Led by early reports of Allport and colleagues (Allport et al., 1994; Allport &
Wylie, 2000) a number of researchers have studied task properties that result in
asymmetric switch costs—larger switch cost when switching in one direction between
two tasks than when switching in the other direction. Based initially on ﬁndings from
Stroop-like tasks, Allport proposed that differential dominance of a task (word reading
versus color naming in the Stroop task) can affect the efﬁciency of task switching, such
that it is easier to switch to a less dominant task than to a more dominant task. Meuter
and Allport (1999) have found similar asymmetric switch costs when bilinguals switch
between their dominant and non dominant languages. Across these studies, Allport and
colleagues concluded that pre-experimental experience with individual tasks can
inﬂuence and indeed be used to understand the processes involved in task switching.
These conclusions recently have been called into question. Monsell, Yeung, and Azuma
(2000) reviewed the effects of task dominance or relative task strength under a variety of
different manipulations of the construct, including pre-experimental experience, stimulus-
response compatibility, and practice within the experiment. Their survey presented a
more complex picture of differential switch costs, demonstrating conditions in which
asymmetrical switch costs can be eliminated or even reversed. Interestingly these

ﬁndings of asymmetrical switch costs associated with task dominance or relative task

10

strength suggest that not only characteristics of speciﬁc tasks, but also the relationships

among tasks may be critical in determining the processes involved in task switching.
1.2 Representational Task Space

As can be seen from this brief review, researchers have begun investigating how
task characteristics may affect task switching. However, there has been even less
research investigating how the larger task environment inﬂuences switching processes.
This lack of consideration of what I will call “task space” is surprising since the speciﬁc
processes under study in task switching experiments are executive control processes.
These processes are generally conceived of as acting at a level separate from and above
individual task demands. It should be apparent, therefore, that understanding and even
manipulating not only the elements (individual tasks) within the task space, but also the
relationships among those elements is critical to studying executive control processes.
Indeed, these relationships among tasks deﬁne the task space in which executive control
processes act and may prove to impact these processes to a greater extent than the
characteristics of individual tasks taken in isolation.

1.2.1 Task Switching within the Context of a Task Space

A notable exception to the disregard for relationships among tasks is the recent
work of Kleinsorge and colleagues. In a series of studies, they addressed the question of
how the organization and structure of task space inﬂuence the processes of task
switching. Kleinsorge and Heuer (1999; Kleinsorge, Heuer, & Schmidtke, 2001a)
deﬁned a task environment where four potential tasks were formed through an orthogonal
manipulation of the type of judgment to be made on a stimulus, either spatial or

numerical, and the response mapping, either compatible or incompatible. They referred

11

to this set up as a dimensionally organized task space and focused on the hierarchical
nature of the relationships among tasks: task judgment is superordinate to task mapping,
which is in turn superordinate to speciﬁc responses made with either the leﬁ or right
hand. The hierarchy in the task space that they described appears to be based on the
temporal order in which task components or dimensions can be carried out: knowing
which judgment is to be performed logically must occur prior to choosing and making
responses.

In these experiments, numerical stimuli appeared on a display where the position
of the stimulus indicated which judgment was to be made and color of the stimulus
indicated which judgment-to-response mapping was to be used to respond to the current
stimulus. Task performance was analyzed based on whether the trial involved switches
of the judgment portion of the task, the judgment-to-response mapping of the task, or the
speciﬁc responses. The basic ﬁnding was that switches of a higher-order dimension
resulted in automatic switches of subordinate dimensions. For example, if the trial called
for a switch in the judgment being made from a numerical to a spatial judgment, the
performance was fastest if there was also a switch at the level of the judgment-to-
response mapping However, if a switch at the lower level was not called for,
participants’ responding was slowed. They interpret this result as evidence for
generalizing of the switch operation, such that when a switch is carried out on a higher
level of task structure it generalizes and results in switches at lower levels whether they
are indicated by current task demands or not.

In follow-up work, Kleinsorge, Heuer, and Schmidtke (2001b) altered the task

space by expanding the task dimensions such that three judgments and three response

12

mappings were possible for each stimulus. This manipulation resulted not only in an
overall increase in response time, which would be expected as the number of potential S-
R mappings increases, but also in a qualitatively different pattern of data for the different
types of switches when compared to the earlier experiments involving binary task
possibilities. In this more complex task space, where a switch ﬁom a task involved a
choice of switched-to tasks rather than simply a change to the only other task available,
the “generalizing switching operation” identiﬁed in the earlier work no longer occurred.
A switch in the hi gher-order judgment component of the task did not result in a cost
associated with not switching the subordinate task component. Rather, increases in total
numbers of switches at any level in the hierarchy resulted in a monotonic increase in
response time. The exact same shift between two speciﬁc tasks occurred differently
when a third task was introduced into the task environment even when that task was not
involved in the particular pair of trials deﬁning the switch condition. What is clear from
these ﬁndings is the fact that changing the overall task environment (e. g. increasing the
number of potential tasks) appears to result in a qualitatively different pattern of
performance suggestive of a change in the executive control processes governing the
shifts between individual pairs of tasks. Although they do not cast their analysis in terms
of task space, Allport and Wylie (1999, 2000) have evidence in line with the proposal
that increasing the number of tasks available in task space inﬂuences switching among
tasks. In a series of studies, they introduced participants to different tasks performed on
the same set of stimuli in a sequential fashion across phases of the experiment.

Introducing new tasks, and thus manipulating the task space in which the earlier tasks

l3

were presented, resulted in large changes in the switch costs associated with individual
tasks.

While the work of Kleinsorge and colleagues lays some important ground work
for considering the issue of task space in examining executive control processes involved
in task switching, their approach to describing task space is limited. The focus of
Kleinsorge’s description of task space is on the hierarchical logic and temporal order of
task components: does the switch involve a change in judgment or a change in response
mapping? This approach begins to consider how task space might be organized in terms
of relationships among tasks that can be described in terms of components of the task
themselves. However, Kleinsorge considers only the order of the elements in the task set
as the manipulation of the task space, asking whether the changes in the task set are a
result of components that occur early or late in the performance of the task. While the
order of components in a task set is certainly one important way of deﬁning task space,
order is only one dimension of what will obviously be a complicated set of inter-task
relationships. Thus this approach is only a start toward providing a full model of task
space that might prove useful in furthering our understanding of the executive control
processes that act within this space.

1.2.2 Analogies from Other Areas of Cognition

In light of the limited consideration of representational task space in the literature
on task switching and executive control, drawing on analogies from other areas of
cognition is useful. Several highly researched cognitive processes and the
representational environments in which they occur may provide a framework for building

an understanding of how task space might be expected to inﬂuence executive control

14

processes. Two such areas of study are orienting of visual attention and retrieval of
information from semantic memory.

Orienting of visual attention. An analogy between orienting of attention in visual
displays and executive control processes involved in task switching is compelling.
Indeed, the very idea of task switching conjures images of movement or translation from
one task set to another in much the same way that orienting moves attention from one
location in visual space to another. Cashing out this analogy, the attentional mechanisms
involved in orienting map to the executive control processes active in task switching,
while the feature and spatial parameters of the visual display map to tasks in task space.
Consider how such an analogy might serve to focus the somewhat jumbled theoretical
models of executive control. Posner’s (1980) simple model of visual orienting includes
three components: disengaging attention from the current focus, shiﬁing attention to a
new location, and engaging attention at the new location. The description of executive
control processes involved in task switching might easily be placed in a parallel form:
disengage executive attention from the current task set, switch executive attention to the
new task set, and engage executive attention at the new task set. Using such a model as a
starting point provides structure for considering the various effects currently pouring
from the laboratories investigating task switching. For instance, disengaging a task set
may include active inhibition such as the phenomenon of backward inhibition described
by Mayr and Keele (2000). Other effects may best be thought of as inﬂuencing the time
that it takes to engage a new task set, such as Rogers and Monsell’s (1995) stimulus-cued
completion hypothesis, which explains residual switch costs in terms of a process that

cannot occur until the stimulus for the upcoming task appears. Meiran’s (2000a)

15

demonstration that the complexity of the response mapping associated with an upcoming
task affects residual switch costs might be thought of as a manipulation that varies the
time necessary to engage the upcoming task set. This analogy is not completely unused
in the task switching literature. In particular, the idea that task sets are engaged and
disengaged as part of the task switching process has shown up in discussions of task
switching (Allport & Wylie, 1999). Further this analogy may offer new approaches to
considering task switching. For example, the idea that one aspect of task switching may
involve a shift of executive attention through a task space might be suggested by this
analogy. This idea will be considered more fully below.

The second aspect of the analogy between attentional orienting and executive
control attempts to explicate task space by comparing it to visual space. Two known
effects of features of visual displays may be informative as to how to begin carving up
and measuring task space: distance effects and grouping effects. One easily apparent
aspect of visual displays is the distance between elements in the display, both in plane
and in depth. Researchers studying both overt and covert shifts of attention have
manipulated distance in order to measure the time necessary to shift attention as a
function of the distance between elements in the display (Tsal, 1983; but also see Kwak,
Dagenbach, & Egeth, 1991). If task space is to be thought of in analogy to visual space,
then manipulations of distance in task space may provide insight into the executive
control processes that act within task space. However, unlike visual space where there is
a clear three-dimensional Euclidean metric for distance, ﬁnding, manipulating, and
measuring distance in what will surely be a multidimensional task space may prove a

difﬁcult problem.

16

Grouping of elements in visual displays impacts the processing of these elements.
Recently it has been clearly demonstrated that such grouping factors affect attentional
processes. Kramer and Jacobson (1991) examined the effects of grouping on the
distribution of attention in a visual display. Distractor lines that were grouped with a
target line based on shared color or connectivity resulted in greater compatibility
effects—deﬁned by faster responding when the distractors were the same as the target
compared to when the distractors were different from the target—than distractor lines that
were grouped separately from the target. These results indicate a greater degree of
attention to distractors when they were grouped with the target line. Other research
suggests that locations in visual displays can be grouped within object boundaries and
that shifts of attention from one location to another within an object occur more easily
than shifts of attention between locations in different objects (Egly, Driver, & Rafal,
1994). By analogy, task space may also be prone to effects of grouping whereby tasks
sharing a common feature may have overlapping processing much as the distractor and
target lines in the Kramer and Jacobson study, or tasks that are grouped within some
common boundary (or region of task space) may be easier to switch between. These
examples are clearly speculative and the analogy is likely not to completely capture the
complexities of executive control in task switching. However, the link between processes
of attentional orienting and executive control may serve as a reasonable starting point for
trying to understand executive control in task switching environments, in particular, the
effects that manipulations of task space have on implementation of these control

mechanisms.

17

Retrieval from semantic space. Models of semantic memory make up another
highly studied area of cognitive psychology that may instruct a conceptualization of task
space and the processes that operate on and within it. Unlike visual attention where the
spatial representation has a clearly Euclidean metric, deﬁning semantic space is a more
difﬁcult problem and may thus make a better analogy to the idea of task space, which is
likely to be at least as complex.

During the early 1970’s, several theories describing the representational space
underlying semantic memory were developed. One of the earliest and most inﬂuential of
these models was the spreading activation model of Collins and Quillian (1969), later
expanded by Collins and Loﬁus (1975), which proposed a hierarchical network
organization of semantic memory. Within this model, semantic space is represented as a
series of nodes (concepts) that are connected by links that describe the relationships
between nodes. These links also provide a loose measure of semantic space in that links
between more highly related semantic items are conceived of as being shorter, evidenced
by faster response times for related items in sentence veriﬁcation and semantic priming
tasks.

In contrast to the hierarchical network models are feature models, which propose
that concepts within semantic memory are represented in terms of feature lists. The
similarity relations among concepts are captured by their relative locations in a
multidimensional space deﬁned by the dimensions from which feature values are drawn.
The best known of these models is that of Smith, Shoben, and Rips (1974). They
proposed that storage of semantic items involves not only the individual lexical word

forms, but also lists of semantic features including deﬁning and characteristic features.

18

These feature models provide a far simpler, or at least more systematic, structure for the
representation of semantic memory than do network models. However, this is coupled
with a more complex set of processes involved in retrieving and comparing semantic
items in memory (see Kintsch 197 8 for a review of these two classes of models). In
analogy to task space, the organization of semantic memory in terms of a list of features
corresponds rather nicely to an organization of tasks within task space in terms of a list of
task components. Such a model of task space implies a structure to task space where
shared or similar components between tasks may inﬂuence how shifting between tasks
occurs.

In addition to describing the representation of semantic concepts, these and other
researchers investigated the processes involved in retrieval of semantic memories from
this representational space. Just as orienting mechanisms may serve as an analogy to
executive control in task switching, so might the processes of retrieving semantic
information from different locations in semantic space. Early work with the lexical
decision task by Schvaneveldt and Meyer (1973) serves as a good example. They
compared two models of how retrieval from semantic memory is affected by the
presentation of associated words: a spreading activation model where semantic items
receive excitation when an associated or “nearby” semantic item is retrieved from
memory and a location shifting model where semantic items are retrieved from one
memory location at a time and a shift to a new location requires varying amounts of time
depending on the distance between the two memory locations. The task used to test these
two models involved lexical decision for three simultaneously presented items, such that

participants responded “yes” only if all three items were words. By manipulation of the

19

associations between words in the list and the order of the associated and unassociated
items, the researchers were able to demonstrate that associative beneﬁt is better explained
by a spreading activation model than by a location shifting model.

Drawing on such research may inform development of models of how executive
control works to shift between multiple tasks in a task switching experiment, raising
questions such as whether switch costs result from residual activation that causes
interference between competing task sets or from the time necessary to shift between the
locations of two tasks in representational task space. The rich body of theoretical and
experimental work on semantic priming that followed the early work just reviewed offers
numerous ﬁndings that may beneﬁt consideration of the executive control processes
engaged in task switching. An obvious example is Posner and Snyder’s (1975) proposal
that while automatic priming results from spreading activation ala Schvaneveldt and
Meyer (1973), an additional source of controlled priming based on conscious
expectations arises ﬁom location shifting—explicit movement of attention from one
concept to another. This seminal idea stimulated Neely’s (1977) examination of the
relationship between controlled and automatic inﬂuences on priming as a result of
expectancy and timing of elements in the priming task. Several current models of task
switching evoke dual processes reminiscent of this work on priming (Meiran et al., 2000;
Sohn & Carlson, 2000). Other work demonstrating the combination of facilitory and
inhibitory processes involved in semantic priming effects (Houghton & Tipper, 1994;
Dagenbach & Carr, 1994) may inform current investigations of facilitation (Allport &
Wylie, 1999; Sohn & Carlson, 2000) and inhibition (Mayr & Keele, 2000) in task

switching.

20

In sum, while the notion of a representational task space has been largely
unstudied, the development of an understanding of task space may beneﬁt from
consideration of other areas of study where more work has been devoted to questions of
how representational space is organized and the impact that it has on the cognitive
processes that act within it. I have brieﬂy reviewed issues of orienting of visual attention
and structure and retrieval of semantic memory in order to present some possible
analogies to how task space might be represented and the mechanisms by which
executive control processes may act within this task space. Clearly there will not be a
direct mapping between either visual attention or semantic memory and processes
engaged in task switching, rather they should be thought of simply as a starting point for
considering task switching processes within a model of task space. I will turn next to an
examination of the dimensions that might deﬁne a representational task space.

1.2.3 The Structure of Task Space

Two factors deﬁne task space: the speciﬁc task sets that are included in the task
space and the relationships among these individual task sets. Task sets can generally be
thought of as involving all information necessary for performing a particular task. This
view of task sets dovetails nicely with component processing models where cognitive
tasks are thought to require a sequence or combination of separate processes that are
more or less distinct from each other in terms of their activity and implementation and
that can be manipulated independently. Thus task sets are built up from separate
component processes. At a highest level approximation, the task set would include three
major components: perception or encoding of the stimulus, manipulations of or

judgments about the stimulus, and response selection and programming. Additionally

21

there may be some information concerning the circumstances under which the task set
becomes relevant for guiding task performance (e. g. cue to task links).

The representation of tasks as collections of component operations suggests that
executive processes involved in task switching might act on individual components of a
task in different ways and on different time scales. Despite early reports to the contrary,
recent reports provide wide spread support for this view. Early evidence suggested that
switch costs were not affected by the number of task components changing from one task
to another, in contrast to results that might be expected if executive control processes
acted separately on individual task components in a serial or additive fashion. Allport et
al. (1994, Experiment 1) reported that task switches involving one component of a task—
either the dimension of the stimulus display on which the decision was being made or the
nature of the decision being made—resulted in switch costs equal to that of task switches
involving both task components. Recently however, consideration of task sets and the
processes that act upon them in terms of separable components has received support in a
number of experiments. Hiibner, Futterer, and Steinhauser (2001) presented viewers with
“global/local” stimuli where small “local” digits were positioned in a display in such a
way as to form a larger “global” digit. In the different tasks, participants made either
odd/even or less than/ greater than judgments on either the local or global dimension of
the stimulus. Like in the Allport et al. study, task switches could involve either a change
in the dimension of the stimulus or in the nature of the judgment or both. Unlike Allport
et al., the task was cued prior to the onset of the stimulus and the response times for

individual trials were recorded rather than the time to complete the entire list. Htibner et

22

al. consistently found larger switch costs in switches involving two components of the
task set as compared to one component.

A particularly interesting approach to separating task components and switching
processes has been developed by Badre and colleagues (Badre, J onides, Hernandez, Noll,
Smith, & Chenevert, 2000). Their approach investigated differences in switching
between objects held in working memory and between operations that manipulated those
mental representations. Participants began each run with two numbers or counters that
they were to hold in working memory. Over successive trials in a run, instructions were
given to either increment or decrement one or the other of the counters. Based on
instructional cues, the participants switched either the operation being performed
(increment or decrement), the counter on which the operation was being performed, both
the counter and the operation, or neither. Their results indicated separate costs for
switching operations and switching counters (mental representations). Furthermore,
these separate effects were additive in conditions involving both types of switches. This
research cleverly demonstrates that components of the task set have independent effects
on the overall time associated with switching tasks.

Finally, Meiran (2000a) recently proposed a model of task switching that
depended on a division of task sets into “stimulus task sets” and “response task sets”. In
a pair of experiments, he independently manipulated whether stimuli or responses were
univalent or bivalent. The effect of changes to the stimulus set on switch costs varied as
a function of the preparation interval, while changes to the response set affected the
residual switch cost. Taken together the ﬁndings from these and other studies indicate

the importance of considering individual components of the task sets when examining the

23

processes involved when switching from one task to another. The idea that the executive
control processes involved in switching between tasks may depend on the particular
components of the task set that differ between the task will be pursued below.

Individual task sets only partially deﬁne task space. The second aspect of task
space is the relationships among the task sets: how do we organize a group of tasks
within a larger task space that might inﬂuence how we are able to transition among those
tasks? While it is relatively straightforward (though not necessarily simple) to describe
the component operations that make up a task set, it is less clear how the relationships
among tasks should be described. Drawing from the literature on the representation of
concepts in semantic space discussed above, there are two potential ways to conceive of
the relationships among task sets. The ﬁrst would be to consider individual task sets as
nodes in a larger network made up of all task sets. Describing the links between task sets
would then serve to deﬁne task space. However, it is not clear a priori how links between
task sets should be deﬁned or how they are formed.

Alternatively, a representation of task space might be developed based on the
analogy to the feature model of Smith et al. (1974). The feature model approach is
appealing for a number of reasons. As noted above, task sets involve a number of
different components, making an easy analogy to feature lists in semantic memory.
Following from the feature model, the relationships among tasks then can be described as
a function of the task components or cognitive operations shared by each task. If the
mental representation of a task is thought of in terms of the set or group of cognitive
operations that must be carried out or implemented for task performance, then describing

the relationships among tasks in terms of which of these operations are shared across

24

tasks appears to be a straight forward approach. Task space then may be thought of as a
multidimensional space in which individual components of task sets, or cognitive
operations, form the dimensions and a particular task is placed within this task space
based on the cognitive operations involved in performing that task.

Thus having deﬁned task space, the next step in studying the effect that task space
has on the executive control processes acting within that space is to ﬁnd a way to deﬁne
and manipulate task space. There are several potential ways to manipulate task space
either through changes to the task components (the elements included within task space)
or the relationships between tasks that deﬁne the broader task environment. For example,
the number of tasks in task space could be increased or the difﬁculty of speciﬁc tasks
could be altered. The approach I have taken in the current research focuses not on the
characteristics of individual tasks, but rather on the relationships among tasks. I propose
to use the concept of similarity as a measuring stick for the relationships among tasks.

1.3 Task Similarity

The concept of similarity has been widely used within cognitive psychology and
the effects of similarity have been examined in many areas of cognition (Tversky, 1977).
Similarity is primarily conceptualized in terms of either distance based models relying on
methods of multidimensional scaling or shared component models based on methods of
task analysis. Multidimensional scaling procedures have been widely applied to study
cognitive similarity particularly in the areas of perception and categorization (for review
see Ashby & Penin, 1988; cf. Nosofsky & Johansen, 2000). The method of
multidimensional scaling is linked to models of mental representations that emphasize the

structure of the mental space where concepts are stored and evaluated (Keele, 1973).

25

Thus distance in mental space can be deﬁned in terms of the similarity of concepts on
each of multiple dimensions that deﬁne the concepts. Shared component models of
similarity emphasize the importance of shared features of concepts particularly in
circumstances where dimensions may be difﬁcult to specify (Tversky, 1977). Such
shared components models have been more widely used in studies in learning and
transfer of training of motor and other complex skills, where the analysis of task
components is the critical descriptor of tasks (Proctor & Dutta, 1995).

Similarity within task space can be represented in ways that reﬂect both
deﬁnitions of similarity: as shorter distances between two tasks in a multidimensional
task space, or alternatively as shared components between tasks. Tasks may be
conceived of as positioned within a multidimensional task space. The dimensions on
which a given task space are deﬁned represent the individual cognitive operations
necessary to carry out particular tasks. The distance of any two tasks within task space is
determined by the similarity of the tasks on each of the dimensions or cognitive
operations. Thus tasks that have similar or shared operations on a given task dimension
are closer in task space. The shared operations approach to similarity ﬁts nicely with the
feature model approach to task space in which differing cognitive operations involved in
performance of a task describe that task within a larger task space. Similarity can
therefore be operationally deﬁned as shared task components which result in an overlap
of two tasks within task space: for any given component process, tasks increase in
similarity when they share that process and decrease in similarity when they do not. This

is the approach to deﬁning task similarity that I will take in the current line of research.

26

Predictions about the effect that similarity between tasks will have on the
processes involved in switching between tasks fall into two categories. Conventional
wisdom has it that task similarity will interfere with performance of tasks that are
presented close together in time. In reviewing the effect of task similarity in dual task
experiments, Pashler (1998) states that “there is little doubt that task similarity can
exacerbate interference” (p. 294). Additionally, similarity has been linked to
confusability between perceptual stimuli (Ashby & Perrin, 1988). Similarity between
two tasks may result in confusion and interference when switching between tasks.
Within this context, the manipulation of task similarity in the task switching paradigm
might be expected to result in larger switch costs when switching between similar tasks.

On the other hand, the current conceptualization of tasks as a part of a larger task
space and executive control processes involved in task switching as implementing
movement or shifting within task space suggests the opposite effect. Similarity between
two tasks can be conceived of as either decreasing the distance between those tasks
within a multidimensional task space or as grouping those tasks together within task
space. Then by analogy to the visual orienting and semantic priming literature discussed
above, task similarity would be predicted to facilitate switching between two tasks. Just
as decreasing the distance between two targets decreases the time associated with shifting
between targets in visual space, shifting in task space may also occur more easily when

tasks are “nearby”.
1.4 Overview of the Current Research

In the following chapters, I present four experiments that examine the notion of

similarity in task space and its relationship to the processes involved in task switching.

27

The basic experimental design presents multiple tasks in a task switching environment
where the tasks are cued on each trial and occur in a random order. The tasks involve
simple discriminations of multivalent visual stimuli. The task space developed in each
experiment is deﬁned by the relationships between pairs of tasks in the experiment. In
the ﬁrst two experiments, task similarity is deﬁned in terms of shared attentional
components of the tasks—similar tasks involve discrimination of stimulus features that
are processed under the same attentional control setting. Experiment 1 introduces a task
space involving four tasks, divided into two pairs of tasks where the tasks are similar
within a pair but are dissimilar across pairs. Experiment 2 builds on the ﬁndings of a task
similarity effect from Experiment 1, trying to localize the similarity effect to automatic or
controlled processes involved in task switching. Experiments 3 and 4 mirror the ﬁrst two
experiments, but with task similarity deﬁned in terms of response modality rather than

attentional components.

28

2 Task Similarity as Shared Attentional Control Settings

The initial step in developing a task switching experiment suitable for examining
the question of task similarity in task space was to generate a set of tasks that
manipulated similarity. Considering the current deﬁnition of similarity among tasks as
shared cognitive operations, there are clearly multiple ways of deﬁning similarity through
systematic manipulation of tasks such that various cognitive operations are either shared
or not between pairs of tasks. The tasks developed for the ﬁrst two experiments
implemented a deﬁnition of task similarity that was based on a shared attentional
component of the task sets. The tasks involved the presentation of a simple visual
stimulus that could afford multiple two-choice discrimination tasks. The stimuli were
colored rectangles that varied along the dimensions of height, width, color, and
brightness. Each task required the participant to indicate with a simple key press the
value of the stimulus on one of the four dimensions. These dimensions were chosen
based on the fact that they can be grouped into two pairs (height and width; color2 and
brightness) where within a pair the tasks involve attention to a similar perceptual
dimension, while between pairs the tasks involve attention to dissimilar dimensions. The
similarity of these dimensions is rooted in models of selective attention.

Garner (1974) put forward a model of selective attention that categorized
perceptual dimensions as either separable or integral. According to this model, separable
dimensions can be selectively attended, while integral dimensions cannot be selectively
Footnote 2 The use of the term color here describes the stimulus dimension more rightly
referred to as hue. The more generic term color was used when naming the tasks and

providing instructional cues to participants, because it was thought that this would be
simpler for participants who might be unfamiliar with the term hue.

29

attended. Evidence for this distinction came from a large body of psychophysical and
attentional experimentation. Typically such experiments involved a discrimination task
based on a single dimension such as height (Felfoldy, 1974). The stimuli varied not only
on the attended dimension, but also on a second dimension that the participant was
instructed to ignore. The presentation of stimuli then varied such that speciﬁc values of
the secondary dimension were either uncorrelated with values of the primary dimension,
or correlated such that each value of the dimension on which the decision was being
made was always paired with the same value of the secondary dimension. If the
secondary dimension was integral with the primary dimension (e. g. width in the case of
attention to height) then correlated stimulus values resulted in a speed-up of responses on
the discrimination task compared to uncorrelated values. On the other hand, if the two
dimensions were separable (e. g. color in the case of attention to height) then response
times did not vary based on whether the values were correlated across the two
dimensions. Garner interpreted these results as indicating that attending to certain
stimulus dimensions automatically implies attention to and processing of other
dimensions. He referred to these dimensions as integral.

Based on Gamer’s work, the stimulus dimensions for the tasks in Experiments 1
and 2 were chosen such that they formed two pairs of tasks in which the attended
stimulus dimensions are integral within a pair, but separable across pairs. The
operational deﬁnition of similarity in this task space then becomes a shared task
component in terms of attention to the perceptual dimension on which the decision is
being made. Since attention to height results in automatic processing of width and vice

versa, and attention to color results in automatic processing of brightness and vice versa,

30

the task component of attention to perceptual features is common for the two tasks within
a pair—Height and Width, or Color and Brightness. When tasks are combined across
pairs—such as Height and Color or Width and Brightness—then the tasks are dissimilar
with respect to attention to the perceptual dimension on which the decision is being
made.

Another way to think of this task component is as an attentional control setting
(Folk, Remington, & Johnston, 1992). Folk et al. introduced the notion of attentional
control settings in order to account for why certain stimulus characteristics such as color
or abrupt onset appear to capture attention exogenously in some experimental situations
but not in others. They suggested that the exogenous system “can be ‘conﬁgured’ or ‘set’
to respond selectively to a property” and that “any particular system conﬁguration, or
‘attentional control setting’ is assumed to be a function of current behavioral goals.” (p.
1041) The remarkable similarity of this description of attentional control settings and
Monsell and Rogers’ (1995) discussion of task set reconﬁguration is noteworthy,
suggesting a link between the two lines of research that is worth considering (Klein &
Shore, 2000). Within the context of the current experiments, the Height and Width tasks
might be said to require the same attentional control setting, for example form. Likewise,
the Color and Brightness tasks may be considered to involve an attentional control setting
for color. The task cue for a given trial sets up the current behavioral goals for the
participant in the task switching environment including what aspect of the upcoming
stimulus is to be attended. If attentional control settings are shared by tasks within a pair,

switching between similar tasks (e. g. Height and Width) does not require a change in the

31

attentional control setting, whereas switching between dissimilar tasks (e. g. Height and
Color) does require that the attentional control setting be reconﬁgured.

Having operationally deﬁned task similarity, it is now possible to consider how
task similarity inﬂuences task space. The four tasks and the relationships among these
tasks described above serve as the building blocks for the task space as it might be
constructed by a participant in this experiment. Figure 1 presents one possible

conceptualization of this task space.

Task Space

Attention
to Form

Attention
to Color

 

 

 

 

 

Figure 1. Schematic diagram of four tasks from Experiment 1 presented in a
representational task space. Ovals represent the task set for each task. Overlap between
tasks sets indicates a shared task component—attentional control setting.

32

Each task has an associated task set, represented by the ovals. The task sets
include all cognitive operations necessary for carrying out a given task, such as decision
rules, S-R mappings, and, of particular interest for the current experimental tasks,
attentional control settings. The similarity among tasks is represented as an overlap in
the task sets for two tasks. Conceptually this overlap represents the component that is
shared by two task sets. As indicated in Figure 1, in the case of Experiments 1 and 2 this
overlap would be in the attentional control setting. In terms of this two-dimensional
representation of task space, the overlapping in task sets results in less overall distance
between the two task sets than for two task sets that are non overlapping. Translating this
two-dimensional image into a multidimensional task space suggests a model in which
similar task sets are “closer” in task space. The empirical question then becomes whether
distance between task sets manipulated by similarity between tasks affects the executive

control processes that shift between task sets in a task switching paradigm.

2.1 Experiment 1

The primary goals of Experiment 1 were to develop a task switching paradigm
where the overall task space could be manipulated based on the similarity among
individual tasks and to investigate the impact that such task similarity manipulations
might have on the executive control processes engaged during task switching. The
manipulation of task similarity was accomplished through the choice of tasks that
involved either attention to similar features of a stimulus (deﬁned by their integratality)
or attention to dissimilar features (deﬁned by their separability). These tasks were then
presented in a random order with the task cued on each trial. This design enabled

comparisons of performance on various types of trial pairs including task repetitions, task

33

switches from a similar task, and task switches from a dissimilar task. These
comparisons illuminate how the relationships between tasks impact the processes
involved in switching between those tasks.

2.1.1 Method

Participants. Participants in Experiment 1 were 20 undergraduate psychology
students who participated in partial fulﬁllment of course requirements. All participants
reported normal or corrected-to-normal visual acuity and color vision. All participants
were naive to the purpose of the experiment.

Apparatus. Presentation of stimulus displays and recording of responses were
controlled using the E-Prime software running on a Dell Dimension (Psychological
Software Tools, Inc., 2000). Responses were made using a standard keyboard.

Stimuli. All stimuli were presented against a white background. Cue stimuli were
printed in black uppercase 18 point font presented just above the center of the screen.
The four cue words were HEIGHT, WIDTH, COLOR, and BRIGHT. The target stimuli
were 16 rectangles that varied along the four task dimensions. The rectangles were four
different shapes formed by crossing two heights (32 or 48 pixels) and two widths (64 or
96 pixels). Each size rectangle appeared in four different color/brightness combinations:
light green (BGR 85 255 85), light blue (90 90 254), dark green (0 1.20 0), and dark blue
(0 0 170). The colors used were developed within the Microsoft Ofﬁce drawing palette
by ﬁrst setting the hue for each of the colors. Next the luminance was initially set at 80
for the dark shades and 160 for the light shades. However the perceived luminance was
not equal, with the green colors appearing brighter. A brief psychophysical experiment

was carried out in order to better match the colors in terms of perceived luminance.

34

Using the dark blue and light green stimuli as standards (the largest span of perceived
luminance), a sample of nine viewers adjusted the luminance on the dark green and light
blue stimuli until they matched as closely as possible the dark blue and light green
standards, respectively, in terms of perceived luminance. The mean judgments of the
nine viewers were then used to develop the stimuli used in the experiment.

Procedure. A cueing paradigm was used to present tasks in a randomly ordered
sequence. The trial line is represented in Figure 2. Trials began with the presentation of
the instructional cue. Following a 500 ms cue to target interval (CTI), the target appeared
just below the instructional cue and both displays remained on the screen until a response
was made. The screen was then blank for a 100 ms response to cue interval (RCI) until
the start of the next trial. The experimental session began with single task practice blocks
of 16 trials for each of the four tasks, presented in order based on the response ﬁnger
(index, middle, ring, little) for the task. Four different task-to-ﬁnger mappings were
counterbalanced across subjects. The four mappings were chosen such that each task was
mapped to each ﬁnger in one of the mappings. Further, two of the mappings linked
similar tasks to adjacent ﬁngers and two of the mappings linked dissimilar tasks to
adjacent ﬁngers. Responses were made on the home keys of a standard keyboard. Two
blocks of trials involving mixed trials of all four tasks followed. In the ﬁrst mixed
practice block, a “cheat sheet” showing the mapping of tasks, stimulus values, and
responses was given at the top of the screen. During this block participants were
encouraged to work slowly and accurately, and to focus on memorizing the appropriate
responses for each task. The second block of practice trials was identical to the data

collection blocks and participants were encouraged to work more quickly. Participants

35

then completed 12 blocks of data collection. Trials were run in blocks of 128 (64
task/target combinations x 2 cycles). Trials occurred in random order without
replacement for each cycle. Average response time and accuracy feedback were
provided for participants at the end of each block and participants recorded these values
on a performance sheet so that they could track their progress through the experiment.
Participants were given goals of working as accurately and quickly as possible, making
sure that their accuracy stayed above 90% and trying to decrease response time from
block to block throughout the experiment. The experimenter was present during the
practice blocks and during the ﬁrst block of data collection, but then left the room during

the remainder of data collection.

 

HEIGHT

 

 

HEIGHT

 

 

Target Present
until Response

 

 

 

 

Response to Cue
lnterva|100 ms

 

 

 

 

Figure 2. Example trial line for cued task paradigm used in Experiment 1. Elements are
not presented to scale.

36

2.1.2 Results

Data processing. Response time and accuracy data were collected for all trials.
Because trials were presented in a random order within each block, trials ﬁrst had to be
grouped into conditions based on the task indicated on Trial N, the current trial, and the
task indicated on Trial N-l, the previous trial, resulting in 16 trial conditions. Of the 16
trial conditions, four (one per task) involved task repetitions (e. g. a Height trial that
followed a Height trial), four involved task switches from a similar task (e. g. a Height
trial that followed a Width trial), and eight involved task switches from a dissimilar task
(e. g. a Height trial that followed a Color trial and a Height trial that followed a Bright
trial). Mean response times per condition were calculated as follows. The ﬁrst two trials
in each block were excluded from data analysis. Error trials and the two trials following
errors were all excluded from calculations of the mean response times. Overall response
accuracy was high (96.8%, range 91.9% to 97.9%). Following removal of trials by this
procedure, there remained on average 89.5% percent of the data, ranging for individual
participants from 77.1% to 94.6%. Data were then trimmed through removal of all
response times less than 200 ms, which were considered anticipations, and greater than
three standard deviations of the cell means for each trial condition. Response time
trimming resulted in the removal of a further 2.1% (range 1.6% to 2.7%) of the trials.
Following the trimming procedure, participants’ mean response times for the 16 trial
conditions were calculated based on an average of 84 trials per cell (range 56 to 117.)
Table 1 shows means and standard errors for response times (A) and accuracy (B) for

each trial condition.

37

 

 

 

Trial N
Trial N-l Height Width Color Bright
Height 61 5:1:22 796:1:50 81 8:1:50 980:1:65
Width 882i45 569d:19 857i51 946i56
Color 93 33:52 883i61 5473:20 913i48
Bright 962i56 882:I:64 794i40 601i16
B
Trial N

Trial N-l Height Width Color Bright
Height 98.1i0.5 96.2i0.7 96.2i0.7 96.5iO.7
Width 96.7:t0.7 98.3i0.5 96.5:t0.6 96.2:l:O.7
Color 96.4i0.9 96.4i0.8 98.1:t0.4 96.6:t0.6
Bright 96.5i0.7 96.4:t0.7 96.9i0.4 97.1106

 

Table 1. Mean i Standard Error for Response Times (A) and Accuracies (B) for Trial
Conditions Based on Task on Trial N and Trial N-l for Experiment 1

Switch cost analysis. In order to address the primary question of this experiment
concerning the effect of task similarity on switching between tasks, switch costs (task
switch minus task repetition) were calculated. Throughout the following analyses, unless
otherwise indicated, the term similarity is being used to refer to the relationship between
the tasks performed on Trial N and Trial N-l. Figure 3 shows mean switch costs
separated by similarity and the identity of the task performed on Trial N. These data

were entered into a 2 similarity (similar or dissimilar) by 4 task (Height, Width, Color, or

38

Bright) repeated-measures AN OVA. The analysis revealed a main effect of similarity

F (1,19) = 15.6, p < .01, MSE = 9619.3, with switch costs signiﬁcantly smaller following
a switch from a similar task (M = 263 ms) than following a switch from a dissimilar task
(M = 324 ms). Irnportantly this task similarity effect did not interact with which task was
being performed, F (3,57) = 1.2, p = .32, MSE = 2972.5, nor was there a signiﬁcant main
effect of task, F(3,57) = 2.0, p = .12, MSE = 20622.8. The fact that the similarity effect
did not interact with task indicates that the similarity effect cannot be attributed to
speciﬁc tasks that are more or less difﬁcult to switch from or to, but rather that the effect
results from the particular relationships between speciﬁc pairs of tasks. Combined these

results indicate that task switching is facilitated by similarity between tasks.

400
350 «
300 4»-
250 ,
200 .
150 «
100 -

50 .7 .

 

 

 

 

Switch Cost (ms)

 

 

 

 

 

 

 

 

 

 

 

Height Width Color Bright
Task

g Switch from Similar I Switch from Dissimilar

Figure 3. Mean switch costs when switching from similar and dissimilar tasks for each
task type in Experiment 1. Error bars represent 95% conﬁdence intervals calculated
based on the error term for the interaction of similarity by task as described in Loftus and
Masson (1994).

39

The response accuracy data were also examined in order to rule out the possibility
that a speed accuracy trade-off was driving the similarity effect. Response accuracy was
highest on task repetition trials (M = 97.9%) followed by switches ﬁom similar tasks (M
= 96.6%) and switches from dissimilar tasks (M = 96.4%). Irnportantly, this pattern of
data, in particular the difference between switches from similar and dissimilar tasks,
provided no evidence for a speed accuracy trade-off in responding that could bring into
question the interpretation of the response time data in the previous analysis. The 2
similarity (similar or dissimilar) by 4 task (Height, Width, Color, or Bright) repeated-
measures ANOVA on the response accuracy data showed no signiﬁcant effects.

Stimulus transition analysis. In addition to the transitions between tasks that
occur across trials, there are also stimulus transitions and it is important to rule out the
possibility that the switch costs and similarity effects seen in the ﬁrst analysis result from
the repetition or switching of stimulus values rather than the higher-level similarity of
task transition. For example, if repeating a stimulus value along an attended dimension
resulted in faster processing of the stimulus on the second trial, this could lead to faster
responding on switches from similar trials. Based on the integrality of the stimulus
dimensions, the value of the stimulus dimension attended on Trial N and Trial N-l may
beneﬁt from such facilitation if the value of the stimulus dimension is repeated across
trials. Therefore, a secondary analysis was performed examining the effects of stimulus
transition along with task transition. Stimulus transitions were deﬁned based on the
individual dimension of the stimulus that was attended either on Trial N or Trial N-l. So
in the case of a task repetition, the attended dimension was the same for both trials and

stimulus repetition versus switch would be deﬁned based on the value of the stimulus

40

dimension on which the task was being performed. In the case of a task switch, stimulus
transitions were analyzed both as a function of the stimulus dimension attended on Trial
N and as a function of the stimulus dimension attended on Trial N-l. Examining
stimulus transitions for the stimulus dimension on the current trial examines the inﬂuence
of repeating a stimulus dimension that was previously task irrelevant but is now task
relevant; while examining stimulus transitions for the stimulus dimension relevant to the
task on Trial N-l assesses the inﬂuence of repeating a previously task relevant dimension
that is not relevant for the current task.

In order to test whether the similarity effect found in the analysis of switch costs
could be accounted for by stimulus repetition rather than shifts between higher order task
sets, three variables were entered into a 2 by 2 by 2 repeated measures ANOVA:
similarity of tasks on Trial N and Trial N-l (similar or dissimilar), stimulus dimension
transition (repetition or switch), and trial on which the stimulus dimension was attended
(Trial N or Trial N-l). The main effect of stimulus transition was non signiﬁcant,

F (1,19) = .48, p = .5, MSE = 785.9, and stimulus transition did not interact with
similarity, F (1,19) = 1.4, p = .25, MSE = 636.3. Based on these ﬁndings the task
similarity effect found in the ﬁrst analysis cannot be attributed to the repetition of
stimulus dimensions.

In a separate analysis, the effect of stimulus transition was examined for task
repetitions. When the task relevant stimulus dimension was repeated participants
responded more quickly (M = 55 7 ms) than when the relevant stimulus dimension
switched between trials (M = 610 ms), t(19) = 3.7, p < .01. Because each dimension was

mapped to a different pair of ﬁngers, it was only in the case of task repetitions that the

41

stimulus repetition also involved a response repetition. The faster responses associated
with stimulus repetitions when the task is also repeated may be the result of a repetition
in the response rather than in processing of the repeated stimulus. The current design
does not allow for separation of these two factors.

Within trial compatibility analysis. While the primary analysis shows a
facilitation associated with task similarity in cases of switching between similar tasks, it
is possible that similar tasks might still result in interference of task performance if
examined from a different perspective. In the analysis of switch cost performed above,
the trials are categorized in terms of the identity of the task on Trial N and Trial N-l. The
focus in this analysis was on the effect of the task transition, without consideration of the
effect that speciﬁc stimulus characteristics might have on performance of a trial.
However, task performance on a given trial may also be affected by the speciﬁc
characteristics of that trial such as the stimulus on which the task is being performed. In
particular the stimuli in the current experiment are tetravalent meaning that they can
afford four possible responses, only one of which is correct for the task indicated for a
given trial. However, the responses indicated by the values of the stimulus dimensions
relevant for the other tasks may inﬂuence the performance of the current task. For
example, if the task is to judge the height of the rectangle and the rectangle is tall this
might indicate a response with a ﬁnger on the left hand. This same rectangle will also
have a value for each of the other stimulus dimensions--width, color, and brightness--and
those values would indicate an appropriate response for the other three tasks. These
responses might be either compatible with the current response (i.e. also left hand

responses) or incompatible with the current response (i.e. right hand responses).

42

Researchers have examined whether such compatibility of responses on stimulus
dimensions relevant for other tasks affects performance in task switching paradigms.
Typically, compatible or congruent stimuli result in faster responding than incompatible
or incongruent stimuli (Rogers & Monsell, 1995). The current situation differs slightly
from this previous work where congruency referred to the exact same response being
indicated by both dimensions of a stimulus since all tasks were mapped to the same two
button press responses (bivalent response mapping). In the current experiment, each
response is unique and compatibility simply refers to the hand with which the response
indicated for unattended dimensions would be made. However, it may still be possible to
examine whether this measure of compatibility inﬂuences responding for a particular task
by stimulus combination. Critically the question can be asked of whether the effect of the
unattended dimension varies as a function of the similarity of that dimension to the
current task dimension. So for example, will the compatibility of the width dimension
have a greater impact on the performance of the Height task than the compatibility of the
color or brightness dimensions?

This question was addressed in a 2 by 2 by 4 repeated measures ANOVA
examining the similarity of the task irrelevant dimension to the task relevant dimension
(similar or dissimilar) and the compatibility of the response indicated by the task
irrelevant dimension (compatible or incompatible), as a function of the task being
performed on the trial. The interaction of dimension similarity and compatibility was
signiﬁcant, F (1,19) = 16.8, p < .01, MSE = 324.1. For the similar dimension (e.g. width
in the case of a Height trial) compatible dimensions resulted in faster responding (M =

810 ms) than incompatible dimensions (M = 818 ms). For the dissimilar dimension (e. g.

43

color or brightness in the case of a Height trial) compatible dimensions resulted in slower
responding (M = 818 ms) than incompatible dimensions (M = 809). This result is
suggestive of an interference effect for similar stimulus dimensions as described above.
However, the magnitude of the effect is small and additionally the pattern did not hold for
all four tasks, but varied signiﬁcantly for individual tasks as seen in a signiﬁcant three-
way interaction, F(3,57) = 6.2, p < .01, MSE = 1378.8. The size and variability of this
effect may result from the fact that compatibility was deﬁned in terms of hand that would
be used to response to a value of a stimulus dimension. Since the responses were
univalent this incompatibility may not result in any great degree of interference. Perhaps
if the tasks were designed such that responses were bivalent with the same ﬁngers being
used for all tasks, the size of this compatibility effect would be more robust and show a
larger effect of similarity of the unattended dimensions on processing the currently task
relevant stimulus dimension.
2.1.3 Discussion

The facilitation of task switching based on the similarity of the tasks is a novel
ﬁnding with interesting implications. This discovery highlights the importance of
considering task space and the role that it plays in determining how coordination of tasks
and transitions among tasks occur in multitask environments. This ﬁnding differs from
previous research that demonstrated that the speciﬁc task characteristics such as practice
with or difﬁculty of a task can affect the switch costs, in that the similarity effect is not
task speciﬁc, but rather results from the relationships between tasks. Performance on
each of the four tasks varied as a ﬁrnction of the similarity between the task on the

current trial and the task on the previous trial, rather than as a function of the

44

characteristics of any individual task. This result clearly indicates that relationships
between tasks have an inﬂuence on the process of switching between tasks. F urtherrnore,
the fact that the task similarity effect resulted in a beneﬁt when switching between similar
tasks, while predicted by the task space model presented here, is an unexpected ﬁnding
based on previous research showing that similarity generally decreases performance in
dual task situations (Pashler, 1998).

The mechanisms of the task similarity effect are not known; however, two
potential mechanisms may be speculated. First, the effect may be attributable to a
reduction in the amount of controlled preparation that must be carried out in order to
switch between similar task sets as compared to dissimilar task sets. In relation to the
model of task space developed above, this might be thought of as a decrease in the
distance between two task sets resulting in a reduction in the time necessary to switch
between the two task sets. This interpretation would be reminiscent of the location
shifting model pr0posed by Schvaneveldt and Meyer (1973) for retrieval from semantic
space. The same process may also be conceptualized in terms of the number of cognitive
processing units or modules that have to be changed from the performance of one task to
another (i.e. how many features differ between the two task sets). If there are fewer
components in the task set that must be reconﬁgured, less reconﬁguration is necessary. If
reconﬁguration of individual task components occurs in a serial fashion or in parallel but
at a speed proportional to the number of components being reconﬁgured, then this would
translate into a larger preparation interval necessary to fully activate the appropriate task

set for the upcoming trial in the case of switches between dissimilar tasks.

45

Within models of task switching, task set reconﬁguration is a thought to be an
active or controlled process. These controlled processes can be carried out prior to the
onset of the task stimulus when the upcoming task is known. Thus one possible
mechanism for the task similarity effect is through a decrease in the controlled processes
that must be carried out in order to prepare for an upcoming task.

Second, the task similarity effect may involve a passive or automatic priming of
the upcoming task when a similar task is performed on the previous trial. Several models
of task switching propose some sort of automatic mechanism resulting in switch costs;
however, these models take various forms. Allport et al.’s (1994) task set inertia model
focused on a passive mechanism of proactive interference by which task performance
following a switch is slower because of interference from residual activity associated
with the S-R mappings from the previous task. Other models cast this automatic process
in terms of repetition priming where aspects of a task are preformed more rapidly on task
repetitions because they are primed from the previous trial (Sohn & Anderson, 2001;
Sohn & Carlson, 2000). Both cost associated with interference following a task switch
and beneﬁt following a task repetition would result in the pattern of responding typically
referred to as switch cost. However, when considering how the task similarity effect
might result from automatic processes, the repetition priming model appears to offer a
more probable mechanism than the proactive interference suggested in the task set inertia
model. If the task sets for similar tasks involve an overlap, as depicted in Figure 1, then
in the case of a task switch from a similar task, residual activation associated with the
task set for the Trial N-l task would include some activation of the Trial N task set. The

result of the residual activation would thus be to boost the starting level of activation for

46

the current task set and the result would be “repetition priming” for the portion of the task
set that overlapped with the task set activated on the previous trial.

Whether the mechanisms behind the task similarity effect turn out to involve
controlled or automatic processes, the implication of the ﬁndings of the current
experiment is clear. The task space in which tasks are presented affects the process of
switching between tasks. These results support the idea that examination and
manipulation of task space provides a useful tool for understanding the executive control

processes that act on the tasks and task space.
2.2 Experiment 2

The aim of Experiment 2 was to further examine the task similarity effect by
separating out potential effects associated with a controlled preparation for an upcoming
task and with automatic priming from the task set activated on the previous trial in order
to identify the mechanisms by which the task similarity effect occurs. One approach for
separating automatic and controlled processes in the task switching paradigm is through
parametric manipulation of the time over which these processes are engaged (Meiran et
al., 2000). The logic behind such temporal manipulations is as follows: a controlled
process that is carried out in order to prepare for an upcoming task cannot occur until the
identity of the upcoming task is known. Manipulating the time between the instructional
cue and target stimulus for the current trial (preparation interval or one target interval,
CTI) impacts the degree of controlled processing that can occur prior to the onset of the
target stimulus. On the other hand, any automatic effects resulting from priming
associated with residual activation of the previous task set would be expected to decrease

over the entire delay interval between the response to the previous trial and the onset of

47

the target stimulus for the current trial. So manipulating the time between the response
ending the previous trial and the presentation of the target stimulus for the current trial
(delay interval or response target interval, RTI) affects the amount of residual activation
associated with the task set from Trial N-l and thus the degree to which this residual
activation affects performance on Trial N. The cueing paradigm used in Experiment 1,
where tasks are presented in random order with an instructional cue indicating the task to
be performed on a given trial, is well suited for the independent manipulation of delay
interval and preparation interval. First, the overall delay interval can be lengthened while
holding the preparation interval constant by increasing the amount of time between the
response to the previous trial and the onset of the cue for the upcoming trial (response cue
interval, RCI). Alternatively, the preparation interval can be manipulated by varying
CTI, while holding the delay interval constant.

Previous studies that used manipulations of these temporal measures have
demonstrated the effect of both controlled and automatic processes on task switching by
showing reductions in switch costs associated with both increasing preparation interval
and increasing delay interval. Meiran (1996; Meiran etal., 2000) found decreases in
switch cost associated with increases in both RTI and CTI in a cueing procedure. Sohn
and Anderson (2001) showed decreases in switch costs with increasing RTI even in
conditions when no foreknowledge about the upcoming task was provided. In neither
case did the switch cost completely disappear even with very long delay and preparation
intervals, thus indicating the presence of the residual switch cost.

While this previous research has demonstrated that both automatic and controlled

processes are engaged during task switching and inﬂuence the magnitude of the switch

48

cost, it is not known whether these processes might be selectively acting upon speciﬁc
component operations of the task set. Rubinstein et al. (2001) suggested that it was
important to specify “when, where, and to what extent” these two types of mechanisms
are involved in task switching. It may be that the similarity effect discovered in the ﬁrst
experiment will provide a tool for answering these questions. In the research discussed
above manipulating timing of events in a cueing paradigm, the changes in the switch cost
associated with different timing parameters is interpreted as reﬂecting the sum of the
processes taking place during task switching. If, however, we examine not the overall
changes in switch cost, but rather the changes in the similarity effect, this measure will
reﬂect the speciﬁc processes acting on the component of the task set on which similarity
is deﬁned. Thus by looking at the impact of temporal manipulations on the size of the
similarity effect, we can begin to isolate the processes, controlled or automatic, that act
on a particular component of the task set, in the case of the current tasks—the attentional
control setting.

Experiment 2 uses manipulations of the timing of elements in the trial line in
order to address the question of what mechanisms underlie the task similarity effect. The
manipulation of the delay interval through increasing the RC1 while maintaining the same
CTI varies the automatic inﬂuences on task performance while holding constant the
controlled inﬂuence. Manipulating both the CTI and the RC1 and thus varying
preparation interval within a constant delay interval allows examining controlled
processing while holding automatic inﬂuences stable. If the task similarity effect results
from a reduction in the amount of reconﬁguration that must occur between similar tasks,

then changes in the time available for controlled reconﬁguration processes to be carried

49

out (i.e. an increase in preparation interval) should affect the task similarity effect. If,
however, the task similarity effect arises ﬁ'om the repetition priming of a portion of the
task set that overlaps between the current task and the previous task, then changes in the
amount of automatic processes (i.e. an increase in the overall delay interval) should affect
the task similarity effect.

2.2.1 Method

Participants. Participants were drawn from the same population as Experiment 1.
A total of 75 individuals participated. Data from three participants was removed due to
low accuracy (below 90%) or failure to complete all blocks in the experiment, resulting
in 24 participants per between-subjects condition.

Stimuli and apparatus. The equipment was identical to Experiment 1. The
stimuli in this experiment were a subset of those used in Experiment 1, reﬂecting the
change to a three task design. The switch to a three task version of the experiment was
made so that the experimental sessions would not be made overly long by changes in the
timing of the trials, while still allowing for an adequate number of trials per cell in the
experimental design. The task cues HEIGHT, WIDTH, and COLOR were used. The
target stimuli consisted of eight of the rectangles from Experiment 1 (the light luminance
stimuli). The three tasks by eight target stimuli resulted in 24 unique cue target pairs.

Procedure. The sequence of trial events was the same as in Experiment 1, but the
timing of the trial events now varied and was manipulated between subjects. The levels
of the timing variable were 100 ms RCI/SOO ms CTI (Group US, identical to Experiment
1), 700 ms RCI/500 ms CTI (Group 7/5), and 100 ms RCI/l 100 ms CTI (Group 1/1 1).

The timing conditions are illustrated in Figure 4.

50

 

Group 1/5

'3' -.'- ‘‘‘‘‘‘‘ . .. '. ' . '- '. -_ ' ‘5 ,. -:
.

 

 

 

 

 

 

 

 

 

 

 

 

 

 

Group 7/5

Group 1/11

1 l l l l l l l l A l l l
T 100 300 500 700 900 1 100
Response Time in milliseconds

Trial N-1

[:l Response to Cue Cue to Target
Interval " Interval

Figure 4. Timing of elements in the trial line for the three groups in Experiments 2 and 4.
Group 1/ 5 and Group 7/5 have the same preparation interval, but different delay intervals.
Comparing these conditions equates for the amount of controlled processing that can take
place once the identity of the upcoming task is known, but varies the amount of time over
which decay of activity associated with the task set from Trial N-l occurs. Group 7/5
and Group 1/11 have the same delay interval but different preparation intervals.
Comparing these conditions varies the amont of controlled processing while equating for
the decay of activity associated with the task set from Trial N-l.

Participants responded using the index, middle, and ring ﬁngers of both hands.
The mapping of task to ﬁnger pair was counterbalanced across participants. The left
hand responses were made on the c, d, and s keys, while right hand responses were made
on the m, k, and 1 keys. This offset hand position provided a more comfortable wrist
position than the use of the home keys. The session structure was similar to Experiment

1. Following practice with individual task and mixed task blocks, participants performed

51

12 blocks of trials for which data were collected. The trials were run in blocks of 96
trials (24 cue target pairs x 4 cycles). The entire session lasted between 60 and 90
minutes depending on the trial length based on the timing manipulation.
2.2.2 Results

Data processing. The procedures for data processing were the same as used in
Experiment 1. Response accuracy was high (96.9%, range 92.8% to 100%). Following
removal of error and subsequent two trials 89.3% (range 77.5% to 97.9%) of the trials
remained. Response time trimming with a low end cut off of 200 ms and high end based
on 3 standard deviations of participants’ mean response time per trial conditions (deﬁned
by task type on Trial N and Trial N-l) resulted in the further removal of 2.2% (range
1.5% to 3.2%) of the trials. As in Experiment 1, trials were initially sorted into
conditions based on task type on Trial N and Trial N-l , resulting in nine different trial
conditions. Means and standard errors of response times (A) and accuracies (B) for all
nine trial conditions separated by timing condition are given in Table 2. Calculations of
mean response times for individual participants were based on an average of 112 trials
per cell in the design (range 70 to 143 trials). All three tasks could occur as task
repetitions, however, in the case of the Color task, all switch trials were the result of a
switch from a dissimilar task. Only for the Height and Width tasks could the trials
involve either a switch from a similar task or a switch ﬁ'om a dissimilar task. For this
reason the primary analyses of task similarity involved analysis of data from only Height

and Width trials.

52

A

Timing Condition 100/500

 

 

 

 

 

Trial N
Trial N-l Height Width Color
Height 606120 706i26 578:1:23
Width 721i32 575:1:16 566i25
Color 75 3142 741:1:35 490i 1 6
Timing Condition 700/500

Trial N
Trial N-l Height Width Color
Height 522i20 626i33 477i18
Width 614:1:37 494il 8 473i17
Color 65 22t42 668i42 413:1:12
Timing Condition 1 00/ l 100

Trial N
Trial N-l Height Width Color
Height 550i29 645:1:39 517:22
Width 65 5&39 5 18i23 534d:22
Color 649i40 630i40 449i16

 

Table 2. Mean i Standard Error for Response Times (A) and Accuracies (B) for Trial
Conditions Based on Task on Trial N and Trial N-l Separated by Timing Condition for
Experiment 2

53

Table 2 (cont’d).

 

 

 

 

 

B
Timing Condition 100/500

Trial N
Trial N-l Height Width Color
Height 96.5i0.7 96.5i0.2 97.7:t0.4
Width 97.1:t0.4 96.9:t0.5 97.5:t0.5
Color 96.4i0.5 96510.5 98.0i0.4
Timing Condition 700/500

Trial N
Trial N-l Height Width Color
Height 96.7i0.4 95.73207 97.2105
Width 96.4:t0.6 97.7d:0.3 97.1i0.4
Color 95.7:t0.6 96.1i0.6 98.5i0.3
Timing Condition 100/1100

Trial N
Trial N-l Height Width Color
Height 95.8i0.6 95.7i0.6 98.0i.04
Width 96.3:1:O.5 97.2d:0.5 98.5i0.5
Color 95.8i0.7 96.5i0.6 98.6i0.3

 

Switch cost analysis. The primary question of this experiment is whether the task
similarity effect seen in Experiment 1 varies as a ﬁrnction of the temporal parameters of

the elements in the trial. In order to address this question, a 2 similarity (similar or

54

dissimilar) by 2 task (Height or Width) by 3 timing condition (Group US, Group 7/5, or
Group 1/ l 1) mixed ANOVA was conducted on switch costs, calculated by subtracting
response times for task repetitions from response times for task switch conditions.
Switch costs were signiﬁcantly greater for the Width task (M = 143 ms) than for the
Height task (M: 115 ms), F(1, 69) = 11.9,p < .01, MSE = 4602.0. However, the task
variable did not enter into any signiﬁcant interactions. Interestingly, while the data are
suggestive of a decrease in overall switch costs for both increases in delay interval and
increases in preparation interval, a ﬁnding that would be in line with previous research
showing decreasing switch costs, the main effect of timing condition was non signiﬁcant,
F (2, 69) = 0.37, p = .69, MSE = 59854.0. The main effect of task similarity was found in
the expected direction based on the results from Experiment 1, with smaller switch costs
for switches from similar tasks (M = 119 ms) than for switches from dissimilar tasks (M
= 138 ms), F(1, 69) = 5.3, p < .05, MSE = 5015.7, thus providing a replication ofthe
results from Experiment 1. The smaller magnitude of the switch costs in the current
experiment as compared to Experiment 1 may reﬂect the fact that there were only three
rather than four possible tasks in the task space. Critically for the focus of the current
experiment, the similarity effect interacted signiﬁcantly with the timing manipulation,
F(2, 69) = 4.8, p < .05, MSE = 5015.7. This interaction can be seen in the graph of the
data in Figure 5, which shows switch costs collapsed across tasks for switches from
similar and dissimilar tasks as a function of the timing condition. As demonstrated in the
graph of the data, the similarity effect appears to be present for Group US and Group 7/5,

but to be absent for Group l/l l.

55

400
350 +
300...“— 777---, - _-,L-,-__.,,,.
250 .7 , . 7 . - 2
200 ..
150 l“
100
50 -

 

 

Switch Cost (ms)

 

 

 

 

 

 

 

 

 

 

 

 

 

 

Group 1/5 Group 7/5 Group 1/11
E] Switch from Similar I Switch from Dissimilar

Figure 5. Mean switch costs when switching from similar and dissimilar tasks for each
timing condition in Experiment 2. Error bars represent 95% conﬁdence intervals
calculated based on the error term for the similarity effect for ANOVAs calculated for
each between subjects condition separately.

The timing conditions were chosen to separately manipulate the delay interval
(during which residual activation associated with the previous task would presumably
decay) and the preparation interval (during which the participant could prepare for the
upcoming task) and thus to manipulate the degree of automatic and controlled processing,
respectively. Therefore, the interaction of similarity by timing condition found in the
ﬁrst analysis, was broken down in separate analyses. The ﬁrst analysis compared the
Group US and Group 7/5 data, focusing on automatic mechanisms for the similarity
effect, and the second analyses compared the Group 7/5 and Group 1/11 data, focusing on
controlled mechanisms for the similarity effect. The analysis of the Group US and Group

7/5 conditions showed a main effect of similarity, F (l, 46) = 11.8, p < .01, MSE =

5408.2, that did not interact with the timing condition, F (l , 46) = .90, p = .38, MSE =

56

5408.2. This pattern of results indicates that the similarity effect does not vary with a
change in the overall delay interval between the response from the previous task and the
onset of the target for the current task. However, the analysis of the Group 7/5 and
Group 1/11 conditions revealed a signiﬁcant interaction of similarity and timing
condition, F (1 , 46) = 11.1, p < .01, MSE = 4147.4. This pattern of results conﬁrms that
an increase in preparation interval from 500 ms to 1100 ms resulted in a signiﬁcant
decrease, and in fact, the elimination of the similarity effect.

As in Experiment 1, the response accuracy data were also considered in order to
assess the possibility of a speed accuracy trade-off. Again, response accuracy did not
suggest a speed accuracy trade-off with accuracy highest on task repetition trials (M =
96.8%) followed by switches from similar tasks (M = 96.3%) and switches from
dissimilar tasks (M = 96.2%). The 2 similarity (similar or dissimilar) by 4 task (Height,
Width, Color, or Bright) ) by 3 timing condition (Group US, Group 7/5, or Group 1/11)
mixed AN OVA was performed on the response accuracy data. The only signiﬁcant
effect was the interaction of similarity by task, F(1, 69) = 7.2, p < .01, MSE = .00027.
For the Height task response accuracy was in the expected direction with responding
more accurate for switches from similar tasks than for switches from dissimilar tasks,
96.6% and 95.9%, respectively. However, for the Width task response accuracy was in
the opposite direction, with response accuracies of 96.0% and 96.4% for similar and
dissimilar switches respectively. The cause of this cross over interaction is unclear and
the effect is not seen in the analysis of the response time data where there was no

interaction of similarity by task.

57

Stimulus transition analysis. As in Experiment 1, the effects of stimulus
dimension repetitions were assessed in order to rule out the possibility that the similarity
effects found in task transitions might result from repetition of an attended stimulus
dimension. A 2 by 2 by 2 by 3 mixed ANOVA was conducted looking at similarity of
the task transition from Trial N and Trial N-l (similar or dissimilar), stimulus dimension
transition (repetition or switch), trial on which the stimulus dimension was attended
(Trial N or Trial N-l), and timing condition (Group US, Group 7/5, or Group 1/11). As
in Experiment 1, the similarity of the task transition did not enter into any signiﬁcant
interactions with stimulus transition, thus indicating that the similarity effect found in the
ﬁrst analysis cannot be the result of stimulus repetition. The main effect of repeating the
value of a stimulus dimension was signiﬁcant, F(1, 69) = 7.8, p < .01, MSE = 1788.6,
with responding slower on trials when an attended stimulus dimension was repeated (M =
675 ms) than when none of the attended stimulus dimensions were repeated (M = 668
ms). This effect must be considered in light of a signiﬁcant interaction of the stimulus
transition effect with the trial on which that dimension of the stimulus was attended.
When the stimulus repetition is deﬁned based on the dimension of the stimulus that is
attended on Trial N, for which the response time data is being collected, then repeating
the stimulus dimension slows performance (M = 679 ms) compared to switching the
value of the stimulus dimension (M = 665 ms). However, if the stimulus transition is
deﬁned based on the dimension of the stimulus that is attended on Trial N-l , then
stimulus transition has no effect on response time (M = 671 for both conditions). This
interaction indicates when switching to a task the processing of the stimulus dimension

relevant for the performance of that task is affected by the previous exposure to that task

58

dimensions. Speciﬁcally if the value of the stimulus dimensions on the current task is
repeated from the previous task (when that stimulus dimension was not task relevant)
then performance of the task is slower. This pattern would be expected if processing of a
recently ignored stimulus value is slowed.

Stimulus transitions were also analyzed for task repetition trials where the same
stimulus dimension was attended on both Trial N and Trial N-l. In this case when the
task is repeated, stimulus repetitions resulted in response repetition as well. Unlike
Experiment 1 where there was a signiﬁcant beneﬁt for stimulus repetitions under these
conditions, a 2 stimulus transition (repeat or switch) by 3 task (Height, Width, or Color)
by 3 timing condition (Group US, Group 7/5, or Group 1/11 mixed ANOVA revealed no
signiﬁcant effects involving stimulus repetition. In sum, repetition of the stimulus
dimensions across Trial N and Trial N-l showed very little impact of task performance
and could not be a cause of the task similarity effects described above.

Within trial compatibility analysis. Trials were again analyzed for the
compatibility of the response indicated by the unattended stimulus dimensions with the
response indicated by the currently task relevant dimension. A 2 compatibility of the
response by 2 similarity of the unattended and attended dimensions by 2 task (Height or
Width) repeated measures AN OVA was conducted. There was a signiﬁcant main effect
of compatibility, F (1, 69) = 30.0, p < .01, MSE = 674.2, such that responding was faster if
the unattended stimulus dimension indicated a response with the same hand as the
response indicated by the attended stimulus dimension (M = 627 ms) than if a response
with the opposite hand was indicated (M = 639 ms). This effect did not interact with

either of the other two factors.

59

Trial N—2 sequence analysis. As well as examining switch costs for task
transitions deﬁned based on the task performed on Trial N and Trial N-l, trials can also
be categorized based on the task transitions over the course of three trials, Trial N, Trial
N-l, and Trial N-2 . Table 3 gives means and standard errors for response times (A) and

accuracies (B) for each of the 27 conditions generated by combining the three tasks

 

 

 

 

A
Trial N Task
Height Width Color
Trial N-l Height Width Color Height Width Color Height Width Color
Group US
Trial N—2
Height 599i21 737i32 812i48 681i27 5873:18 766st41 558i23 555i29 494i 1 6
Width 618i23 700:37 740i44 7363:28 5741:19 771i34 577:25 549:1:21 496:1:17
Color 5983:20 730132 713i4l 700i27 564i 1 6 688134 597i23 595i28 480116
Group 7/5
Trial N-2
Height 5 13¢ 18 625i40 67 1 $42 5943:31 507i20 681i43 470120 469i19 414i12
Width 542123 602i36 644145 656i36 479$ 18 672i44 467116 466i16 420:1:13
Color 51 1i20 6063:37 643i44 627i33 495i18 650142 492i19 487i20 402i 1 2
Group 1/ 11
Height 548:30 6833:43 674i43 635i40 5201:25 655i48 505i21 523i22 447i16
Width 55 li31 630:1:40 660i42 664i37 520:1:26 620135 5193:22 524i22 455i17
Color 5493:30 656i39 613i39 6383:42 512i23 613i42 530i25 558i23 444i 1 7

 

Table 3. Mean d: Standard Error for Response Times (A) and Accuracies (B) for Trial
Conditions Based on Task on Trial N, Trial N-l, and Trial N-2 Separated by Timing
Condition for Experiment 2

60

Table 3 (cont’d).

B

Trial N-l

Height

Height

Trial N Task
Width

Color

 

Width

Color

Height Width

Color

Height

Width

Color

 

Trial N-2

Height
Width

Color

97.0108
96.2109
96511.0

97.1106
96.9107
97.3105

96.4105
96.4106
96.5108

Group US

96.5107 96.7106
96.6109 97.0108
96.4106 96.9105

97.2106
95.0110
96.8106

98. 110.4
97.6107
97.6106

98.0104
97.5107
97.2106

98.0106
98.2104
97.7106

 

Trial N-2

Height
Width

Color

96.8108
96.6106
96.5107

96.7107
96.6107
95.8109

95.7107
96.1109
95.2109

Group 7/5

95511.1 96.9106
94.71l.l 98.1104
96.7108 98.3106

96.0107
96.9108
95.9109

97.1107
97.0106
97.5107

97.3105
97.1106
96.9106

98.7103
98.2104
98.6107

 

Trial N-2

Height
Width

Color

97.5106
94.7108
95.6108

96.3107
96.1107
96.4106

95.8107
95611.0
96011.0

Group 1/11

96.5106 97. 110.6
95.8108 98.0-10.6
95110.9 96.9106

95.7109
96.4107
97.6106

98.5105
98.0104
97.7106

98.5104
98.9106
98.0105

99.0103
98.4104
98.1104

 

across each trial sequence. In looking at runs of three trials when the tasks on Trials N

and N-l differ from each other (B-C), three types of sequences can occur: unique tasks

sequence where each trial involves a different task, A-B-C; alternating tasks sequence

where the task is the same on Trial N and Trial N-2, C-B-C; and repeat-switch tasks

sequence where the task is the same on Trial N-l and Trial N-2, B-B-C. The effect of the

three trial sequences was assessed in a 3 task sequence (unique tasks, alternating tasks, or

repeat-switch tasks) by 2 similarity of Trial N and Trial N-l (similar or dissimilar) by 2

61

task (Height or Width) by 3 timing condition (Group US, Group 7/5, or Group 1/11)
mixed ANOVA. The main effect of sequence was signiﬁcant, F (2, 138) = 42.8, p < .01 ,
MSE = 3823.2. Response times were the slowest in the alternating tasks sequence (M =
694 ms), intermediate in the unique tasks sequence (M = 675 ms), and fastest in the
repeat-switch tasks sequence (M = 647 ms). This pattern of results was mirrored in an
analysis of response accuracy (M = 96.1%, 96.2%, and 96.3%, respectively) although no
effects in the response accuracy analysis reached signiﬁcance. Thus, at the most general
level, this pattern of results replicates previous ﬁndings for these sequence effects
(Arbuthnott & Frank, 2000; Arrington, Altmann, & Carr, 2001; Mayr & Keele, 2000).
However, the sequence effect did not hold across all of the various conditions. Two
different three-way interactions involving sequence were signiﬁcant: sequence by
similarity by timing condition, F (4, 138) = 3.4, p < .05, MSE = 2742.4, and sequence by
similarity by task, F (2, 138) = 7.0, p < .01, MSE = 2333.3. In each three-way interaction
the pattern of the sequence effect shown above held for all cells in the design except one:
dissimilar by Group 1/11 condition in the ﬁrst interaction and dissimilar by Width
condition in the second interaction. In these cases the unique tasks sequence resulted in
slower responding than the alternating task sequence. So while there appears to be some
systematicity in response times based on three trial sequences, the effect is dependent on
task and timing conditions. This ﬁnding will be more thoroughly considered in the
General Discussion.
2.2.3 Discussion

The results of Experiment 2 clearly indicate that the task similarity effect, deﬁned

within the current context as a shared attentional control setting, arises from a mechanism

62

governed by controlled preparation for the upcoming task. The cost of switching from a
dissimilar task as compared to a similar task was completely eliminated when the time
available for preparing for the upcoming task was increased, while at the same time being
unaffected by a similar increase in the overall delay interval. This result can be
accounted for by considering that the task set reconﬁguration that occurs during switches
from a dissimilar task includes an added component that is not present for switches from
a similar task: the need to reconﬁgure the attentional control setting. When the
preparation interval is short, 500 ms in the case of Group US and Group 7/5, then the
reconﬁguration of the attentional control setting cannot be completed prior to the onset of
the target stimulus and thus the performance of the task following a switch from a
dissimilar task is slowed. The longer 1100 ms preparation interval for Group 1/11
provides enough time for the reconﬁguration of the attentional control setting. Thus in
the case of both a switch from a similar task and a switch from a dissimilar task the
attentional control setting is established by the time the target for the current task appears,
making performance following either type of switch equally rapid. This ﬁnding is in
direct contradiction to claims by Allport and Wylie (1999) that there is no evidence that
preparation for the upcoming task involves enabling operations involved in perceptual
processing.

The fact that the similarity effect does not decrease between Group US and 7/5,
suggests the attentional control setting does not change or degrade with the increase in
delay time. Such degradation would be expected to result in a reduction of the similarity
effect, which is not seen in the current data. The logic behind the manipulation of the

delay interval in the current experiment is that the residual activity associated with a task

63

set decays over time resulting in the decrease of any automatic processes that might affect
performance on Trial N (see Meiran et al., 2000 for a similar discussion). Therefore, the
conclusion from the current results must then be that automatic processes that result from
priming associated with residual activity from the previous task set are not involved in
attentional aspects of the task. However, these conclusions are certainly limited to the
time interval chosen for the current experiment and it is possible that over this time
interval there was not enough passive decay in activation of the attentional control setting
to affect performance. A longer delay interval might result in a decrease of the similarity
effect even when the preparation interval is relatively short such as the 500 ms interval in
the current experiment. If the task similarity effect is seen as a beneﬁt in terms of not
having to reconﬁgure the attentional control settings between trials, then anything that
disrupted the attentional control setting between Trial N-l and Trial N might eliminate
the similarity effect. Thus a longer delay or a disruption during the delay, would result in
equivalent switch costs for switches from similar and dissimilar tasks, since in both cases
the attentional control setting would have to be conﬁgured prior to performance of the
task on Trial N.

As noted earlier, changes in the similarity effect are indicative of processes acting
on particular aspects of the task sets. Understanding such processes allows us to begin to
carve up the larger task space. The most straight forward interpretation of the current
ﬁndings is that the establishment of the attentional control setting occurs through a
controlled process and thus requires time to perform once the identity of the upcoming

task is known.

64

It is worth noting that while there were signiﬁcant changes in the similarity effect
with changes in timing, the overall switch cost did not decrease as a function of either
increases in delay interval or increases in preparation interval. This ﬁnding appears to be
in conﬂict with previous reports of decreases in switch costs following similar
experimental manipulations (DeJong, 2000; Goschke, 2000; Meiran et al., 2000). One
methodological difference between the current experiment and others that show decreases
in switch cost with increases in delay or preparation interval is that the timing intervals
are manipulated between subjects in the current experiment. In previous research
participants saw multiple timing conditions either mixed randomly within a block
(DeJong, 2000; Meiran et al., 2000) or across blocks in the same experimental session
(Goschke, 2000). It may be that strategic effects that arise when participants know there
are variable timings of trial elements impact the controlled processes of preparation such
that preparation varies as a function of the timing of the trial. Such strategic effects

would not occur when participants are only exposed to one timing condition.

2.3 Interim Conclusions

Taken together Experiments 1 and 2 provide evidence that manipulating the
relationships between tasks can impact the executive control processes involved when
switching between tasks. While this result may seem intuitive, the question is not one
that has been widely considered previously in the task switching literature. Experiment 1
demonstrated that task similarity provides a systematic beneﬁt for switches between
similar tasks as compared to switches between dissimilar tasks. These ﬁndings can be
accounted for within the model of task space put forward in the introduction. Similar

tasks are closer in a multidimensional task space resulting in faster transitions between

65

similar tasks as compared to dissimilar tasks that are further from each other in task
space. Building on the initial ﬁndings from the ﬁrst experiment, Experiment 2 showed
how the similarity effect could be used to examine the nature of the controlled and
automatic processes involved in the speciﬁc component of the task set on which
similarity was deﬁned. Thus these experiments demonstrate not only that the
manipulations of the representational task space impact processes of executive control,
but that this effect can be used to begin to describe the relationship between individual
task components (such as attentional control settings) and controlled versus automatic
processes involved when switching between tasks. While these conclusions are being
cast in terms of a general quality of task space, they have only been shown to occur in
terms of a manipulation of the attentional component of the task set. In the following
chapter, I extend the study of the task similarity effect to another group of tasks using a

different component of the task set to deﬁne task similarity: the response modality.

66

3 Task Similarity as Shared Response Modality

The next two experiments follow the general structure and logic of the ﬁrst pair of
experiments; however, the tasks are now designed so that the similarity among tasks is
deﬁned based on the response component of the task set rather than the attentional
component. In order to test the hypothesis that task similarity facilitates task switching, it
is necessary to consider task similarity deﬁned in terms of other component operations,
extending beyond the attentional component used to deﬁne similarity in the ﬁrst two
experiments. For several reasons, response modality is a good component operation on
which to manipulate similarity. Response modality is a clearly separate component of the
task set from the attentional component that served to deﬁne similarity in the ﬁrst two
experiments. Thus manipulating similarity through shared response mode provides a ﬁne
compliment to the earlier experiments. Further, response selection has been implicated in
psychological refractor period (PRP) experiments as the process that results in the central
bottleneck in dual task performance (Pashler, 2000). The implications of this ﬁnding in
terms of the role that executive control must play in response selection makes this
component of the task set particularly interesting when considering processes involved in
task switching.

Experiment 3 introduces four new tasks that form two pairs of similar tasks, with
similarity a function of the response modality. In each task participants make a simple
judgment about whether a rectangle is tall or short and indicate that decision with a
response that varies based on the current task. Two tasks involve manual responses: one
using the ﬁrst or index ﬁnger of the right and left hand to indicate target height; the other

using the second or middle ﬁnger of each hand. The other two tasks involve vocal

67

responses: one where the participants respond by saying a number, either “one” or “two”;
one where the participant respond by saying a letter, either “A” or “B”. The responses
for each task are unique and were chosen such that there were equally arbitrary mappings
of the decision about the target stimulus to the response. Pre-potent responses such as
using “short” and “tall” for vocal responses would result in a task where there is already
an established S—R mapping that would set the vocal task apart from the manual tasks.
While the tasks are equivalent in terms of the perceptual, encoding, and decision
processes, they are different in terms of the response.

Task similarity for these new tasks can again be deﬁned in terms of a shared
component, which in this case is the response modality. While the individual responses
for each task are unique (just as the stimulus dimensions on which the decisions were
being made for the tasks in the ﬁrst two experiments were unique), the tasks are grouped
based on the response modality: manual or vocal. The separation of manual and vocal
responses into two separate response streams is supported by both psychological and
neurophysiological evidence. Within the dual task literature, a number of studies have
examined situations involving both manual and vocal responses. McLeod (1977)
provided an early demonstration that manual and vocal responses involved different
processing streams. In a dual task environment, participants performed a primary task
involving visual tracking that required a manual joystick response. The secondary task
was the presentation of a tone that participants responded to either by making a manual
response or a vocal response. The results showed interruption of the primary task only
when responses to the secondary task were made in the same modality. Based on these

ﬁndings, McLeod concluded that “the vocal and manual responses are produced by

68

independent processors, but the two streams of manual responses are produced by
interacting processes” (p. 659). Other behavioral evidence further supports this
separation of response modality including work using the Stroop task (Keele, 1973) and
the PRP paradigm (Pashler, 1990).

In addition to the behavioral evidence cited above, response similarity is also
implemented at the level of the neurophysiological systems underlying the individual
responses. Not only are there clearly muscular level differences between the effector
systems of these two response modalities, but cortical control of these effector systems is
separated. The motor cortex is organized somatotopically such that representations of
different body regions are laid out in an organized fashion in the motor cortex (Schieber,
1999). The cortical representation of the ﬁrst and second ﬁngers involves adjacent and
indeed overlapping regions of motor cortex (Sanes, Donoghue, Thangaraj, Edelman, &
Warach, 1995). The ﬁnger representations are separated in the so-called sensorimotor
homunculus from the motor regions controlling speech (Huang, Carr, & Cao, 2001).
Thus in the current set of tasks, response similarity also may be thought of in terms of the
neural regions involved in production of the response.

The organization of tasks into a representational task space for the present
experiments could occur in much the same fashion as was explained in detail for the ﬁrst
experiment (refer to Figure 1). The similarity of the task sets for tasks involving the
same response modality would be indicated by overlap. The overlap now would
represent the response modality of the task rather than the attentional control setting.
Recall that the overlap in task set also results in an overall decrease in the “distance”

between two tasks, thus providing a conceptual representation of the degree of similarity

69

between two tasks. Based on both the model of task space put forward in the
introduction and the results from the previous experiments, it would be expected that task
similarity would facilitate switches between tasks. However, it is interesting to note that
the similarity in response modality has more often been associated with interference of
task performance in dual task environments. Consider the evidence from Pashler (1990)
showing greater PRP effects—slowed responding to the second of two stimuli presented
in quick succession—when each task involved responses in the same response modality
as compared to separate response modalities. The PRP paradigm is in many respects
similar to the task switching paradigm except that the performance of the tasks is
overlapping in time rather than sequential as it is in task switching. In the PRP paradigm,
response similarity results in costs in performance of the second task rather than beneﬁts.
This ﬁnding would suggest a different prediction than the current model of task space for
the effect that task similarity deﬁned in terms of response modality would have on
switching between tasks. The differences between the PRP and task switching paradigms
will be considered in greater detail in the General Discussion.

As the following two experiments will be compared to the ﬁrst two, which they
mirror in design, it is important to examine the two groups of tasks used across the pairs
of experiments in order to clarify the differences between task environments. In the ﬁrst
pair of experiments, the tasks varied based on the stimulus dimension on which the
stimulus was being judged. Thus, task switching required possible changes in the
attentional control setting and the decisions being made for each task, while the responses
were always made in the same response modality, manual. In the second pair of

experiments, this pattern is reversed. The attentional aspects of the task and the judgment

70

being made are the same for all tasks, but the response modality changes among the
tasks. Clearly, the new set of tasks involves fewer differences in component operations
than the set of tasks used in the ﬁrst two experiments. The tasks differ only in terms of
the responses, but are in all other respects the same. While the differences in component
operations among tasks used in the ﬁrst set of experiments included the decision being
made as well as the attentional control setting. However, in both cases the tasks are
grouped as similar or dissimilar based on a single cognitive component of the task that is
shared within pairs of similar tasks—attentional control setting in the previous

experiments and response modality in the upcoming experiments.

3.1 Experiment 3

Experiment 3 introduces two new pairs of tasks in which tasks within a pair are
similar and tasks between pairs are dissimilar. As described above, task similarity is
deﬁned now in terms of response modality rather than attentional control settings as in
Experiment 1. Otherwise, the two experiments are parallel.

3.1.1 Method

Participants. The participants were 21 undergraduate psychology students who
participated in partial fulﬁllment of course requirements. One participant was eliminated
for failure to complete all experimental blocks during the one and one half hour session.
As in the ﬁrst two experiments, all participants reported normal or corrected-to-normal
vision and in addition all participants were native English speakers.

Apparatus and stimuli. The apparatus was the same as in the ﬁrst two
experiments, with the exception that responses were made using the E-Prime button box

and microphone. The microphone was positioned directly in front of the participant

71

approximately two to three inches from the participant’s mouth and the button box was
positioned just beyond the microphone in easy reach of the participant. The stimuli were
also similar to those used in the earlier experiments. Instructional cues were FIRST,
SECOND, NUMBER, and LETTER. The two target stimuli were the wide, light blue
rectangles from the ﬁrst experiment varying only in terms of height, either tall (48 pixels)
or short (32 pixels). The combination of four task instructional cues and two targets
resulted in eight unique trial types.

Procedure. The trial, block, and experiment procedures paralleled that of the ﬁrst
experiment. Instructional cues appeared 500 ms before the onset of the target and both
stimuli remained on the screen until a response (button press or voice key activation) was
detected. The screen then remained blank during the 100 ms RC1. Blocks were
constructed of 128 trials. In order to equate the number of possible task repetitions with
those in Experiment 1, trials were presented randomly without replacement for eight
examples of each of the eight trial types (16 instances of each task) for 64 trials (equal to
the number of possible trial types in Experiment 1). Then the cycle was repeated.

The experimental session began with an instruction and practice period during
which participants were introduced to and given practice with each of the tasks in
separate blocks of 16 trials. Tasks were presented in a set order: First, Second, Number,
and Letter. Responses to the First and Second tasks were button presses made with the
ﬁrst and second digits, respectively. The responses to the Number task were “one” and
“two” and to the Letter task were “A” and “B”. Stimulus to response mappings for each
task were counterbalanced across subjects. During the practice for the two vocal tasks,

participants were given instructions on the appropriate placement of and volume

72

necessary for the microphone. Participants then completed two blocks of trials in which
all four tasks were presented in a randomly mixed order. During the ﬁrst mixed block of
practice trials, participants were given verbal feedback and redirection by the
experimenter in addition to having a “cheat sheet” presented at the top of the screen that
indicated the appropriate response for each task and stimulus value. Participants were
encouraged to work slowly and accurately in order to correctly memorize the appropriate
responses for each of the tasks. During the second mixed block of practice trials,
participants were encouraged to work more quickly. Participants then complete ten
experimental blocks of trials. As before, participants were given feedback at the end of
each block as to the accuracy and average response time for that block and recorded this
information on sheets provided for this purpose. Participants were again given goals of
working as accurately and quickly as possible, making sure that their accuracy stayed
above 90% and trying to decrease response times from block to block throughout the
experiment.

Unlike the ﬁrst two experiments, the experimenter remained in the room with the
participant throughout the experiment. The experimenter sat to the right and behind the
participant and coded the participant’s vocal responses using ﬁve keys on the standard
keyboard: one key for each appropriate vocal response and a ﬁnal key used to indicate
microphone errors such as failure to register an appropriate vocal response or triggering
to non response noises. Experimenters received practice in coding vocal responses with
at least one practice subject prior to the start of data collection. In addition, audiotape
recordings of the sessions were made for later checking of the accuracy of the vocal

responses.

73

3.1.2 Results

Data processing. Experimenter coding of the participants’ vocal responses was
cross-checked to audio tapes of each experimental session. Trials on which the
participant stuttered or made multiple utterances, typically occurring in the form of self
correction, were considered microphone errors. Trials on which the participant clearly
stated a single response were counted as useable trials and coded as either correct or
incorrect for the current task and stimulus. Following coding of the vocal responses, the
data processing procedures were the same as the ﬁrst two experiments. The ﬁrst two
trials in each block were removed. Response accuracy was high (96.9%, range 93.9% to
99.0%). Error trials (both microphone and response errors) and the two subsequent trials
were removed leaving 91.4% (range 83.7% to 97.1%) of the trials for the calculation of
mean response times. In addition the response time trimming procedures based on a low
cutoff of 200 ms and a high cutoff of three standard deviations above the mean for trials
conditions resulted in the further removal of 2.2% (range 1.4% to 2.7%) of the data.
Table 4 gives means and standard errors for response times (A) and accuracies (B) for the
16 trial conditions created based on the task indicated on Trial N and Trial N-l. The
calculations were based on an average of 70 (range 48 to 89) trials per cell in the design.

Switch cost analysis. Again, the primary analysis focused on addressing the
question of whether task transitions were affected by the similarity of the tasks being
performed. Switch costs were calculated by subtracting task repetition conditions from
task switch conditions, separately for switches from similar tasks and switches from
dissimilar tasks. These values were entered into a 2 similarity (similar or dissimilar) by 4

task (First, Second, Number, or Letter) repeated measures ANOVA. The results mirror

74

A

 

 

 

Trial N
Trial N- 1 First Second Number Letter
First 5 5 1120 764142 84713 8 829144
Second 728145 580129 873148 838145
Number 764168 828155 622117 777137
Letter 759168 847153 793141 617116
B

Trial N
Trial N-l First Second Number Letter
First 98.5104 97.6104 96.5107 97.4105
Second 97.2104 97.6105 96.0107 97.6105
Number 97.2105 97.9104 98.9103 97.8104
Letter 98.1104 97.8105 97.4104 98.9103

 

Table 4. Mean 1 Standard Error for Response Times (A) and Accuracies (B) for Trial
Conditions Based on Task on Trial N and Trial N—l for Experiment 3

those found in Experiment 1. The main effect of task similarity was signiﬁcant, F (1,19) =
15.4, p < .01, MSE = 8640.1, with switch costs smaller when the task transition involved
a switch from a similar task (M = 173 ms) than a switch from a dissimilar task (M = 231
ms). The magnitude of the switch costs is somewhat smaller than the comparable
conditions in Experiment 1, potentially due to the fact that the attended stimulus
dimensions and the decision being made were the same for each of these four tasks.
Additionally, the similarity effect did not interact with task, F(3,57) = 0.8, p = .51, MSE =

4036.6, indicating that the similarity effect was equivalent for all tasks. Nor did switch

75

costs differ based on the task for which they were calculated, F (3,57) = 0.5, p = .69, MSE
= 16062.1. Figure 6 shows the switch costs for switches from similar and dissimilar tasks

separated by task performed on Trial N.

400
350 -, . . -, ,_ -_
300 +4 —- -1. -1 1 .11“. 1 “1. 1____-____.
250 .
200
150 r
100 .
50 4
O

 

 

 

Switch Cost (ms)

 

 

 

 

 

 

 

 

 

 

First Second Number Letter
Task

Cl Switch from Similar I Switch from Dissimilar

Figure 6. Mean switch costs when switching from similar and dissimilar tasks for each
task type in Experiment 3. Error bars represent 95% conﬁdence intervals calculated as in

Experiment 1.

Once again, the response accuracy data were examined in order to rule out the
possibility that a speed accuracy trade-off was driving the similarity effect. The pattern
was very similar to Experiment 1, with response accuracy highest on task repetition trials
(M = 98.5%) followed by switches from similar tasks (M = 97.5%) and switches from
dissimilar tasks (M = 97.3%). This pattern of data again rules out the possibility that the
similarity effect found in the response time data might result from a speed accuracy trade-
off. The 2 similarity (similar or dissimilar) by 4 task (First, Second, Number, or Letter)

repeated measures AN OVA on the response accuracy data showed no signiﬁcant effects.

76

 

Stimulus transition analysis. As in the ﬁrst two experiments, the possibility that
the similarity effect found in the switch cost analysis might arise from processes
associated with stimulus level transitions rather than task transitions was considered.
Unlike the ﬁrst two experiments, only two stimuli were used in the current experiment
and the same decision (height) was made in all four tasks. Therefore the stimulus
transition analysis was relatively simple. Again two separate analyses were conducted.
The ﬁrst considered the two task switch conditions (switch ﬁ'om similar and switch ﬁ'om
dissimilar) thus allowing for the consideration of the similarity effect in light of possible
stimulus transition effects. Differences in the mean response times for the stimulus
repetition and stimulus switch conditions were relatively small (6 ms and -10 ms for the
switch from similar and dissimilar tasks, respectively) suggesting that stimulus transition
properties had little impact when a task transition occurred. This observation was
conﬁrmed in the 2 stimulus transition (repetition or switch) by 2 similarity (similar or
dissimilar) repeated measures ANOVA. The main effect of stimulus transition was non
signiﬁcant, F (1, 19) = 0.05, p = .83, MSE = 4724.7. In addition the interaction of the
stimulus transition and task transition was non signiﬁcant, F (1, 19) = 1.9, p = .18, MSE =
2821.4, conﬁrming that the similarity effect seen in the ﬁrst analysis was not a function
of stimulus repetition.

A separate analysis conducted to examine the stimulus transition effect in the case
of task repetitions conﬁrmed that participants were signiﬁcantly faster to perform a trial
on which the stimulus was repeated from the previous trial showing a repetition beneﬁt of
55 ms, F(1, 19) = 19.0,p < .01, MSE = 6378.0. As noted previously when the task

repeats, a stimulus repetition also involves a response repetition, therefore the current

77

effect may be viewed as a beneﬁt of response repetition. Interestingly, this stimulus or
response repetition effect did interact with task, F (3, 57) = 4.3, p < .01, MSE = 1838.9.
Stimulus repetition beneﬁts were larger in the First and Second ﬁnger tasks (62 ms and
92 ms respectively) than in the Number and Letter tasks (47 ms and 23 ms, respectively).
This difference in stimulus repetition beneﬁt across response modalities was conﬁrmed in
an analysis of the repetition beneﬁt comparing manual and vocal tasks, t(19) = 2.4, p <
.05. This ﬁnding would further suggest that response repetition rather than stimulus
repetition is the cause of the beneﬁt since response characteristics rather than stimulus
encoding demands differed between the two groups of tasks.
3.1.3 Discussion

The results of the current experiment are strikingly similar to those of Experiment
1. The task similarity effect replicated in an entirely new group of tasks for which the
deﬁnition of similarity depended upon the response component of the task set rather than
the attentional component. This ﬁnding strengthens the argument that the higher order
variable of task similarity rather than any characteristic speciﬁc to the individual tasks led
to the differences in switch costs found in both experiments. Further, this ﬁnding
reinforces the notion that task space organized based on relationships among tasks can
have regular and predictable effects on executive control processes. While ﬁtting nicely
in the current framework, this ﬁnding might be considered surprising based on the known
interference effects in dual task situations involving responses in the same modality
(Pashler, 1990). Consideration of how the task similarity effect may provide insight into
differences between task switching paradigms and other experimental paradigms

involving multiple tasks will be considered more fully in the General Discussion.

78

The mechanisms for the task similarity effect again cannot be determined within
the current experimental design and may involve either controlled or automatic processes.
Similarity of the responses between two tasks may reduce the task set reconﬁguration
that must take place in order to prepare for an upcoming task. If the participant has just
made a vocal response on the Letter task, saying “A” for example, the switch to the
number task does not require the participant to reconﬁgure the motor response system to
a new modality, but rather the Number task involves the same response modality.
Alternatively residual activation associated with the task set for the previous task may
result in a beneﬁt in performing a similar response through priming of the response
modality of the current task. Thus, performing the First ﬁnger task may lead to residual
activation in the neural regions that control not only movement of the ﬁrst ﬁnger, but also
the regions associated with moving the second ﬁnger. These possible mechanisms are

examined in Experiment 4.

3.2 Experiment 4

The previous experiment demonstrated that task similarity, when deﬁned in terms
of response modality, facilitates switching between tasks. Experiment 4 more closely
examines the potential controlled and automatic processes that might drive the similarity
effect. Experiment 4 follows the same logic as Experiment 2, with manipulations of the
timing parameters of the task being implemented to isolate automatic and controlled
processes. The results from Experiment 2 indicate that the similarity effect deﬁned on
attentional control settings disappear with increasing preparation time, suggesting that the
attentional control setting for a task is a function of a controlled preparatory process. If

task sets are composed of component processes that are independent, then there is no

79

reason to believe that the task similarity effect when similarity is deﬁned in terms of the
response modality will behave in the same way as when similarity is deﬁned in terms of
the attentional control setting. Meiran (2000a) demonstrated that manipulations of the
complexity of the response set could be seen in the size of the residual switch costs
presumably unrelated to preparation interval. On the other hand, the effects of changes in
the stimulus set were linked to changes in preparation interval. These results suggest that
the similarity effect deﬁned in terms of response modality may be impacted by changes
in overall delay interval.

3.2.1 Method

Participants. Participants were drawn from the same population as Experiment 3.
Seventy-six individuals participated in this experiment. Data were removed from four
participants who either failed to maintain an alert state throughout the session or failed to
restrain task irrelevant vocalization resulting in excessive microphone errors. Thus there
were 24 participants in each between-subj ects condition.

Procedure. The stimuli and apparatus were the same as those in Experiment 3
and the tasks were a subset of the tasks used in Experiment 3: FIRST, SECOND, and
NUMBER. The trial procedure was the same except for the manipulation of the timing
intervals. The timing intervals were manipulated between subjects and involved the same
timing intervals as were used in Experiment 2: 100 ms RCI/SOO ms CTI (Group US,
identical to Experiment 3), 700 ms RCI/SOO ms CTI (Group 7/5), and 100 ms RCI/1100
ms CTI (Group 1/11).

The session structure was similar to previous experiments. Following practice

with individual task and mixed task blocks, participants performed 12 blocks of trials for

80

which data were collected. The trials were run in blocks of 96 trials (8 one target pairs
run in 4 cycles of 24 trials). This presentation was intended to equate the total possible
repetitions of the same task with that possible in Experiment 2. The entire session lasted
between 60 and 90 minutes depending on the trial length, which was a function of the
timing manipulation. The experimenter again remained in the room throughout the
session and coded the participants’ vocal responses during data collection using three
keys on the number pad to code each of the vocal responses and microphone errors. The
data collection portion of the experiment was audiotaped.
3.2.2 Results

Data processing. Experimenter coding of the vocal responses made by the
participants was compared to audio tapes of the experimental session for blocks on which
the experimenter indicated at the time of coding that a coded response was incorrect.
This restricted checking procedure was ad0pted after complete cross-checking of two
data ﬁles per experimenter suggested that experimenter coding and awareness of coding
errors was highly accurate (approximately 99.9%). Since the participants’ vocal
responses were also highly accurate, the chance of mistakenly including a trial that
should have been excluded from the data analysis (one on which the participant
responded incorrectly and the experimenter miscoded the response) was extremely low.
Thus the less extensive cross-checking of data ﬁles was deemed appropriate.
Microphone errors and incorrect vocal responses were coded in the same manner as
Experiment 3. Following coding of the vocal responses, the data processing procedures
were the same as the earlier experiments. The ﬁrst two trials in each block were

removed. Response accuracy was high (97.6%, range 93.8% to 99.8%). Error trials

81

(both microphone and response errors) and the two subsequent trials were removed
leaving 91.7% (range 81.3% to 98.8%) of the trials. In addition the response time
trimming procedures based on a low cutoff of 200 ms and a high cutoff of three standard
deviations above the mean for trial conditions based on the tasks indicated on Trial N and
Trial N-l resulted in the further removal of 2.1% (range 1.0% to 2.8%) of the data. Table
5 gives the means and standard errors for response times (A) and accuracies (B) for the
nine trial conditions separated by the between subjects timing condition. Mean response
time calculations per participant were based on an average of 112 (range 73 to 142) trials
per cell in the design.

Switch cost analysis. As in Experiment 2, the primary analysis of this experiment
examined the effect of timing condition on the similarity effect. Switch costs were
calculated for the First ﬁnger and Second ﬁnger tasks where both switch from similar
task and switch from dissimilar task conditions could be deﬁned. A 2 similarity (similar
or dissimilar) by 2 task (First ﬁnger or Second ﬁnger) by 3 timing condition (Group US,
Group 7/5, or Group 1/11) mixed ANOVA was conducted.

While the comparison of results from Experiments 1 and 3 show the same pattern
of effects of similarity even with the changes in tasks that altered how similarity was
being deﬁned, several interesting differences between the results of the current
experiment and the parallel analyses from Experiment 2 can be seen. These results
indicate differences between response processes (current experiment) and attentional
processes (Experiment 2) involved in task switching. The main effects of all three
variables were signiﬁcant. The similarity effect was in the expected direction with

switches from similar tasks showing smaller switch costs (M = 125 ms) than switches

82

from dissimilar tasks (M = 191 ms), F(1, 69) = 58.8,p < .01, MSE = 5333.9. In addition,

switch costs were larger for the Second ﬁnger task (M = 180 ms) than for the First ﬁnger

task (M: 136), F(1, 69) = 37.5,p < .01, MSE = 3751.8. Finally, the timing condition had

 

 

 

 

 

A
Timing Condition 100/5 00

Trial N
Trial N- 1 First Second Number
First 543116 702121 741121
Second 669121 560115 758120
Number 748128 846132 61211 7
Timing Condition 700/500

Trial N
Trial N-l First Second Number
First 525131 724157 654138
Second 652149 539132 651139
Number 71 8162 790165 570125
Timing Condition 100/1 100

Trial N
Trial N-l First Second Number
First 471115 564125 558117
Second 548123 472114 564119
Number 556128 597133 524114

 

Table 5. Mean 1 Standard Error for Response Times (A) and Accuracies (B)for Trial
Conditions Based on Task on Trial N and Trial N-l Separated by Timing Condition for

Experiment 4

83

Table 5 (cont’d).

 

 

 

 

 

B
Timing Condition 100/500

Trial N
Trial N-1 First Second Number
First 98.1103 98.3102 98.7103
Second 98.2104 97.7103 97.6105
Number 97.4104 98.0103 99.1103
Timing Condition 700/500

Trial N
Trial N-l First Second Number
First 98.1103 97.0104 96.7108
Second 97.0105 96.8104 95.6109
Number 96.4104 96.4106 97.8107
Timing Condition 100/1100

Trial N
Trial N-1 First Second Number
First 97.8104 97.4105 98.5102
Second 97.3105 97.3105 98.0103
Number 97.3105 97.8105 99.0102

 

a signiﬁcant effect on switch cost F(2, 69) = 4.4, p < .05, MSE = 64513.5, such that the
switch costs were reduced for Group 1/11 (M = 95 ms) compared to the other two
conditions (M = 190 and 189 ms for Group US and Group 7/5, respectively). Unlike
Experiment 2 where the overall switch costs did not differ with respect to timing of the

trial presentation, the current results indicated that switch costs decreased with an

84

increase in the preparation interval, a ﬁnding in line with previous research on the effects
of preparation on switch costs (Meiran, 1996).

The critical two-way interaction of similarity and timing condition was
signiﬁcant, F(2, 69) = 9.2, p < .01, MSE = 5333.9. This interaction can be seen in Figure
7 showing the mean switch costs for switches ﬁ'om similar tasks and switches from
dissimilar tasks separated by timing condition. The pattern of data suggests decreases in
the similarity effect across both changes in delay interval and changes in preparation
interval. Again this interaction was broken down in two separate AN OVAs in order to

isolate the effects due to automatic processes and controlled processes.

400
350 -1 ,, ,7
300 1“
250 ,
200 .1.
150 ._
100 g..-
50 .____ ..

 

 

 

 

Switch Cost (ms)

 

 

 

 

 

 

 

 

 

 

 

 

 

 

Group 1/5 Group 7/5 Group 1/11

El Switch from Similar I Switch from Dissimilar

Figure 7. Mean switch costs when switching from similar and dissimilar tasks for each
timing condition in Experiment 4. Error bars represent 95% conﬁdence intervals
calculated as in Experiment 2.

85

Comparing Group US to Group7/5 found that increasing the overall delay interval
between the response to the previous trial and the onset of the target for the current trial
signiﬁcantly decreased the size of the similarity effect, F (1, 46) = 4.3, p < .05, MSE =
5930.7. Additionally, comparing Group 7/5 and Group 1/11 found that increasing the
preparation interval from onset of task cue to onset of target stimulus also decreased the
similarity effect, F (1, 46) = 5.0, p < .05, MSE = 4819.0. These ﬁndings conﬁrm that
increases in delay interval as well as increases in preparation interval result in decreases
in the similarity effect when similarity is deﬁned in terms of the response modality
(manual or vocal). Unlike the ﬁndings from Experiment 2 where the reduction in the
similarity effect resided completely in the controlled processes, the current analyses
indicated both automatic and controlled processes impact the similarity effect.

The two-way interaction of similarity and timing demonstrated that both active
preparation and residual activation from previous trials inﬂuence the response component
of the task set during task switching. Additionally, there was also a three-way interaction
of similarity, timing condition, and task (First ﬁnger or Second ﬁnger), F (2, 69) = 4.7, p
< .05, MSE = 1438.3. The nature of this three factor interaction can be seen in examining
the ANOVAs isolating controlled and automatic processes. In the comparison of Group
7/5 and Group 1/11 examining controlled processes changing over increased preparation
interval, the decrease in the similarity effect associated with the increase in preparation
interval did not interact with the task variable, F(1, 46) = 1.3, p = .25, MSE = 1505.7,
such that both First ﬁnger and Second ﬁnger tasks resulted in a decreases in the similarity
effect of 57 and 32 ms, respectively. On the other hand, in the comparison of Group US

and Group 7/5 the increase in delay interval resulted in signiﬁcantly different effects on

86

the similarity variable for the First ﬁnger and Second ﬁnger tasks, F (1, 46) = 6.7, p < .05,
MSE = 1978.2. The First ﬁnger task had a signiﬁcantly smaller decrease in the similarity
effect (12 ms) than did the Second ﬁnger task (78 ms). This interaction of task with the
variables of similarity and timing is a departure from ﬁndings in the earlier experiments
where interactions of the similarity effect with tasks were not discovered. Understanding
why the task interacted with timing and similarity will be discussed more thoroughly
below and may help to shed light on how task speciﬁc characteristics affect the larger
task space in which executive control is acting.

As in previous experiments, the response accuracy data were examined in order to
rule out the possibility of a speed accuracy trade-off. The pattern was very similar to the
previous experiment, with response accuracy highest on task repetition trials (M = 98.0%)
followed by switches from similar tasks (M = 97.7%) and switches from dissimilar tasks
(M = 97.2%). The 2 similarity (similar or dissimilar) by 4 task (First, Second, Number,
or Letter) by 3 timing condition (Group US, Group 7/5, or Group 1/11) mixed ANOVA
showed no signiﬁcant effects. This pattern of data again rules out the possibility that the
similarity effect found in the response time data might result from a speed accuracy trade-
off.

Stimulus transition analysis. The analysis of stimulus transition effects closely
mirrored those from Experiment 3. Analysis of the stimulus transition effect was again
performed separately for task repetitions and task switches. For the First ﬁnger and
Second ﬁnger tasks where task switches from similar and dissimilar tasks could be
deﬁned, the effect of stimulus repetition was analyzed as a function of similarity of task

transition to insure that the similarity effect found in the switch cost analysis could not be

87

attributed to stimulus repetition effects. The 2 similarity (similar or dissimilar) by 2
stimulus transition (repetition or switch) by 2 task (First or Second) by 3 timing condition
(Group US, Group 7/5, or Group 1/11) mixed ANOVA, did produce a signiﬁcant
interaction of similarity by stimulus transition, F(1, 69) = 7.4, p < .01, MSE = 2285.5,
such that the beneﬁt for repeating the stimulus was present (15 ms) for switches from the
similar, other ﬁnger task and absent (-6 ms) for switches from the dissimilar, vocal task.
This effect may reﬂect the fact that a stimulus repetition for a switch from First ﬁnger to
Second ﬁnger task (or vice versa) involved a response with the same hand although a
different ﬁnger on that hand as opposed to a stimulus switch involving a change in both
the ﬁnger and the hand between trials. As discussed more thoroughly below, sequential
responses using the same hand may beneﬁt from the overlap of neural representation of
the two effectors, while switching hands would not provide the same beneﬁt. While this
signiﬁcant interaction is in the direction that would increase the similarity effect when
comparing switches between similar vs. dissimilar tasks, the size of the stimulus
repetition effect was relatively small and cannot account for the larger similarity effect.
For the task repetitions as seen Experiment 3, there was a signiﬁcant beneﬁt for
stimulus repetition of 38 ms, F(1, 69) = 33.2, p < .01, MSE = 4778.0. The stimulus
repetition beneﬁt differed signiﬁcantly for the three tasks, F (4, 138) = 40.0, p < .01, MSE
= 1363.2, with the beneﬁt being largest for the Second ﬁnger task (73 ms), smaller for the
First ﬁnger task (46 ms), and non existent for the Number task (-4 ms). Stimulus
transition also interacted with the timing condition, F (2, 69) = 4.5, p < .05, MSE =
4778.0, with the stimulus repetition beneﬁt being signiﬁcantly smaller for Group 1/11 (12

ms) than for both Group US and Group 7/5 (60 ms and 43 ms, respectively).

88

Trial N-2 sequence analysis. As in Experiment 2, the data were again divided into
conditions based on the sequence of tasks of Trials N, N-l, and N-2. Table 6 gives

means and standard errors for response times (A) and accuracies (B) for each of the 27

conditions generated by combining the three tasks across each trial sequence.

 

 

 

 

A
TrialNTask
First Second Number

Trial N-l First Second Number First Second Number First Second Number
Group US

Trial N-2

First 556120 655120 776131 671121 561116 857134 725122 740123 610113

Second 548117 651122 753133 709121 564117 863134 709122 737122 617119

Number 529116 698125 710125 728125 556117 817132 790125 808120 611117
Group 7/5

Trial N-2

First 500129 647152 744169 711158 546132 798162 636134 641136 573126

Second 541134 633144 720160 718155 525136 815169 654140 644142 573128

Number 529133 676154 692162 742159 546133 752166 670139 678141 559122
Group l/ll

Trial N-2

First 472116 558126 565129 559128 468115 601137 558118 550118 520114

Second 467116 542123 564129 573127 470114 609131 552116 565121 517114

Number 473117 547124 539128 560124 479116 578132 566119 582121 538114

 

Table 6. Mean 1 Standard Error for Response Times (A) and Accuracies (B) for Trial

Conditions Based on Task on Trial N, Trial N-l, and Trial N-2 Separated by Timing
Condition for Experiment 4

89

Table 6 (cont’d).

 

 

 

 

B
TrialNTask
Second Number
Trial N-l Number First Second Number First Second Number
Group US
Trial N-2
First 97.5105 99.0103 97.7105 98.0105 98.7104 99.1103 99.5103
Second 98.1105 97.2106 98.0104 98.1104 97.2106 99.0104 97.6108 99.0104
Number 98.7104 97.4105 97.9105 97.6106 98.7103 98.2104 95.8109 98.8105
Group 7/5
Trial N-2
First 95.4107 96.9108 96.7106 96.3107 96.8109 96911.1 97.8109
Second 97.4104 96.3107 96.8106 96.9107 96.2108 96711.1 96.2107 97.2106
Number 98.5104 97.3105 97.4104 96.6106 96.3108 95611.0 94011.0 98.1108
Group l/ll
Trial N-2
First 97.3106 98.2106 97.3105 97.7106 98.9103 98.3104 98.8104
Second 97.6106 96.9107 96.6108 96.6109 97.8105 98.4104 97.8105 98.8104
Number 98.4105 97.7106 97.4105 97.7106 97.7107 98.2104 98.0105 99.5103

 

The effect of the three trial sequences was assessed in a 3 task sequence

(alternating tasks, unique tasks, or repeat-switch tasks) by 2 similarity of Trial N and

Trial N-l (similar or dissimilar) by 2 task (First ﬁnger or Second ﬁnger) by 3 timing

condition (Group US, Group 7/5, or Group 1/11) mixed ANOVA. While the effect of

sequence was found to be signiﬁcant, F (2, 138) = 21.7, p < .01, MSE = 4478.6, the

pattern of data was clearly qualiﬁed by a signiﬁcant interaction with similarity, F (2, 138)

90

= 7.1, p < .01, MSE = 3086.4. There were no other signiﬁcant interactions involving the
sequence effect. When the task switch prior to Trial N involved a switch from the
dissimilar, Number task, the sequence effect showed the same pattern of response times
as was seen in the Experiment 2: slowest responding on alternating tasks sequence,
intermediate for unique tasks sequence, and fastest for repeat-switch tasks sequence (Ms
= 729, 715, and 681 ms, respectively). However, when the task switch prior to Trial N
involved a switch from the similar task, First ﬁnger to Second ﬁnger or Second ﬁnger to
First ﬁnger, the pattern differed: response times were now intermediate for the alternating
tasks sequence, slowest for the unique tasks sequence, and again fastest for the repeat-
switch tasks sequence (Ms = 643, 659, and 628 ms, respectively). So while the beneﬁt
for switching away from a task that has just been repeated held across all conditions, the
comparison of the alternating tasks and unique tasks conditions shows that the similarity
of the intervening task may have an inﬂuence on whether so-called backward inhibition
effects are present in the current set of tasks. The complimentary analysis on response
accuracy data did not show this complex interaction of similarity by sequence. The only
signiﬁcant effect in this analysis was the sequence effect, F (2, 138) = 4.7, p < .05, MSE =
.00065, with the pattern of data the same as was found in Experiment 2, least accurate
responding on alternating tasks, intermediate for unique tasks, and best for repeat-switch
tasks, (M = 97.0%, 97.3%, and 97.7%, respectively). These ﬁndings and the comparison
to the comparable analysis from Experiment 2 will be considered in the General

Discussion.

91

3.2.3 Discussion

The results of Experiment 4 form an interesting contrast to the parallel analyses
performed in Experiment 2. When task similarity is deﬁned in terms of response
modality, the size of the task similarity effect decreased not only with increasing
preparation interval, which suggests controlled processes are involved in producing the
similarity effect, but also with increasing overall delay interval. The interpretation of this
second result is that some automatic processes are involved in the task similarity effect in
the current group of tasks. Since similarity is deﬁned in terms of the response modality
the implication is that automatic processes associated with task switching costs arise at
least in part from the response component of the task set. This ﬁnding is in contrast to
the results in Experiment 2 where only an increase in preparation interval affected the
magnitude of the task similarity effect, suggesting automatic processes that dissipate over
the present time interval (500 to 1100 ms) are not involved in switching between the
attentional control settings of the task set. The differences between the current results
and those from Experiment 2 are important because they support the idea that the
processes involved in task switching may be localized to speciﬁc components of the task
set. This idea will be considered more fully in the General Discussion.

While the current results clearly indicate that both controlled and automatic
processes are involved in the response component of task switching, it appears to be a
more complicated matter to understand exactly what those effects might be. The three-
way interaction of task (First ﬁnger or Second ﬁnger), similarity, and timing condition
should be considered as it may provide some insight into when automatic and controlled

processes inﬂuence switching between responses. Motor control for the ﬁrst and second

92

ﬁngers is unequal, with individuals having better control over the ﬁrst ﬁnger, evidenced
by a greater degree of independent motion for the ﬁrst ﬁnger (Hager-Ross & Schieber,
2000). The larger switch costs associated with the Second ﬁnger task in the current
experiments likely reﬂect this unequal degree of control. As the less controlled and less
independent response, the Second ﬁnger task may receive greater priming from recent
ﬁrst ﬁnger movement than vice versa, thus explaining the large similarity effect for the
Second ﬁnger task in Group US. That this similarity effect might be reduced as a
function of increased delay interval is not surprising. Recent fMRI research shows a
positive linear relationship between the frequency of ﬁnger tapping and the magnitude of
the response in primary motor regions (Rao et al., 1996). This relationship indicates a
buildup in the activation in motor areas with rapid movement suggesting that the neural
activity associated with a ﬁnger movement dissipates over time resulting in less buildup
of activation as movements are spaced ﬁrrther apart. In this case, the longer delay
interval in Group 7/5 may allow for more dissipation of activity associated with the
recent ﬁrst ﬁnger movement and thus provide less priming of the second ﬁnger
movement. Such a mechanism would account for the changes in the similarity effect
with a change in delay interval for the Second ﬁnger task. However, the better controlled
ﬁrst ﬁnger response may not receive the same sort of beneﬁt following the Second ﬁnger
task. While this account is speculative, the current data suggest such an asymmetry may
be occurring. The effect of stimulus repetition was greater for the Second ﬁnger task (M
= 24 ms) than for the First ﬁnger task (M = 8 ms). Thus when a same hand response is

involved in responding for Trial N-l and Trial N, allowing for overlap in neural

93

representations for the ﬁngers involved and priming of motor regions, the beneﬁt was
greater for the Second ﬁnger task.

The preceding account for the differences between the First ﬁnger and Second
ﬁnger tasks might be tested by performing the experiments using different digits. The
ﬁrst ﬁnger and thumb both have a high degree of control of independent movement and
more separate neural representations (Hager-Ross & Schieber, 2000). Repeating the
experiment using responding with those digits might show similarity effects that are not
inﬂuenced by overall delay interval. On the other hand, using the third and fourth digits
where control is poor in both cases and there is greater overlap in neural representations
would be expected to result in similarity effects that decrease over increasing delay
intervals for both tasks. Additionally it would be interesting to discover how automatic
and controlled processes affect the vocal modality. These various possibilities indeed
paint a complicated picture of mechanisms affecting the response component of the task

set that are yet to be worked out.

94

4 General Discussion

Across four task switching experiments similarity between tasks was shown to
have a systematic inﬂuence on time associated with switching between tasks. In
Experiments 1 and 2 task similarity was deﬁned as a shared attentional control setting,
while in Experiments 3 and 4 similarity was deﬁned as a shared response modality.
Despite these differences in the implementation of the task similarity manipulation, the

conclusions were identical: task similarity facilitates task switching.
4.1 Task Switching as Movement through Task Space

The manipulation of similarity among tasks was designed to be a tool for
describing the larger task space in which these tasks occurred. The current
conceptualization of task space is founded on the belief that the mental representations on
which executive control processes act to coordinate task performance in multitask
environments should be conceived in terms of both the sets of cognitive operations
necessary to perform each task and the relationships among all of the tasks included in
the task environment. This proposal arises ﬁ'om an understanding that executive control
processes are separate from the cognitive operations directly involved in the performance
of any speciﬁc task. Therefore, limiting consideration to the representations of individual
tasks cannot provide for a ﬁrll accounting of the mental representations on which
executive control processes act. Rather, it is necessary as well to consider the
relationships among the individual tasks, since these relationships further deﬁne the
mental representations engaged by executive control processes. By analogy to the visual

attention literature, executive control may be thought of as orienting cognitive resources

95

to different locations in task space through processes of disengaging task sets, moving
through task space, and engaging other task sets.

With this belief as a starting point, I proposed a simple model of task space. Task
space can be represented in terms of a multidimensional space. The primary dimensions
of this space are deﬁned by the component cognitive operations that make up each of the
tasks included in the task environment. These dimensions include but are not limited to
the stimulus encoding, attentional processes, any mental transformations or judgments
that are carried out on stimuli, and response selection, preparation, and implementation.
The relationships among tasks can then be described in terms of their positions relative to
each other within task space. By analogy to feature-based conceptions of similarity, tasks
that share a component process are more similar, and overlap more in the task space, than
tasks that do not share the same component process.

The implication of this model for the study of task switching is that the controlled
and automatic processes involved in switching among tasks should be sensitive to the
structure of task space. In order to test this assumption, I manipulated task space and
measured the effect that these manipulations had on the processes involved in task
switching. Task similarity was chosen to manipulate task space because similarity is a
variable speciﬁcally deﬁned in terms of the relationships between two or more tasks
rather than as a characteristic speciﬁc to a particular task. Thus task similarity can be
used to represent the relationship between two tasks within task space. The more similar
two tasks are the closer they are in task space. In combination with the notion that the

executive control processes serve to move resources between locations in task space, this

96

model of task space predicts that greater similarity between two tasks should reduce the
time that it takes to switch between those tasks.

Across the four experiments presented here, this prediction was supported. The
discovery of the task similarity effect demonstrates that manipulations of task similarity
have systematic inﬂuences on the executive control and automatic processes acting
within that space. These ﬁndings further support the model of task space presented in the
introduction. Understanding that the executive control and automatic processes that
coordinate performance in multitask environments are directly impacted by the task
space in which they act should have powerful implications for the investigation and

understanding of these processes.
4.2 Linking Processes to Task Components

Along with providing a measuring stick for task space, the task similarity effect
also offers a new tool for linking various executive control and automatic processes to
speciﬁc components of task sets and task performance. As discussed in the introduction,
a variety of models propose either automatic or controlled processes or both as the
mechanism behind switch cost in task switching paradigms. While there has been some
speculation as to what speciﬁc task components are acted upon by various processes, the
empirical work in this area has been limited (Meiran, 2000b; Rubinstein et al., 2001).
The second critical component of the current research was the application of the task
similarity effect to this question.

Although task similarity might be deﬁned in a variety of different ways, the
current research does so based on shared cognitive components between tasks. This way

of deﬁning task similarity also provides the ability to localize the effect of task similarity

97

to a particular component of the task set. By implication then, any changes in the size of
the task similarity effect must be the result of processes that impact the particular
cognitive component on which similarity is deﬁned. Building on this logic, the current
research examined the impact of trial timing allowing for varying degrees of controlled
preparation for the upcoming task and passive dissipation of activity associated with the
previous task that might have an automatic inﬂuence on performance of the current task.
Based on the results of Experiments 2 and 4, the beginnings of a model linking
controlled and automatic processes to individual components of the task set can be
developed. Experiment 2 found that when similarity is deﬁned in terms of the attentional
control setting, the similarity effect does not change with variation in delay interval, but
completely disappears with a long preparation interval. This ﬁnding suggests that part of
the controlled task preparation that occurs once the upcoming task is known is the
establishment of the attentional control setting appropriate for performance of the task.
On the other hand, Experiment 4 demonstrated that when similarity is deﬁned by shared
response modality the similarity effect changes both as a result of preparation interval
and overall delay interval. These ﬁndings in unit suggest both controlled and automatic
processes are acting on the response component of the task set. In combination, the
results from these two experiments allow us to begin to link speciﬁc processes to
different components of the tasks that they inﬂuence. Controlled preparation for an
upcoming task must involve activating or enabling both the attentional systems necessary
to process the stimulus information relevant to the task and the motor systems involved in
responding appropriately to the task. On the other hand, only the response component of

the task appears to have an automatic inﬂuence in the form of priming from one trial to

98

the next that dissipates over time. Thus a model may be envisioned in which various
cognitive operations engaged in task performance are activated or deactivated by a
variety of separable controlled and automatic processes. Filling in the details of this

model of task switching will require further investigation.

4.3 Task Sequence Effects

In addition to the switch cost analyses, task performance for Experiments 2 and 4
was also analyzed as a function of the sequence of tasks over a three trial run. (The
comparable analyses were not performed for Experiments 1 and 3 due to the small
number of trials per cell in the four task versions of the experiments, which would make
interpretation of any pattern of results suspect.) The analyses focused on three
sequences: unique tasks, A-B-C; alternating tasks, C-B-C; and repeat-switch tasks, B-B-
C. The focus of previous studies of the Trial N-2 sequence effects has been on the
comparison of unique tasks and alternating tasks sequences as a marker of backward
inhibition. A series of studies by Mayr (2002; Mayr & Keele, 2000) has attributed slower
responding on alternating tasks to the inhibition of the task set that has recently been
switched away from following performance of Trail N-2, but must be reinstated on Trial
N. In addition to this apparent inhibition on alternating tasks sequences in comparison to
the unique tasks sequence, which serves as a control, there is also evidence for a
facilitation in responding on the repeat-switch tasks sequence as compared to the unique
tasks sequence (Arbuthnott & Frank, 2000; Arrington et al., 2001).

The comparisons of alternating tasks and unique tasks sequences produced a
complex pattern of results that differed across the two experiments. In Experiment 2, the

most general level of analysis showed slowed responding in the alternating tasks

99

sequence, thus generally replicating the pattern of results associated with the inhibition of
task set under conditions of alternating tasks. However, the sequence effect did enter into
several signiﬁcant interactions with the similarity, task, and timing interval variables
indicating a complex set of factors that inﬂuence this effect. While Mayr and Keele
(2000) found no changes in backward inhibition as a result of changing the timing of the
trials, the current ﬁndings indicate that under some circumstances temporal
manipulations do affect the alternating tasks effect. Recently, Arbuthnott and Woodward
(2002) also have found task speciﬁc ﬂuctuations in the alternating tasks effect. In
Experiment 4, the alternating tasks effect showed an interesting interaction with
similarity. The cost typically associated with returning to a recently abandoned task set
was found when a dissimilar task intervened (e. g. First — Number -— First). However, the
reverse effect was found when the similar task intervened (e. g. First — Second — First) and
response times were actually faster on these trials when compared to the unique tasks
sequence (6. g. Number — Second — First). This ﬁnding suggests that the alternating tasks
effect, which is considered as a marker of backward inhibition, is not universal. In
particular, when the tasks are highly similar, differing only in terms of the particular
effector (ﬁrst or second ﬁnger) with which the response is made, the need to implement a
mechanism of active inhibition of the previous task set may not occur when switching
between tasks. This conclusion is speculative and further research will need to be
conducted to more fully examine the conditions under which inhibitory processes are or
are not involved in task switching.

Unlike the previous comparison of alternating tasks and unique tasks sequences,

the comparison of repeat-switch tasks and unique tasks sequences showed a consistent

100

beneﬁt across all tasks and timing conditions in both experiments, indicating that the
effect is both replicable and robust. The mechanisms of this facilitated switching
following a task repetition are unknown. It may result from processes associated with the
ease of performing the task on Trial N-l due to the fact that it is a task repetition trial.

On task repetitions the activation of the task set may not need to be as strong for
appropriate task performance, thus switching away from that task set may prove to be an
easier process.

In sum, task performance in the multitask environment presented in these
experiments is clearly affected by the extended sequence of events prior to the
performance of a given trial including not only the preceding trial, but also more distal
trials. Producing a model of the processes inﬂuencing task performance in multitask

environments will necessarily required consideration of these issues.
4.4 Similarity in Task Switching and Other Dual Task Paradigms

The task switching paradigm is just one of a number of experimental
methodologies that have been used extensively in recent research focusing on issues of
control of cognitive processes in dual task environments. Other popular paradigms
include the psychological refractory period (PRP) paradigm and the attentional blink
(AB) paradigm. The experiments used in the current research exemplify the typical task
switching methodology (but see Allport & Hsieh 2001 for an example of an alternative
method for examining task switching). Individual trials include the presentation of a cue
indicating the task on a given trial (sometimes this element is excluded or combined with
the stimulus display), the stimulus, and the response. Only after the completion of each

of these elements in the trial is the next trial initiated. The tasks are thus presented

lOl

sequentially with each task demanding a speeded response. In the PRP paradigm, the
tasks also include the presentation of a stimulus and a speeded response, however the
presentation of the stimulus for the second task can occur prior to the response to the ﬁrst
task. This simultaneous presentation allows for overlap in the performance of the two
tasks. Finally the AB paradigm (Raymond, Shapiro, & Amell, 1992) imbeds the target
stimuli for the two tasks within a stream of distractor items presented in a rapid serial
visual presentation (RSVP) format. Responses to both tasks are withheld until the end of
the stimulus presentation and do not require speeded responding. Thus, while all three
paradigms require performance of two tasks close together in time, there are distinct
methodological differences among the paradigms.

These methodologies and associated phenomena traditionally have been
considered separately. However, there has recently been a ﬂurry of interest in relating
ﬁndings and theories developed from various dual task paradigms used to study control
of attention and performance: AB paradigm and PRP paradigm (J olicceur, Dell’Acqua, &
Crebolder, 2000; Ruthruff & Pashler, 2001), AB paradigm and task switching (Chun &
Potter, 2001), and PRP paradigm and task switching (Pashler, 2000). Understanding the
commonalities among these various paradigms appears to be critical in furthering our
models of executive control and automatic processes acting in a variety of dual task
environments both in and out of the laboratory. When comparing research from disparate
paradigms, effects such as the impact of similarity on task performance can serve as
critical tests of the continuity of the processes engaged across paradigms.

Some evidence for the impact that task similarity has on performance in the PRP

paradigm was introduced previously. When the two tasks involve responses in the same

102

modality (manual-manual) interference is greater than when the two tasks involve
responses in different modalities (manual-vocal; Pashler 1990). However, additional
research focusing on similarity in the stimulus modality (visual-visual vs. visual-
auditory) does not provide strong support for a parallel effect of similarity on dual task
interference when similarity is deﬁned in terms of sensory modality of the stimulus
(reviewed in Pashler 1998). These ﬁndings have been interpreted within the context of a
bottleneck at a central stage of processing. Stages prior to the bottleneck such as
perceptual processing are thought to occur in parallel and thus similarity of stimulus
features does not affect processing. On the other hand, processes that are involved in the
central bottleneck, in particular response selection, occur in a serial fashion. The fact that
similarity deﬁned at this stage of processing impacts the interference effect in the PRP
paradigm suggests that this serial processing stage is in some way affected by the
similarity of task components that require this stage of processing. The results ﬁom the
present experiments appear to differ from these results in that similarity facilitates
switching when it is deﬁned both in terms response modality and in terms of the
attentional control settings of the tasks, which would presumably be occurring prior to a
central bottleneck associated with response selection. Thus one might ask what the
difference is between these two very similar experimental paradigms that would lead to
the discrepant results. In the task switching paradigm, from one task to the next all task
components are processed in series due to the sequential presentation of stimuli. Perhaps
the serial nature of performance in task switching experiments is a key element in the

facilitation associated with similarity in task switching.

103

The role of similarity between tasks in the context of the AB paradigm has also
been studied recently. In a meta-analysis of AB experiments, Visser, Bischof, and
DiLollo (1999) examined lag-1 sparing (high level of accurate performance on T2 if it
appears immediately following T1) as a function of the type of switch occurring between
T1 and T2. They found lag-1 sparing occurred with greater regularity when there was
either no switch between T1 and T2 or a switch along a single dimension. Thus
similarity appeared to facilitate processing of T2 when it occurred in close temporal
succession (within approximately 100 ms) to T1. The relationship between lag-1 sparing
and similarity was however independent of the AB measured over a longer time interval
(approximately 200 to 600 ms). Visser et al. account for this ﬁnding of a similarity effect
at lag-1 in a model that proposes top-down control of perceptual ﬁlters such that when
two tasks are similar the imperative stimuli for both T1 and T2 can enter into the stream
of information processing simultaneously if presented close in time (a notion that appears
consistent with Folk et al.’s description of attentional control settings). This account
differs from the discussion above in that similarity appears to have its effect through
allowing for parallel processing of stimuli where as in the central bottleneck account of
the PRP effect, the similarity effect appears to be acting at the central bottleneck where
processing is serial. Interestingly, Visser et al. also note that the similarity results in the
AB task are in contrast to established interference effects resulting from similarity in dual
task situations. The results of the present set of experiments also provide an example of a
processing beneﬁt associated with similarity suggesting that proposal of greater
interference effects in dual task environments associated with similarity may need to be

reconsidered.

104

Clearly the role that similarity plays in task performance in dual task situations is
complex. The current experiments present a consistent ﬁnding that similarity facilitates
switching between tasks. However, there may also be multiple inﬂuences of similarity at
various stages of task performance in the current experiments as well. In Experiment 1
there was the hint from the analysis of within trial compatibility effects that similarity
may result in greater interference of task performance when similar stimulus dimensions
indicate incompatible responses. This effect was small and would need to be replicated
(which it was not in Experiment 2) before more speciﬁc claims can be made about the
ﬁnding. Further consideration of the role that similarity plays across various dual task

environments appears to be a useful avenue for future investigations.

4.5 Future Directions

One of the important contributions of the early stages of model development is
the questions that are generated. The current research represents a starting point ﬁom
which other questions may be fruitfully pursued. Let me speculate brieﬂy about what
some of these questions might be. Only two task components were examined in the
current research. While the task similarity effect was consistent across both components,
it will be important to consider ﬁuther task components. For example, would similarity
of different cognitive manipulations provide the same beneﬁt? J ersild’s (1927) initial
instantiation of the task switching method had participants switching between addition
and subtraction operations. Building on that method and the work of Badre et al. (2000),
it would be possible to construct a set of tasks that involved different mathematical tasks
(6. g. +3, +5, -2, -4) that were grouped in terms of similarity of the mathematical

operations being performed. Based on the task space model, switches between the two

105

addition task might be expected to occur more rapidly than between addition and
subtraction tasks.

The preceding discussion presents a model in which controlled and automatic
processes can be linked to separate components of task sets. While it appears possible to
separate out task components and examine the processes that act upon them, it would also
be useful to consider the interactions among task components. In the current set of
experiments, the similarity effect was found both when similarity was the result of shared
attentional control settings and when similarity was the result of shared response
modality. One interesting question that might be posed is whether the effect of similarity
at these different components of the task set would result in additive effects if presented
in the same task environment. If chosen properly, tasks might involve similarity along
more than one dimension in task space. For example, a combination of the tasks
involving shared attentional control settings (Height and Width vs. Height and Color)
with a manipulation of the response modality (manual or vocal) could allow for tasks that
are similar in terms of both attentional and response demands, similar on just one of the
task components, or similar on neither task component. In such a design it would be
possible to examine whether the similarity effects resulting from the manipulation of
speciﬁc aspects of the task set combined in an additive fashion.

Finally, the task space approach to task switching should be expanded beyond
simply looking for multiple and interacting effects of task similarity. It will become
important to consider other approaches to deﬁning task space in order to further explicate
the role that task space plays in determining how executive control is implemented. The

tasks that confront us in real world environments are often not easily deﬁned in terms of

106

similarity. For example tasks may better be deﬁned in terms of goals that they achieve or
the environments in which they have been experienced in the past. Consideration of this

larger application of task space to task switching should prove exciting.

4.6 Conclusion

Over the last decade, the task switching methodology has developed rapidly and
been widely applied to questions of cognitive control. Our understanding of the variables
that inﬂuence task switching has advanced. However, consideration of the underlying
representations has remained relatively limited. In this dissertation, I presented one
model of how task space might be organized in terms of a multidimensional space with
various cognitive operations serving as the basic dimensions of this space. I then
demonstrated that manipulating task space through varying the similarity among tasks
has systematic inﬂuences on the processes engaged when switching between tasks.
Finally, I used the task similarity effect to address questions of how and when executive
control and automatic processes act on the individual cognitive operations involved in
task performance. These explorations in task space supply structure to the expanding

ﬁeld of investigations in cognitive control.

107

References

Allport, A., & Hsieh, S. (2001). Task switching: Using RSVP methods to study an
experimenter-cued shift of set. In K. Shapiro (Ed.), The limits of attention: Temporal
constraints in human information processing (pp. 36-65). Oxford: Oxford University
Press.

 

 

Allport, A., Styles, E. A., & Hsieh, S. (1994). Shifting intentional set: Exploring
the dynamic control of tasks. In C. Umilta & M. Moscovitch (Eds), Attention and

Performance XV: Conscious and Nonconscious Information Processing (pp. 421-452).
Cambridge: MA: The MIT Press.

 

 

Allport, A., & Wylie, G. (1999). Task-switching: Positive and negative priming of
task-set. In G. W. Humphreys, J. Duncan, & A. Treisman (Eds), Attention, space and
action: Studies in cognitive neuroscience (pp. 273-296). Oxford: Oxford University
Press.

 

 

Allport, A., & Wylie, G. (2000). Task switching, stimulus-response bindings, and
negative priming. In S. Monsell & J. Driver (Eds), Control of Cognitive Processes (Vol.
18, pp. 35-70). Cambridge, MA: The MIT Press.

 

Altmann, E. M., & Gray, W. D. (2002). Forgetting to remember: The functional
relationship of decay and interference. Psychological Science, 13, 27-33.

 

Arbuthnott, K., & Frank, J. (2000). Executive control in set switching: Residual
switch cost and task-set inhibition. Canadian Journal of Experimental Psychology, 54,
33-41.

 

Arbuthnott, K. D., & Woodward, T. S. (2002). The inﬂuence of cue-task
association and location on switch cost and alternating-switch cost. Canadian Journal of
Experimental Psychology, 56, 18-29.

 

 

Arrington, C. M., Altmann, E. M., & Carr, T. H. (2001). Task switching and task
similarity: Old-task inhibition and new-task facilitation. Paper presented at the
Psychonomic Society Annual Meeting, Orlando, FL.

 

Ashby, F. G., & Perrin, N. A. (1988). Toward a uniﬁed theory of similarity and
recognition. Psychological Review, 95, 124-150.

 

Badre, D. T., Jonides, J ., Hernandez, L., Noll, D. C., Smith, E. E., & Chenevert,
T. L. (2000). Behavioral and neuroimaging evidence of dissociable switching
mechanisms in executive functioning Paper presented at the Cognitive Neuroscience
Society Meeting, San Francisco.

 

 

108

Chun, M. M., & Potter, M. C. (2001). The attentional blink and task switching
within and across modalities. In K. Shapiro (Ed.), The limits of attention: Temporal
constraints in human information processing (pp. 2035). Oxford: Oxford University
Press.

 

Collins, A. M., & Loftus, E. F. (1975). A spreading-activation theory of semantic
processing. Psychological Review, 82, 407-428.

 

Collins, A. M., & Quillian, M. R. (1969). Retrieval time ﬁom semantic memory.
Journal of Verbal Learning and Verbal Behavior, 8, 240-248.

Dagenbach, D., & Carr, T. H. (1994). Inhibitory processes in perceptual
recognition: Evidence for a center-surround attentional mechanism. In D. Dagenbach &
T. H. Carr (Eds), Inhibitory processes in attention, memory, and language (pp. 327-357).
San Diego, CA: Academic Press.

De J ong, R. (2000). An intention-activation account of residual switch costs. In S.
Monsell & J. Driver (Eds), Control of Cognitive Processes (Vol. 28, pp. 357-376).
Cambridge, MA: The MIT Press.

 

De Jong, R., Berendsen, E., & Cools, R. (1999). Goal neglect and inhibitory
limitations: Dissociable causes of interference effects in conﬂict situations. Acta
Psychologica, 101, 379-394.

 

Egly, R., Driver, J ., & Rafal, R. D. (1994). Shifting visual attention between
objects and locations: Evidence from normal and parietal lesion subjects. Journal of
Experimental Psychology: General, 123, 161 -1 77.

Felfoldy, G. L. (1974). Repetition effects in choice reaction time to
multidimensional stimuli. Perception & Psychophysics, 15, 453-459.

 

Folk, C. L., Remington, R. W., & Johnston, J. C. (1992). Involuntary covert
orienting is contingent on attentional control settings. Journal of Experimental
Psychology: Human Perception and Performance, 18, 1030-1044.

 

Garner, W. R. (1974). The processing of information and structure. Potomac,
MD: Lawrence Erlbaurn Associates.

 

Goschke, T. (2000). Intentional reconﬁguration and involuntary persistence in
task set switching. In S. Monsell & J. Driver (Eds), Control of Cognitive Processes (Vol.
28, pp. 331-355). Cambridge, MA: The MIT Press.

 

Hager-Ross, C., & Schieber, M. H. (2000). Quantifying the independence of
human ﬁnger movements: Comparisons of digits, hands, and movement frequencies.
Journal of Neuroscience, 20, 8542-8550.

 

109

Houghton, G., & Tipper, S. P. (1994). A model of inhibitory mechanisms in
selective attention. In D. Dagenbach & T. H. Carr (Eds), Inhibitory processes in
attention, memory, and language (pp. 53-112). San Diego, CA: Academic Press.

 

 

Huang, J ., Carr, T. H., & Cao, Y. (2001). Comparing cortical activations for silent
and overt speech using event-related fMRI. Human Brain Mapping, 15, 39-53.

 

Hubner, R., Futterer, T., & Steinhauser, M. (2001). On attentional control as a
source of residual shift costs: Evidence ﬁ'om two-component task shifts. Journal of
Experimental Psychology: Learning, Memory, and Cognition, 27, 640-653.

 

Jersild, A. T. (1927). Mental set and shift. Archives of Psychology(Whole No.

 

89).

Jolicoeur, P., Dell'Acqua, R., & Crebolder, J. (2000). Multitasking performance
deﬁcits: Forging links between the attentional blink and the psychological refractory

period. In S. Monsell & J. Driver (Eds), Control of cognitive processes: Attention and
performance XVIII (pp. 309-330). Cambridge, MA: MIT Press.

 

Keele, S. W. (1973). Attention and human performance. Paciﬁc Palisades, CA:
Goodyear Publishing Company, Inc.

 

Kintsch, W. (1980). Semantic memory: A tutorial. In R. S. Nickerson (Ed.),
Attention and Performance VIII (pp. 595-620). Hillsdale, NJ: Lawrence Erlbaum
Associates.

Klein, R., & Shore, D. (2000). Relations among modes of visual orienting. In S.
Monsell & J. Driver (Eds), Control of Coggitive Processes (Vol. 28, pp. 195-221).
Cambridge, MA: The MIT Press.

 

Kleinsorge, T., & Heuer, H. (1999). Hierarchical switching in a multi-dimensional
task space. Psychological Research, 62, 300-312.

Kleinsorge, T., Heuer, H., & Schmidtke, V. (2001a). Hierarchical switching in a
multi-dimensional task space is not induced by speciﬁc task cues. Zeitschrift fur
Echologie, 209, 105-117.

 

 

Kleinsorge, T., Heuer, H., & Schmidtke, V. (2001b). Task-set reconﬁguration
with binary and three-valued task dimensions. Psychological Research, 65, 192-201.

 

Kramer, A. F., & Jacobson, A. (1991). Perceptual organization and focused
attention: the role of objects and proximity in visual processing. Perception &
Psychoﬂrysics, 50, 267-284.

 

 

Kwak, H., Dagenbach, D., & Egeth, H. (1991). Further evidence for a time-
independent shift of the focus of attention. Perception & Psychophysics, 49, 473-480.

 

110

Loftus, G. R., & Masson, M. E. J. (1994). Using conﬁdence intervals in within-
subject designs. Psychonomic Bulletin & Review, 4, 476-490.

 

Logan, G. D. (1985). Executive control of thought and action. Acta Psychologica,
Q, 193-210.

 

Mayr, U. (2002). Inhibition of action rules. Psychonomic Bulletin & Review, 9,

 

93-33.

Mayr, U., & Keele, S. W. (2000). Changing internal constraints on action: The
role of backward inhibition. Journal of Experimental Psychology: Generalg 29, 4-26.

 

Mayr, U., & Kliegl, R. (2000). Task-set switching and long-terrn memory
retrieval. Journal of Experimental Psychology: Learning, Memory, and Cognition, 26,
1124-1140.

 

McLeod, P. (1977). A dual task response modality effect: Support for
multiprocessor models of attention. Quarterly Journal of Experimental Psychology, 29,
651-667.

 

Meiran, N. (1996). Reconﬁguration of processing mode prior to task performance.
Journal of Experimental Psychology: Learning, Memory, and Cognition, 22, 1423-1442.

Meiran, N. (2000a). Reconﬁguration of stimulus task sets and response task sets
during task switching. In S. Monsell & J. Driver (Eds), Control of Cognitive Processes
(Vol. 28, pp. 377-399). Cambridge, MA: The MIT Press.

 

Meiran, N. (2000b). Modeling cognitive control in task-switching. Psychological
Research, 63, 234-249.

 

 

Meiran, N., Chorev, Z., & Sapir, A. (2000). Component processes in task
switching. Cognitive Psychology, 41, 211-253.

 

Meuter, R. F. I., & Allport, A. (1999). Bilingual language switching in naming:
Asymmetrical costs of language selection. Journal of Memory and Language, 40, 25-40.

 

Monsell, S., Yeung, N., & Azuma, R. (2000). Reconﬁguration of task-set: Is it
easier to switch to the weaker task? Psycholgjcal Research, 63, 250-264.

 

Neely, J. H. (1977). Semantic priming and retrieval from lexical memory: Roles
of inhibitionless spreading activation and limited-capacity attention. Journal of
Experimental Psychology: General, 106, 226-254.

 

111

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),
Consciousness and self-regulation: Advances in research and theory (Vol. 4, pp. 1-18).
New York: Plenum Press.

 

Nosofsky, R. M., & J ohansen, M. K. (2000). Exemplar-based accounts of
"multiple-system" phenomena in perceptual categorization. Psychonomic Bulletin &
Review 7, 375-402.

 

 

Pashler, H. (1990). Do response modality effects support multiprocessor models
of divided attention? Journal of Experimental Psychology: Human Perception and
Performance, 16, 826-842.

 

 

Pashler, H. E. (1998). The Psychology of Attention. Cambridge, MA: MIT Press.

 

Pashler, H. (2000). Task switching and multitask performance. In S. Monsell & J.
Driver (Eds), Control of cognitive processes: Attention and performance XVIII (pp. 278-
307). Cambridge, MA: MIT Press.

 

Posner, M. I. (1980). Orienting of attention. Quarterly Journal of Experimental
Psychology, 32, 3-25.

 

 

Posner, M. I., & Snyder, C. R. R. (1975). Attention and cognitive control. In R. L.
Solo (Ed.), Information processirpgand cognition (pp. 55-85). Hillsdale, N. J .: Erlbaum.

Proctor, R. W., & Dutta, A. (1995). Skill acquisition and human performance.
Thousand Oaks, CA: Sage.

 

Psychological Software Tools, Inc. (2000). E-prime (Version 1.0 Beta 5).
Pittsburgh, PA.

Rao, S. M., Bandettini, P. A., Binder, J. R., Bobholz, J. A., Hammeke, T. A.,
Stein, E. A., & Hyde, J. S. (1996). Relationship between ﬁnger movement rate and
funcitonal magnetic resonance signal change in human primary motor cortex. Journal of
Cerebral Blood Flow and Metabolism, 16, 1250-1254.

 

Raymond, J. E., Shapiro, K. L., & Amell, K. M. (1992). Temporary suppression
of visual processing in an RSVP task: An attentional blink? Journal of Experimental
Psychology: Human Perception and Performance, 18, 849-860.

 

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., Mayer, 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.

 

 

112

Ruthruff, E., & Pashler, H. E. (2001). Perceptual and central interference in dual-
task performance. In K. Shapiro (Ed.), The limits of attention: Temporal constraints in
human information processing (pp. 100-123). Oxford: Oxford University Press.

 

Sanes, J. N., Donoghue, J. P., Thangaraj, V., Edelman, R. R., & Warach, S.
(1995). Shared neural substrates controlling hand movements in human motor cortex.
Science, 268, 1775-1777.

 

Schieber, M. H. (1999). Voluntary descending control. In M. J. Zigmond, F. E.
Bloom, S. C. Landis, J. L. Roberts, & L. R. Squire (Eds), Fundamental Neuroscience
(pp. 931-949). San Diego: Academic Press.

 

Schneider, W., & Shiffrin, R. M. (1977). Controlled and automatic human
information processing: I. Detection, search, and attention. Psychological Review, 84, l-
66.

 

Schvaneveldt, R. W., & Meyer, D. E. (1973). Retrieval and comparison processes
in semantic memory .

Smith, E. E., Shoben, E. J ., & Rips, L. J. (1974). Structure and process in
semantic memory: A featural model for semantic decisions. Psychological Review, 81,
214-241.

 

Sohn, M.-H., & Anderson, J. R. (2001). Task preparation and task repetition:
Two-component model of task switching. Journal of Experimental Psychology: General,
I3_O, 764-778.

 

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.

 

Spector, A., & Biederman, I. (1976). Mental set and shift revisited. American
Journal of Psychology, 89, 669-679.

 

Stroop, J. R. (1935). Studies of interference in serial verbal reactions. Journal of
Experimental Psychology, 18, 643-662.

 

Sudevan, P., & Taylor, D. A. (1987). The cuing and priming of cognitive
operations. Journal of Experimental Psychology: Human Perception and Performance,
I3, 89-103.

 

Tsal, Y. (1983). Movements of attention across the visual ﬁeld. Journal of
Experimental Psychology: Human Perception and Performance, 9, 523-530.

 

Tversky, A. (1977). Features of similarity. Psychological Review, 84, 327-352.

 

113

Visser, T. A. W., Bischof, W. F., & Di Lollo, V. (1999). Attentional switching in
spatial and nonspatial domains: Evidence from the attentional blink. Psychological
Bulletin, 125, 458-469.

 

 

114

 

 

 

IIIIIIIIIIIIIIIIIIIIIIIIIIIIII

' 11111111111unwilliyyit

   

302