Chapter

5 Attention and
performance

INTRODUCTION
Attention is invaluable in everyday life. We use attention to avoid being
hit by cars when crossing the road, to search for missing objects and to
perform two tasks together. The word “attention” has various meanings
but typically refers to selectivity of processing as emphasised by William
James (1890, pp. 403–404):

Attention is . . . the taking into possession of the mind, in clear and

vivid form, of one out of what seem several simultaneously possible
objects or trains of thought. Focalisation, concentration, of conscious-
ness are of its essence.

William James distinguished between “active” and “passive” modes of

attention. Attention is active when controlled in a top-down way by the
individual’s goals or expectations. In contrast, attention is passive when
controlled in a bottom-up way by external stimuli (e.g., a loud noise). This
distinction remains theoretically important (e.g., Corbetta & Shulman,
2002; see discussion, pp. 192–196).
Another important distinction is between focused and divided atten-
KEY TERMS tion. Focused attention (or selective attention) is studied by present-
Focused attention ing individuals with two or more stimulus inputs at the same time and
A situation in which instructing them to respond to only one. Research on focused or selective
individuals try to attend attention tells us how effectively we can select certain inputs and avoid
to only one source of
information while ignoring
being distracted by non-task inputs. It also allows us to study the selection
other stimuli; also known process and the fate of unattended stimuli.
as selective attention. Divided attention is also studied by presenting at least two stimulus
Divided attention
inputs at the same time. However, individuals are instructed they must
A situation in which two attend (and respond) to all stimulus inputs. Divided attention is also
tasks are performed at the known as multi-tasking (see Glossary). Studies of divided attention provide
same time; also known as useful information about our processing limitations and the capacity of
multi-tasking. our attentional mechanisms.
 Attention and performance 179

There is a final important distinction (the last one, I promise you!) KEY TERMS
between external and internal attention. External attention is “the selec-
Cocktail party problem
tion and modulation of sensory information” (Chun et al., 2011). In con- The difficulties involved
trast, internal attention is “the selection, modulation, and maintenance of in attending to one voice
internally generated information, such as task rules, responses, long-term when two or more people
memory, or working memory” (Chun et al., 2011, p. 73). The connec- are speaking at the same
time.
tion to Baddeley’s working memory model is especially important (e.g.,
Baddeley (2012); see Chapter 6). The central executive component of Dichotic listening task
working memory is involved in attentional control and is crucially involved A different auditory
message is presented to
in internal and external attention.
each ear and attention
Much attentional research has two limitations. First, the emphasis is has to be directed to one
on attention to externally presented task stimuli rather than internally gen- message.
erated stimuli (e.g., worries; self-reflection). One reason is that it is easier to
Shadowing
assess and to control external attention. Second, what participants attend Repeating one auditory
to is determined by the experimenter’s instructions. In contrast, what we message word for word
attend to in the real world is mostly determined by our current goals and as it is presented while a
emotional states. second auditory message
is also presented; it is
Two important topics related to attention are discussed elsewhere.
used on the dichotic
Change blindness (see Glossary), which shows the close links between atten- listening task.
tion and perception, is considered in Chapter 4. Consciousness (including
its relationship to attention) is discussed in Chapter 16.

FOCUSED AUDITORY ATTENTION

Many years ago, British scientist Colin Cherry (1953) became fascinated
by the cocktail party problem – how can we follow just one conversation
when several people are talking at once? As we will see, there is no simple
answer.
McDermott (2009) identified two problems listeners face when attend-
ing to one voice among many. First, there is sound segregation: the listener
must decide which sounds belong together. This is complex: machine-based
speech recognition programs often perform poorly when attempting to
achieve sound segregation with several sound sources present together
(Shen et al., 2008). Second, after segregation has been achieved, the listener
must direct attention to the sound source of interest and ignore the others.
McDermott (2009) pointed out that auditory segmentation is often
harder than visual segmentation (deciding which visual features belong
to which objects; see Chapter 3). There is considerable overlap of signals
from different sound sources in the cochlea whereas visual objects typically
occupy different retinal regions.
There is another important issue – when listeners attend to one audi-
tory input, how much processing is there of the unattended input(s)? As we
will see, various answers have been proposed.
Cherry (1953) addressed the issues discussed so far (see Eysenck, 2015,
for an evaluation of his research). He studied the cocktail party problem
using a dichotic listening task in which a different auditory message was
presented to each ear and the listener attended to only one. Listeners
engaged in shadowing (repeating the attended message aloud as it was pre-
sented) to ensure their attention was directed to that message. However,
the shadowing task has two potential disadvantages: (1) listeners do not
180 Visual perception and attention

normally engage in shadowing and so the task is artificial; and (2) it

increases listeners’ processing demands.
Listeners solved the cocktail party problem by using differences between
the auditory inputs in physical features (e.g., sex of speaker; voice intensity;
speaker location). When these physical differences were eliminated by pre-
senting two messages in the same voice to both ears at once, listeners found
it very hard to separate out the messages based on differences in meaning.
Cherry (1953) found very little information seemed to be extracted
from the unattended message. Listeners seldom noticed when it was spoken
backwards or in a foreign language. However, physical changes (e.g., a
pure tone) were nearly always detected. The conclusion that unattended
information receives minimal processing was supported by Moray (1959),
who found listeners remembered very few words presented 35 times each.

Where is the bottleneck? Early vs late selection

Many psychologists have argued we have a processing bottleneck (discussed
below). A bottleneck in the road (e.g., where it is especially narrow) can
cause traffic congestion, and a bottleneck in the processing system seriously
limits our ability to process two (or more) simultaneous inputs. However,
it would sometimes solve the cocktail problem by permitting listeners to
process only the desired voice.
Where is the bottleneck? Broadbent (1958) argued a filter (bottleneck)
early in processing allows information from one input or message through
it based on the message’s physical characteristics. The other input remains
briefly in a sensory buffer and is rejected unless attended to rapidly (see
Figure 5.1). Thus, Broadbent argued there is early selection.
Treisman (1964) argued the bottleneck’s location is more flexible than
Broadbent suggested (see Figure 5.1). She claimed listeners start with pro-
cessing based on physical cues, syllable pattern and specific words and then
process grammatical structure and meaning. Later processes are omitted or
Interactive exercise: attenuated if there is insufficient processing capacity to permit full stimulus
Treisman analysis.
Treisman (1964) also argued top-down processes (e.g., expectations)
are important. Listeners performing the shadowing task sometimes say a
word from the unattended input. Such breakthroughs mostly occur when
the word on the unattended channel is highly probable in the context of
the attended message.
Deutsch and Deutsch (1963) argued all stimuli are fully analysed, with
the most important or relevant stimulus determining the response. Thus,
they placed the bottleneck much later in processing than did Broadbent
(see Figure 5.1).

Findings: unattended input

Broadbent’s approach predicts little or no processing of unattended audi-
tory messages. In contrast, Treisman’s approach suggests flexibility in
the processing of unattended messages, whereas Deutsch and Deutsch’s
approach implies reasonably thorough processing of such messages.
Relevant findings are discussed below.
 Attention and performance 181

Treisman and Riley (1969) asked listeners to shadow one of two audi- Figure 5.1
tory messages. They stopped shadowing and tapped when they detected A comparison of
a target in either message. Many more target words were detected on the Broadbent’s theory (top),
Treisman’s theory (middle),
shadowed message.
and Deutsch and Deutsch’s
Aydelott et al. (2015) asked listeners to perform a task on attended theory (bottom).
target words. When unattended words related in meaning were presented
shortly before the target words themselves, performance on the target
words was enhanced when unattended words were presented as loudly as
attended ones. Thus, the meaning of unattended words was processed.
There is often more processing of unattended words that have a special
significance for the listener. For example, Li et al. (2011) obtained evidence
that unattended weight-related words (e.g., fat; chunky) were processed
more thoroughly by women dissatisfied with their weight. Conway et al.
(2001) found listeners often detected their own name on the unattended
message. This was especially the case if they had low working memory
capacity (see Glossary) indicative of poor attentional control.
Coch et al. (2005) asked listeners to attend to one of two auditory
inputs and to detect targets presented on either input. Event-related poten-
tials (ERPs; see Glossary) provided a measure of processing activity. ERPs
100 ms after target presentation were greater when the target was presented
on the attended rather than the unattended message. This suggests there
was more processing of the attended than unattended targets.
Greater brain activation for attended than unattended auditory stimuli
may reflect enhanced processing for attended stimuli and/or suppressed
processing for unattended stimuli. Horton et al. (2013) addressed this
 182 Visual perception and attention

issue. Listeners heard separate speech messages presented to each ear with
instructions to attend to the left or right ear. There was greater brain acti-
vation associated with the attended message (especially around 90 ms after
stimulus presentation). This difference depended on enhancement of the
attended message combined with suppression of the unattended message.
Classic theories of selective auditory attention (those of Broadbent,
Treisman, and Deutsch and Deutsch) de-emphasised the importance of
suppression or inhibition of the unattended message shown by Horton
et al. (2013). For example, Schwartz and David (2018) reported suppres-
sion of neuronal responses in the primary auditory cortex to distractor
sounds. More generally, all the classic theories de-emphasise the flexibility
of selective auditory attention and the role of top-down processes in selec-
tion (see below).

Findings: cocktail party problem

Humans are generally very good at separating out and understanding one
voice from several speaking at the same time (i.e., solving the cocktail party
problem). The extent of this achievement is indicated by the finding that
automatic speech recognition systems are considerably inferior to human
speech recognition (Spille & Meyer, 2014).
Mesgarani and Chang (2012) studied listeners with implanted multi-­
electrode arrays permitting the direct recording of activity within the audi-
tory cortex. They heard two different messages (one in a male voice; one
in a female voice) presented to the same ear with instructions to attend to
only one. The responses within the auditory cortex revealed “The salient
spectral [based on sound frequencies] and temporal features of the attended
speaker, as if subjects were listening to that speaker alone” (Mesgarani &
Chang, 2012, p. 233).
Listeners found it easy to distinguish between the two messages in the
study by Mesgarani and Chang (2012) because they differed in physical
characteristics (i.e., male vs female voice). Olguin et al. (2018) presented
native English speakers with two messages in different female voices. The
attended message was always in English whereas the unattended message
was in English or an unknown language. Comprehension of the attended
message was comparable in both conditions. However, there was stronger
neural encoding of both messages in the former condition. As Olguin et al.
concluded, “The results offer strong support to flexible accounts of selec-
tive [auditory] attention” (p. 1618).
In everyday life, we are often confronted by several different speech
streams. Accordingly, Puvvada and Simon (2017) presented three speech
streams and assessed brain activity as listeners attended to only one. Early
in processing, “the auditory cortex maintains an acoustic representation of
the auditory scene with no significant preference to attended over ignored
sources” (p. 9195). Later in processing, “Higher-order auditory cortical
areas represent an attended speech stream separately from, and with signifi-
cantly higher fidelity [accuracy] than, unattended speech streams” (p. 9189).
This latter finding results from top-down processes (e.g., attention).
How do we solve the cocktail party problem? The importance of
top-down processes is suggested by the existence of extensive descending

9781138482210_COGNITIVE_PSYCHOLOGY_PART_1.indd 182 28/02/20 6:44 PM

 Attention and performance 183

pathways from the auditory cortex to brain areas involved in early auditory
processing (Robinson & McAlpine, 2009). Various top-down factors based
on listeners’ knowledge and/or expectations are involved. For example, lis-
teners are more accurate at identifying what one speaker is saying in the
context of several other voices if they have previously heard that speaker’s
voice in isolation (McDermott, 2009).
Woods and McDermott (2018) investigated top-down processes in
selective auditory attention in more detail. They argued, “Sounds pro-
duced by a given source often exhibit consistencies in structure that might
be useful in separating sources” (p. E3313). They used the term “schemas”
to refer to such structural consistencies.
Listeners showed clear evidence of schema learning leading to rapid
improvements in their listening performance. An important aspect of such
learning is temporal coherence – a given source’s sound features are typ-
ically all present when it is active and absent when it is silent. Shamma
et al. (2011) discussed research showing that if listeners can identify one
distinctive feature of the target voice, they can then distinguish its other
sound features via temporal coherence.
Evans et al. (2016) compared patterns of brain activity when attended
speech was presented on its own or together with competing unattended
speech. Brain areas associated with attentional and control processes (e.g.,
frontal and parietal regions) were more activated in the latter condition.
Thus, top-down processes relating to attention and control are important
in selective auditory processing.
Finally, Golumbic et al. (2013) suggested individuals at actual cocktail
parties can potentially use visual information to assist them in understand-
ing what a given speaker is saying. Listeners heard two simultaneous mes-
sages (one in a male voice and the other in a female voice). Processing of
the attended message was enhanced when they saw a video of the speaker
talking.
In sum, listeners generally achieve the complex task of selecting one
speech message from among several such messages. There has been pro-
gress in identifying the top-down processes involved. For example, if lis-
teners can identify at least one consistently distinctive feature of the target
voice, this makes it easier for them to attend only to that voice. Top-down
processes often produce a “winner-takes-all” situation where the processing
of one auditory input (the winner) suppresses the brain activity ­associated
with all other inputs (Kurt et al., 2008).

FOCUSED VISUAL ATTENTION

There has been much more research on visual attention than auditory atten-
tion. The main reason is that vision is our most important sense modality
with more of the cortex devoted to it than any other sense. Here we con-
sider four key issues. First, what is focused visual attention like? Second,
what is selected in focused visual attention? Third, what happens to unat-
tended visual stimuli? Fourth, what are the major systems involved in visual
attention? In the next section (see pp. 196–200), we discuss what the study
of visual disorders has taught us about visual attention.
 184 Visual perception and attention

KEY TERM Spotlight, zoom lens or multiple spotlights?

Split attention
Allocation of attention Look around you and attend to any interesting objects. Was your visual
to two (or more) non- attention like a spotlight? A spotlight illuminates a fairly small area, little
adjacent regions of visual can be seen outside its beam and it can be redirected to focus on any given
space. object. Posner (1980) argued the same is true of visual attention.
Other psychologists (e.g., Eriksen & St. James, 1986) claim visual
attention is more flexible than suggested by the spotlight analogy and
argue visual attention resembles a zoom lens. We can increase or decrease
the area of focal attention just as a zoom lens can be adjusted to alter the
visual area it covers. This makes sense. For example, car drivers often need
to narrow their attention after spotting a potential hazard.
A third theoretical approach is even more flexible. According to the
multiple spotlights theory (Awh & Pashler, 2000), we sometimes exhibit
split attention (attention directed to two or more non-adjacent regions
in space). The notion of split attention is controversial. Jans et al. (2010)
argued attention is often strongly linked to motor action and so attending
to two separate objects might disrupt effective action. However, there is no
strong evidence for such disruption.

Findings
Support for the zoom-lens model was reported by Müller et al. (2003). On
each trial, observers saw four squares in a semi-circle and were cued to
attend to one, two or all four. Four objects were then presented (one in
each square) and observers decided whether a target (e.g., a white circle)
was among them. Brain activation in early visual areas was most wide-
spread when the attended region was large (i.e., attend to all four squares)
and was most limited when it was small (i.e., attend to one square). As pre-
dicted by the zoom-lens theory, performance (reaction times and errors)
was best with the smallest attended region and worst with the largest one.
Chen and Cave (2016, p. 1822) argued the optimal attentional zoom
setting “includes all possible target locations and excludes possible distrac-
tor locations”. Most findings indicated people’s attentional zoom setting
is close to optimal. However, Collegio et al. (2019) obtained contrary
findings. Drawings of large objects (e.g., jukebox) and small objects (e.g.,
watch) were presented so their retinal size was the same. The observer’s
area of focal attention was greater with large objects because they made
top-down inferences concerning their real-world sizes. As a result, the area
of focal attention was larger than optimal for large objects.
Goodhew et al. (2016) pointed out that nearly all research has focused
only on spatial perception (e.g., identification of a specific object). They
focused on temporal perception (was a disc presented continuously or
were there two presentations separated by a brief interval?). Spotlight size
had no effect on temporal acuity, which is inconsistent with the theory.
How can we explain these findings? Spatial resolution is poor in peripheral
vision but temporal resolution is good. As a consequence, a small atten-
tional spotlight is more beneficial for spatial than temporal acuity.
We turn now to split attention. Suppose you had to identify two digits
that would probably be presented to two cued locations a little way apart

9781138482210_COGNITIVE_PSYCHOLOGY_PART_1.indd 184 28/02/20 6:44 PM

 Attention and performance 185

Figure 5.2
(a) Shaded areas indicate
the cued locations; the
near and far locations are
not cued. (b) Probability of
target detection at valid
(left or right) and invalid
(near or far) locations.
Based on information in Awh
and Pashler (2000).

(see Figure 5.2a). Suppose also that on some trials a digit was presented
between the two cued locations. According to zoom-lens theory, the area
of maximal attention should include the two cued locations and the space
in between. As a result, the detection of digits presented in the middle
should have been very good. In fact, Awh and Pashler (2000) found it was
poor (see Figure 5.2b). Thus, attention can resemble multiple spotlights, as
predicted by the split-attention approach.
Morawetz et al. (2007) presented letters and digits at five locations
simultaneously (one in each quadrant of the visual field and one in the
centre). In one condition, observers attended to the visual stimuli at the
upper left and bottom right locations and ignored the other stimuli. There
were two peaks of brain activation corresponding to the attended areas but
less activation corresponding to the region in between. Overall, the pattern
of activation strongly suggested split attention.
Niebergall et al. (2011) recorded the neuronal responses of monkeys
attending to two moving stimuli while ignoring a distractor. In the key
condition, there was a distractor between (and close to) the two attended
stimuli. In this condition, neuronal responses to the distractor decreased
compared to other conditions. Thus, split attention involves a mecha-
nism reducing attention to (and processing of) distractors located between
attended stimuli.
186 Visual perception and attention

KEY TERM In most research demonstrating split attention, the two non-adjacent
stimuli being attended simultaneously were each presented to a different
Hemifield
One half of the visual hemifield (one half of the visual field). Note that the right hemisphere
field. Information from receives visual signals from the left hemifield and the left hemisphere
the left hemifield of each receives signals from the right hemifield. Walter et al. (2016) found per-
eye proceeds to the formance was better when non-adjacent stimuli were presented to different
right hemisphere and
hemifields rather than the same hemifield. Of most importance, the assess-
information from the right
hemifield proceeds to the ment of brain activity indicated effective filtering or inhibition of stimuli
left hemisphere. presented between the two attended stimuli only when presented to differ-
ent hemifields.
In sum, we can use visual attention very flexibly. Visual selective atten-
tion can resemble a spotlight, a zoom lens or multiple spotlights, depending
on the current situation and the observer’s goals. However, split attention
may require that two stimuli are presented to different hemifields rather
than the same one. A limitation with all these theories is that metaphors
(e.g., attention is a zoom lens) are used to describe experimental findings but
these metaphors fail to specify the underlying mechanisms (Di Lollo, 2018).

What is selected?
Why might selective attention resemble a spotlight or zoom lens? Perhaps
we selectively attend to an area or region of space: space-based attention.
Alternatively, we may attend to a given object or objects: object-based
attention. Object-based attention is prevalent in everyday life because visual
attention is mainly concerned with objects of interest to us (see Chapters 2
and 3). As expected, observers’ eye movements as they view natural scenes
are directed almost exclusively to objects (Henderson & Hollingworth,
1999). However, even though we typically focus on objects of potential
importance, our attentional system is so flexible we can attend to an area of
space or a given object.
There is also feature-based attention. For example, suppose you are
looking for a friend in a crowd. Since she nearly always wears red clothes,
you might attend to the feature of colour rather than specific objects or
locations. Leonard et al. (2015) asked observers to identify a red letter
within a series of rapidly presented letters. Performance was impaired when
a # symbol also coloured red was presented very shortly before the target.
Thus, there was evidence for feature-based attention (e.g., colour; motion).

Findings
Visual attention is often object-based. For example, O’Craven et al. (1999)
presented observers with two stimuli (a face and a house), transparently
overlapping at the same location, with instructions to attend to one of
them. Brain areas associated with face processing were more activated
when the face was attended to than when the house was. Similarly, brain
areas associated with house processing were activated when the house was
the focus of attention.
Egly et al. (1994) devised a much-used method for comparing object-
based and space-based attention (see Figure 5.3). The task was to select
a target stimulus as rapidly as possible. A cue presented before the target
 Attention and performance 187

was valid (same location as the target) or

invalid (different location from the target).
Of key importance, invalid cues were in the
same object as the target (within-object cues)
or in a different object (between-object cues).
The key finding was that target detection was
faster on invalid trials when the cue was in the
same object rather than a different one. Thus,
attention was at least partly object-based.
Does object-based attention in the Egly
et al. (1994) task occur fairly “automatically”
or does it involve strategic processes? Object- Figure 5.3
based attention should always be found if it is Stimuli adapted from Egly et al. (1994). Participants saw two
rectangles and a cue indicated the most likely location of a
automatic. Drummond and Shomstein (2010)
subsequent target. The target appeared at the cued location
found no evidence for object-based attention (V), at the uncued end of the cued rectangle (IS) or at the
when the cue indicated with 100% certainty uncued, equidistant end of the uncued rectangle (ID).
where the target would appear. Thus, any From Chen (2012). © Psychonomic Society, Inc. Reprinted with
preference for object-based attention can be permission from Springer.
overridden when appropriate.
Hollingworth et al. (2012) found evidence object-based and space-based
attention can occur at the same time using a task resembling that of Egly
et al. (1994). There were three types of within-object cues varying in the
distance between the cue and subsequent target (see Figure 5.4). There was
evidence for object-based attention: when the target was far from the cue,
performance was worse when the cue was in a different object rather than
the same one. There was also evidence for space-based attention: when the
target was in the same object as the cue, performance declined the greater
the distance between target and cue. Thus, object-based and space-based
attention are not mutually exclusive.
Similar findings were reported by Kimchi et al. (2016). Observers
responded faster to a target presented within rather than outside an object.
This indicates object-based attention. There was also evidence for space-
based attention: when targets were presented outside the object, observ-
ers responded faster when they were close to it. Kimchi et al. concluded
that “object-related and space-related attentional processing can operate
­simultaneously” (p. 48).
Pilz et al. (2012) compared object-based and space-based attention
using various tasks. Overall, there was much more evidence of space-based
than object-based attention, with only a small fraction of participants
showing clear-cut evidence of object-based attention.
Donovan et al. (2017) noted that most studies indicating visual atten-
tion is object-based have used spatial cues, which may bias the allocation
of attention. Donovan et al. avoided the use of spatial cues and found
“Object-based representations do not guide attentional selection in the
absence of spatial cues” (p. 762). This finding suggests previous research KEY TERM
has exaggerated the extent of object-based visual attention. Inhibition of return
When we search the visual environment, it would be inefficient if we A reduced probability of
repeatedly attended to any given location. In fact, we exhibit inhibition of visual attention returning
return (a reduced probability of returning to a region recently the focus of to a recently attended
attention). Of theoretical importance is whether inhibition of return applies location or object.
188 Visual perception and attention

Figure 5.4
(a) Possible target locations
(same object far, same
object near, valid, different
object far) for a given cue.
(b) Performance accuracy at
the various target locations.
From Hollingworth et al. (2012).
© 2011 American Psychological
Association.

more to locations or objects. The evidence is mixed (see Chen, 2012). List
and Robertson (2007) used Egly et al.’s (1994) task shown in Figure 5.4
and found location- or space-based inhibition of return was much stronger
than object-based inhibition of return.
Theeuwes et al. (2014) found location- and object-based inhibition of
return were both present at the same time. According to Theeuwes et al.
(p. 2254), “If you direct your attention to a location in space, you will
automatically direct attention to any object . . . present at that location,
and vice versa.”
There is considerable evidence of feature-based attention (see Bartsch
et al., 2018, for a review). In their own research, Bartsch et al. addressed
 Attention and performance 189

the issue of whether feature-based attention to colour-defined targets is

confined to the spatially attended region or whether it occurs across the
entire visual field. They discovered the latter was the case.
Finally, Chen and Zelinsky (2019) argued it is important to study the
allocation of attention under more naturalistic conditions than those typ-
ically used in research. In their study, observers engaged in free (uncon-
strained) viewing of natural scenes. Eye-fixation data suggested that
attention initially selects regions of space. These regions may provide “the
perceptual fragments from which objects are built” (p. 148).

Evaluation
Research on whether visual attention is object- or location-based have
produced variable findings and so few definitive conclusions are pos-
sible. However, the relative importance of object-based and space- or
­location-based attention is flexible. For example, individual differences are
important (Pilz et al., 2012). Note that visual attention can be both object-
based and space-based at the same time.
What are the limitations of research in this area? First, most research
apparently demonstrating that object-based attention is more important
than space- or location-based attention has involved the use of spatial
cues. Recent evidence (Donovan et al., 2017) suggests such cues may bias
visual attention and that visual attention is not initially object-based in
their absence.
Second, space-, object- and feature-based forms of attention often inter-
act with each other to enhance object processing (Kravitz & Behrmann,
2011). However, we have as yet limited theoretical understanding of the
mechanisms involved in such interactions.
Third, there is a need for more research assessing patterns of attention
under naturalistic conditions. In recent research where observers view artifi-
cial stimuli while performing a specific task, it is unclear whether attentional
processes resemble those when they engage in free viewing of natural scenes.

What happens to unattended or distracting stimuli?

Unsurprisingly, unattended visual stimuli receive less processing than
attended ones. Martinez et al. (1999) compared event-related potentials
(ERPs) to attended and unattended visual stimuli. The ERPs to unattended
visual stimuli were comparable to those to attended ones 50–55 ms after
stimulus onset. After that, however, the ERPs to attended stimuli were
greater than those to unattended stimuli. Thus, selective attention influ-
ences all but the very early stages of processing.
As we have all discovered to our cost, it is often hard (or impossible)
to ignore task-irrelevant stimuli. Below we consider factors determining
whether task performance is adversely affected by distracting stimuli.

Load theory
Lavie’s (2005, 2010) load theory has been an influential approach to
understanding distraction effects. It distinguishes between perceptual

9781138482210_COGNITIVE_PSYCHOLOGY_PART_1.indd 189 28/02/20 6:44 PM

190 Visual perception and attention

and cognitive load. Perceptual load refers to the perceptual demands of a

current task. Cognitive load refers to the burden placed on the cognitive
system by a current task (e.g., demands on working memory).
Tasks involving high perceptual load require nearly all our perceptual
capacity whereas low-load tasks do not. With low-load tasks there are
spare attentional resources, and so task-irrelevant stimuli are more likely
to be processed. In contrast, tasks involving high cognitive load reduce our
ability to use cognitive control to discriminate between target and distrac-
tor stimuli. Thus, high perceptual load is associated with low distractibility,
whereas high cognitive load is associated with high distractibility.

Findings
There is much support for the hypothesis that high perceptual load reduces
distraction effects. Forster and Lavie (2008) presented six letters in a circle
and participants decided which target letter (X or N) was present. The five
non-target letters resembled the target letter more closely in the high-load
condition. On some trials a picture of a cartoon character (e.g., Spongebob
Squarepants) was presented as a distractor outside the circle. Distractors
interfered with task performance only under low-load conditions.
According to the theory, brain activation associated with distractors
should be less when individuals are performing a task involving high per-
ceptual load. This finding has been obtained with visual tasks and distrac-
tors (e.g., Schwartz et al., 2005) and also with auditory tasks and distractors
(e.g., Sabri et al., 2013).
Why is low perceptual load associated with high distractibility? Biggs
and Gibson (2018) argued this happens because observers generally adopt
a broad attentional focus when perceptual load is low. They tested this
hypothesis using three low-load conditions in which participants decided
whether a target X or N was presented and a distractor letter was some-
times presented (see Figure 5.5). They argued that observers would adopt
the smallest attentional focus in the circle condition and the largest atten-
tional focus in the solo condition. As predicted, distractor interference was
greatest in the solo condition and least in the circle condition. Thus, dis-
traction effects depend strongly on size of attentional focus as well as per-
ceptual load.
The hypothesis that distraction effects should be greater when cognitive
or working memory load is high rather than low was tested by Burnham
et al. (2014). As predicted, distraction effects on a visual search task were

Standard condition Solo condition Circle condition

Figure 5.5
Sample displays for three
low perceptual load X X X
* * * *
conditions in which the T T T
task required deciding * * * *
whether a target X or N * *
was presented. See text for
further details.
From Biggs and Gibson (2018).
 Attention and performance 191

greater when participants performed another task placing high demands

on the cognitive system.
Sörqvist et al. (2016) argued high cognitive load can reduce rather than
increase distraction. They pointed out that cognitive load is typically asso-
ciated with high levels of concentration and our everyday experience indi-
cates high concentration generally reduces distractibility. As predicted, they
found neural activation associated with auditory distractors was reduced
when cognitive load on a visual task was high rather than low.
The effects of cognitive load on distraction are very variable. How can
we explain this variability? Sörqvist et al. (2016) argued that an impor-
tant factor is how easily distracting stimuli can be distinguished from task
stimuli. When it is easy (e.g., task and distracting stimuli are in differ-
ent modalities as in the Sörqvist et al., 2016, study), high cognitive load
reduces distraction. In contrast, when it is hard to distinguish between task
and distracting stimuli (e.g., they are similar and/or in the same modality),
then high cognitive load increases distraction.
Load theory assumes the effects of perceptual and cognitive load are
independent. However, Linnell and Caparos (2011) found perceptual and
cognitive processes interacted: perceptual load only influenced attention as
predicted when cognitive load was low. Thus, the effects of perceptual load
are not “automatic” as assumed theoretically but instead depend on cogni-
tive resources being available.

Evaluation
The distinction between perceptual and cognitive load has proved useful
in predicting when distraction effects will be small or large. More specifi-
cally, the prediction that high perceptual load is associated with reduced
distraction effects has received much empirical support. In applied research,
load theory successfully predicts several aspects of drivers’ attention and
behaviour (Murphy & Greene, 2017). For example, drivers exposed to high
perceptual load responded more slowly to hazards and drove less safely.
What are the theory’s limitations? First, the terms “perceptual load”
and “cognitive load” are vague, making it hard to test the theory (Murphy
et al., 2016). Second, the assumption that perceptual and cognitive load
have separate effects on attention is incorrect (Linnell & Caparos, 2011).
Third, perceptual load and attentional breadth are often confounded.
Fourth, the prediction that high cognitive load is associated with high
distractibility has been disproved when task and distracting stimuli are
easily distinguishable. Fifth, the theory de-emphasises several relevant
factors including the salience or conspicuousness of distracting stimuli and
the spatial distance between distracting and task stimuli (Murphy et al.,
2016).

Major attention networks

As we saw in Chapter 1, many cognitive processes are associated with net-
works spread across relatively large areas of cortex rather than small, spe-
cific regions. With respect to attention, several theorists (e.g., Posner, 1980;
Corbetta & Shulman, 2002) have argued there are two major networks.
192 Visual perception and attention

KEY TERM One attention network is goal-directed or endogenous whereas the other is
stimulus-driven or exogenous.
Covert attention
Attention to an object
in the absence of an eye
Posner’s (1980) approach
movement towards it.
Posner (1980) studied covert attention in which attention shifts to a given
spatial location without an accompanying eye movement. In his research,
people responded rapidly to a light. The light was preceded by a central
cue (arrow pointing to the left or right) or a peripheral cue (brief illumina-
tion of a box outline). Most cues were valid (i.e., indicating where the target
light would appear) but some were invalid (i.e., providing inaccurate infor-
mation about the light’s location).
Responses to the light were fastest to valid cues, intermediate to neutral
cues (a central cross) and slowest to invalid cues. The findings were com-
parable for central and peripheral cues. When the cues were valid on only
a small fraction of trials, they were ignored when they were central cues.
However, they influenced performance when they were peripheral cues.
The above findings led Posner (1980) to distinguish between two atten-
tion systems:

(1) An endogenous system: it is controlled by the individual’s intentions

and is used when central cues are presented.
(2) An exogenous system: it automatically shifts attention and is involved
when uninformative peripheral cues are presented. Stimuli that are
salient or different from other stimuli (e.g., in colour) are most likely
to be attended to using this system.

Corbetta and Shulman’s (2002) approach

Corbetta and Shulman (2002) identified two attention systems that are
involved in basic aspects of visual processing. First, there is a goal-directed
or top-down system resembling Posner’s endogenous system. This dorsal
attention network consists of a fronto-parietal network including the intra-
parietal sulcus. It is influenced by expectations, knowledge and current
goals. It is used when a cue predicts the location or other feature of a forth-
coming visual stimulus.
Second, Corbetta and Shulman (2002) identified a stimulus-driven or
bottom-up attention system resembling Posner’s exogenous system. This is
the ventral attention network and consists primarily of a right-­hemisphere
ventral fronto-parietal network. This system is used when an unexpected
and potentially important stimulus (e.g., flames appearing under the door)
occurs. Thus, it has a “circuit-breaking” function, meaning visual atten-
tion is redirected from its current focus. What stimuli trigger this circuit-­
breaking? According to Corbetta et al. (2008), non-task stimuli (i.e.,
distractors) closely resembling task stimuli are especially likely to activate
the ventral attention network although salient or conspicuous stimuli also
activate the same network.
Corbetta and Shulman (2011; see Figure 5.6) identified the brain
areas associated with each network. Key areas within the dorsal attention
network are as follows: superior parietal lobule (SPL), intraparietal sulcus
 Attention and performance 193

Figure 5.6
The brain areas associated
with the dorsal or goal-
directed attention network
and the ventral or stimulus-
driven network. The full
names of the areas involved
are indicated in the text.
From Corbetta and Shulman
(2011). © Annual Reviews. With
permission of Annual Reviews.

(IPS), inferior frontal junction (IFJ), frontal eye field (FEF), middle tem-
poral area (MT) and V3A (a visual area). Key areas within the ventral
attention network are as follows: inferior frontal junction (IFJ), inferior
frontal gyrus (IFG), supramarginal gyrus (SMG), superior temporal gyrus
(STG) and insula (Ins). The temporo-parietal junction also forms part of
the ventral attention network.
The existence of two attention networks makes much sense. The
goal-directed system (dorsal attention network) allows us to attend to
stimuli directly relevant to our current goals. If we only had this system,
however, our attentional processes would be dangerously inflexible. It is
also important to have a stimulus-driven attentional system (ventral atten-
tion network) leading us to switch attention away from goal-relevant stimuli
to unexpected threatening stimuli (e.g., a ferocious animal). More generally,
the two attention networks typically interact effectively with each other.

Findings
Corbetta and Shulman (2002) supported their two-network model by car-
rying out meta-analyses of brain-imaging studies. In essence, they argued,
brain areas most often activated when participants expect a stimulus that
has not yet been presented form the dorsal attention network. In contrast,
brain areas most often activated when individuals detect low-frequency
targets form the ventral attention network.
Hahn et al. (2006) tested Corbetta and Shulman’s (2002) theory by
comparing patterns of brain activation when top-down and bottom-up
processes were required. As predicted, there was little overlap between the
brain areas associated with top-down and bottom-up processing. In addi-
tion, the brain areas involved in each type of processing corresponded rea-
sonably well to those identified by Corbetta and Shulman.
Chica et al. (2013) reviewed research on the two attention systems and
identified 15 differences between them. For example, stimulus-driven atten-
tion is faster than top-down attention and is more object-based. In addi-
tion, it is more resistant to interference from other peripheral cues once
activated. The existence of so many differences strengthens the argument
the two attentional systems are separate.
Considerable research evidence (mostly involving neuroimaging) indi-
cates the dorsal and ventral attention systems are associated with distinct
 194 Visual perception and attention

neural circuits even during the resting state (Vossel et al., 2014). However,
neuroimaging studies cannot establish that any given brain area is necessar-
ily involved in stimulus-driven or goal-directed attention processes. Chica
et al. (2011) provided relevant evidence by using transcranial magnetic
stimulation (TMS; see Glossary) to interfere with processing in a given
brain area. TMS applied to the right temporo-parietal junction impaired
the functioning of the stimulus-driven system but not the top-down one.
In the same study, Chica et al. (2011) found TMS applied to the right
intraparietal sulcus impaired the functioning of both attention systems.
This provides evidence of the two attention systems working together.
Evidence from brain-damaged patients (discussed below, see pp. 196–200)
is also relevant to establishing the brain areas necessarily involved in goal-­
directed or stimulus-driven attentional processes. Shomstein et al. (2010)
had brain-damaged patients complete two tasks, one requiring stimulus-­
driven attentional processes whereas the other required top-down processes.
Patients having greater problems with top-down attentional processing typ-
ically had brain damage to the superior parietal lobule (part of the dorsal
attention network). In contrast, patients having greater problems with
stimulus-driven attentional processing typically had brain damage to the
temporo-­parietal junction (part of the ventral attention network).
Wen et al. (2012) investigated interactions between the two visual atten-
tion systems. They assessed brain activation while participants responded to
target stimuli in one visual field while ignoring all stimuli in the unattended
visual field. There were two main findings. First, stronger causal influences
of the top-down system on the stimulus-driven system led to superior per-
formance on the task. This finding suggests the appearance of an object
at the attended location caused the top-down attention system to suppress
activity within the stimulus-driven system.
Second, stronger causal influences of the stimulus-driven system
on the top-down system were associated with impaired task performance.
This finding suggests activation within the stimulus-driven system pro-
duced by stimuli not in attentional focus disrupted the attentional set
maintained by the top-down system.

Recent developments
Corbetta and Shulman’s (2002) theoretical approach has been developed
in recent years. Here we briefly consider three such developments. First, we
now have a greater understanding of interactions between their two atten-
tion networks. Meyer et al. (2018) found stimulus-driven and goal-directed
attention both activated frontal and parietal regions within the dorsal
attention network, suggesting it has a pivotal role in integrating bottom-up
and top-down processing.
Second, previous research reviewed by Corbetta and Shulman (2002)
indicated the dorsal attention network is active immediately prior to the
presentation of an anticipated visual stimulus. However, this research did
not indicate how long this attention network remained active. Meehan
et al. (2017) addressed this issue and discovered that top-down influences
associated within the dorsal attention network persisted over a relatively
long time period.

9781138482210_COGNITIVE_PSYCHOLOGY_PART_1.indd 194 28/02/20 6:44 PM

 Attention and performance 195

Third, brain networks relevant to attention additional to those within KEY TERM
Corbetta and Shulman’s (2002) theory have been identified (Sylvester
Default mode network
et al., 2012). One such network is the cingulo-opercular network including A network of brain
the anterior insula/operculum and dorsal anterior cingulate cortex (dACC; regions that is active
see Figure 5.7). This network is associated with non-selective attention or “by default” when an
alertness (Coste & Kleinschmidt, 2016). individual is not involved
in a current task; it is
Another additional network is the default mode network including
associated with internal
the posterior cingulate cortex (PCC), the lateral parietal cortex (LP), the processes including mind-
inferior temporal cortex (IT), the medial prefrontal cortex (MPF) and the wandering, remembering
subgenual anterior cingulate cortex (sgACC). The default mode network the past and imagining
is activated during internally focused cognitive processes (e.g., mind-­ the future.
wandering; imagining the future). What is the relevance of this network
to attention? In essence, performance on tasks requiring externally focused
attention is often enhanced if the default mode network is deactivated
(Amer et al., 2016a).
Finally, there is the fronto-parietal network (Dosenbach et al., 2008),
which includes the anterior dorsolateral prefrontal cortex (aDLPFC), the

(a) aDLPFC
IPS

TPJ

LP

VLPFC

anterior insula IT

(b) MCC
dACC

PCC
MPF

Figure 5.7
This is part of a theoretical
approach based on several
functional networks of
sgACC
relevance to attention:
the four networks shown
(fronto-parietal; default
Networks
mode; cingulo-opercular;
Key: Fronto-parietal Default mode and ventral attention) are all
discussed fully in the text.
Cingulo-opercular Ventral attention Sylvester et al., 2012, p. 528.
Reprinted with permission of
Elsevier.
196 Visual perception and attention

KEY TERMS middle cingulate cortex (MCC) and the intraparietal sulcus (IPS). It is
associated with top-down attentional and cognitive control.
Neglect
A disorder involving
right-hemisphere damage
Evaluation
(typically) in which the
left side of objects and/ The theoretical approach proposed by Corbetta and Shulman (2002) has
or objects presented to
several successes to its credit. First, there is convincing evidence for some-
the left visual field are
undetected; the condition what separate stimulus-driven and top-down attention systems, each with
resembles extinction but its own brain network. Second, research using transcranial magnetic stimu-
is more severe. lation suggests major brain areas within each attention system play a causal
Pseudo-neglect role in attentional processes. Third, some interactions between the two net-
A slight tendency in works have been identified. Fourth, research on brain-damaged patients
healthy individuals to supports the theoretical approach (see next section, pp. 196–200).
favour the left side of What are the limitations of this theoretical approach? First, the precise
visual space.
brain areas associated with each attentional system have not been clearly
identified. Second, there is more commonality (especially within the pari-
etal lobe) in the brain areas associated with the two attention networks
than assumed theoretically by Corbetta and Shulman (2002). Third, there
are additional attention-related brain networks not included within the
original theory. Fourth, much remains to be discovered about how differ-
ent attention systems interact.

DISORDERS OF VISUAL ATTENTION

Here we consider two important attentional disorders in brain-damaged
individuals: neglect and extinction. Neglect (or spatial neglect) involves a
lack of awareness of stimuli presented to the side of space on the oppo-
site side to the brain damage (the contralesional side). This occurs because
information from the left side of the visual field proceeds to the right
hemisphere.
Most neglect patients have damage in the right hemisphere and so
lack awareness of stimuli on the left side of the visual field: space-based or
egocentric neglect. For example, patients crossing out targets presented to
their left or right side (cancellation task) cross out more of those presented
to the right. When instructed to mark the centre of a horizontal line (line
bisection task), patients put it to the right of the centre. Note that the right
hemisphere is dominant in spatial attention in healthy individuals – they
exhibit pseudo-neglect, in which the left side of visual space is favoured
(Friedrich et al., 2018).
There is also object-centred or allocentric neglect involving a lack
of awareness of the left side of objects (see Figure 5.8). Patients with
right-hemisphere damage typically draw the right side of all figures in a
multi-object scene but neglect their left side in the left and right visual
fields (Gainotti & Ciaraffa, 2013).
Do allocentric and egocentric neglect reflect a single disorder or sepa-
rate disorders? Rorden et al. (2012) obtained two findings supporting the
single disorder explanation. First, the correlation between the extent of
each form of neglect across 33 patients was +.80. Second, similar brain
regions were associated with each type of neglect. However, Pedrazzini
et al. (2017) found damage to the intraparietal sulcus was more associated
 Attention and performance 197

Figure 5.8
On the left is a copying
task in which a patient with
unilateral neglect distorted
or ignored the left side of
the figures to be copied
(shown on the left). On the
right is a clock drawing
task in which the patient
was given a clock face and
told to insert the numbers
into it.
Reprinted from Danckert and
Ferber (2006). Reprinted with
permission from Elsevier.

with allocentric than egocentric neglect, whereas the opposite was the case
with damage to the temporo-parietal junction.
Extinction is often found in neglect patients. Extinction involves a
failure to detect a stimulus presented to the side opposite the brain damage
when a second stimulus is presented to the same side as the brain damage.
Extinction and neglect are closely related but separate deficits (de Haan
et al., 2012). We will focus mostly on neglect because it has attracted much
more research.
Which brain areas are damaged in neglect patients? Neglect is a het-
erogeneous condition and the brain areas damaged vary considerably
across patients. In a meta-analysis, Molenberghs et al. (2012) found the
main areas damaged in neglect patients are in the right hemisphere and
include the superior temporal gyrus, the inferior frontal gyrus, the insula,
the supramarginal gyrus and the angular gyrus (gyrus means ridge). Nearly
all these areas are within the stimulus-driven or ventral attention network Interactive feature:
(see Figure 5.6) suggesting brain networks are damaged rather than simply Primal Pictures’
specific brain areas (Corbetta & Shulman, 2011). 3D atlas of the brain
We also need to consider functional connectivity (correlated brain
activity between brain regions). Baldassarre et al. (2014, 2016) discovered
widespread disruption of functional connectivity between the hemispheres
in neglect patients. This disruption did not involve the bottom-up and top- KEY TERM
down attention networks. Of importance, recovery from attention deficits Extinction
in neglect patients was associated with improvements in functional con- A disorder of visual
nectivity in bottom-up and top-down attention networks (Ramsey et al., attention in which a
2016). stimulus presented to the
side opposite the brain
The right-hemisphere temporo-parietal junction and intraparietal sulcus damage is not detected
are typically damaged in extinction patients (de Haan et al., 2012). When when another stimulus is
transcranial magnetic stimulation is applied to these areas to interfere with presented at the same
processing, extinction-like behaviour results (de Haan et al., 2012). Dugué time to the side of the
et al. (2018) confirmed the importance of the temporo-parietal junction brain damage.
 198 Visual perception and attention

(part of the ventral attention network) in control of spatial attention in a

neuroimaging study on healthy individuals. However, its subregions varied
in terms of their involvement in voluntary and involuntary attention shifts.

Conscious awareness and processing

Neglect patients generally report no conscious awareness of stimuli pre-
sented to the left visual field. However, that does not necessarily mean those
stimuli are not processed. Vuilleumier et al. (2002b) presented extinction
patients with two pictures at the same time, one to each visual field. The
patients showed very little memory for left-field stimuli. Then the patients
identified degraded pictures. There was a facilitation effect for left-field pic-
tures indicating they had been processed.
Vuilleumier et al. (2002a) presented GK, a male patient with neglect
and extinction, with fearful faces. He showed increased activation in the
amygdala (associated with emotional responses) whether or not these faces
were consciously perceived. This is explicable given there is a processing
route from the retina to the amygdala bypassing the cortex (Diano et al.,
2017).
Sarri et al. (2010) found extinction patients had no awareness of left-
field stimuli. However, these stimuli were associated with activation in
early visual processing areas, indicating they received some processing.
Processing in neglect and extinction has been investigated using
event-related potentials. Di Russo et al. (2008) focused on the processing
of left-field stimuli not consciously perceived by neglect patients. Early
processing of these stimuli was comparable to that of healthy controls with
only later processing being disrupted. Lasaponara et al. (2018) obtained
similar findings in neglect patients. In healthy individuals, the presentation
of left-field targets inhibits processing of right-field space. This was less
the case in neglect patients, which helps to explain their lack of conscious
perception of left-field stimuli.

Theoretical considerations
Corbetta and Shulman (2011) discussed neglect in the context of their
two-system theory (discussed earlier, see pp. 192–196). In essence, the
bottom-up ventral attention network is damaged. Strong support for this
assumption was reported by Toba et al. (2018a) who found in 25 neglect
patients that impaired performance on tests of neglect was associated with
damage to parts of the ventral attention network (e.g., angular gyrus;
supramarginal gyrus). Since the right hemisphere is dominant in the ventral
attention network, neglect patients typically have damage in that hem-
isphere. Of importance, Corbetta and Shulman (2011) also assumed that
damage to the ventral network impairs the functioning of the goal-directed
dorsal attention network (even though not itself damaged).
How does the damaged ventral attention network impair the dorsal
attention network’s functioning? The two attention networks interact and
so damage to the ventral network inevitably affects the functioning of the
dorsal network. More specifically, damage to the ventral attention network
“impairs non-spatial [across the entire visual field] functions, hypoactivates

9781138482210_COGNITIVE_PSYCHOLOGY_PART_1.indd 198 28/02/20 6:44 PM

 Attention and performance 199

[reduces activation in] the right hemisphere, and unbalances the activity of
the dorsal attention network” (Corbetta & Shulman, 2011, p. 592).
de Haan et al. (2012) proposed a theory of extinction based on two
major assumptions:

(1) “Extinction is a consequence of biased competition for attention

between the ipsilesional [right-field] and contralesional [left-field]
target stimuli” (p. 1048);
(2) Extinction patients have much reduced attentional capacity so often
only one target [the right-field one] can be detected.

Findings
According to Corbetta and Shulman (2011), the dorsal attention network in
neglect patients functions poorly because of reduced activation in the right
hemisphere and associated reduced alertness and attentional resources.
Thus, increasing patients’ general alertness should enhance their detection
of left-field visual targets. Robertson et al. (1998) found the slower detec-
tion of left visual field stimuli compared to those in the right visual field was
no longer present when warning sounds were used to increase alertness.
Bonato and Cutini (2016) compared neglect patients’ ability to detect
visual targets with (or without) a second, attentionally demanding task.
Detection rates were high for targets presented to the right visual field in
both conditions. In contrast, patients detected only approximately 50% as
many targets in the left visual field as the right when performing another
task. Thus, neglect patients have limited attentional resources.
Corbetta and Shulman (2011) assumed neglect patients have an essen-
tially intact dorsal attention network. Accordingly, neglect patients might
use that network effectively if steps were taken to facilitate its use. Duncan
et al. (1999) presented arrays of letters and neglect patients recalled only
those in a pre-specified colour (the dorsal attention network could be used
to select the appropriate letters). Neglect patients resembled healthy con-
trols in showing equal recall of letters presented to each side of visual space.
The two attention networks typically work closely together. Bays et al.
(2010) studied neglect patients. They used eye movements during a visual
search to assess patients’ problems with top-down and stimulus-driven
attentional processes. Both types of attentional processes were equally
impaired (as predicted by Corbetta and Shulman, 2011). Of most impor-
tance, there was a remarkably high correlation of +.98 between these two
types of attentional deficit.
Toba et al. (2018b) identified two reasons for the failure of neglect
patients to detect left-field stimuli:

(1) a “magnetic” attraction of attention (i.e., right-field stimuli immedi-

ately capture attention).
(2) impaired spatial working memory making it hard for patients to keep
track of the locations of stimuli.

Both reasons were equally applicable to most patients. However, the

first reason was dominant in 12% of patients and the second reason in

9781138482210_COGNITIVE_PSYCHOLOGY_PART_1.indd 199 28/02/20 6:44 PM

200 Visual perception and attention

KEY TERM 24% of patients. Accordingly, Toba et al. argued we should develop
­multi-component models of visual neglect to account for such individual
Visual search
A task involving the rapid differences.
detection of a specified We turn now to extinction patients. According to de Haan et al.
target stimulus within a (2012), extinction occurs because of biased competition between stimuli.
visual display. If two stimuli could be integrated, that might minimise competition and so
reduce extinction. Riddoch et al. (2006) tested this prediction by present-
ing objects used together often (e.g., wine bottle and wine glass) or never
used together (e.g., wine bottle and ball). Extinction patients identified both
objects more often in the former condition than the latter (65% vs 40%,
respectively).
The biased competition hypothesis has been tested in other ways. We
can impair attentional processes in the intact left hemisphere by applying
transcranial magnetic stimulation to it. This should reduce competition
from the left hemisphere in extinction patients and thus reduce extinction.
Some findings are consistent with this prediction (Oliveri & Caltagirone,
2006).
de Haan et al. (2012) also identified reduced attentional capacity as a
factor causing extinction. Bonato et al. (2010) studied extinction with or
without the addition of a second, attentionally demanding task. As pre-
dicted, extinction patients showed a substantial increase in the extinction
rate (from 18% to over 80%) with this additional task.

Overall evaluation
Research has produced several important findings. First, neglect and
extinction patients can process unattended visual stimuli in the absence of
conscious awareness of those stimuli. Second, most neglect patients have
damage to the ventral attention network leading to impaired functioning of
the undamaged dorsal attention network. Third, extinction occurs because
of biased competition for attention and reduced attentional capacity.
What are the limitations of research in this area? First, it is hard
to produce theoretical accounts applicable to all neglect or extinction
patients because the precise symptoms and regions of brain damage
vary considerably across patients. Second, neglect patients vary in their
precise processing deficits (e.g., Toba et al., 2018b), but this has been
­de-emphasised in most theories. Third, the precise relationship between
neglect and e­xtinction remains unclear. Fourth, the dorsal and ventral
networks generally interact but the extent of their interactions remains to
be determined.

VISUAL SEARCH
We spend much time searching for various objects (e.g., a friend in a
crowd). The processes involved have been studied in research on visual
search where a specified target is detected as rapidly as possible. Initially,
we consider an important real-world situation where visual search can be
literally a matter of life-or-death: airport security checks. After that, we
consider an early very influential theory of visual search before discussing
more recent theoretical and empirical developments.
 Attention and performance 201

IN THE REAL WORLD: AIRPORT SECURITY CHECKS

Airport security checks have become more thorough since 9/11. When your luggage is x-rayed, an
airport security screener searches for illegal and dangerous items (see Figure 5.9). Screeners are
well trained but mistakes sometimes occur.

Figure 5.9
Each bag contains one illegal item. From left to right: a large bottle; a dynamite stick; and a gun part.
From Mitroff & Biggs (2014).

There are two major reasons it is often hard for airport security screeners to detect dangerous
items. First, illegal and dangerous items are (thankfully!) present in only a minute fraction of pas-
sengers’ luggage. This rarity of targets makes it hard for airport security screeners to detect them.
Mitroff and Biggs (2014) asked observers to detect illegal items in bags (see Figure 5.9). The
detection rate was only 27% when targets appeared under 0.15% of the time: they termed this
the “ultra rare item effect”. In contrast, the detection rate was 92% when targets appeared more
than 1% of the time.
Peltier and Becker (2016) tested two explanations for the reduced detection rate with rare targets:
(1) a reduced probability that the target is fixated (selection error); and (2) increased caution about
reporting targets because they are so unexpected (identification error). There was evidence for
both explanations. However, most detection failures were selection errors (see Figure 5.10).

Accuracy
1 Target present Target absent
0.9
0.8
0.7
Accuracy

0.6
0.5
0.4
0.3
0.2
0.1
0
10 50 90
Prevalence

Figure 5.10
Frequency of selection and identification errors when targets were present on 10%, 50% or 90% of trials.
From Peltier and Becker (2016).
202 Visual perception and attention

Second, security screeners search for numerous different objects. This increases search difficulty.
Menneer et al. (2009) found target detection was worse when screeners searched for two catego-
ries of objects (metal threats and improvised explosive devices) rather than one.
How can we increase the efficiency of security screening? First, we can exploit individual dif-
ferences in the ability to detect targets. Rusconi et al. (2015) found individuals scoring high on a
questionnaire measure of attention to detail had superior target-detection performance than low
scorers.
Second, airport security screeners can find it hard to distinguish between targets (i.e., danger-
ous items) and similar-looking non-targets. Geng et al. (2017) found that observers whose training
included non-targets resembling targets learned to develop increasingly precise internal target
representations. Such representations can improve the speed and accuracy of security screening.
Third, the low detection rate when targets are very rare can be addressed. Threat image pro-
jection (TIP) can be used to project fictional threat items into x-ray images of luggage to increase
the apparent frequency of targets. When screeners are presented with TIPs plus feedback when
they miss them, screening performance improves considerably (Hofer & Schwaninger, 2005). In
similar fashion, Schwark et al. (2012) found providing false feedback to screeners to indicate they
had missed rare targets reduced their cautiousness about reporting targets and improved their
performance.

Feature integration theory

Feature integration theory was proposed by Treisman and Gelade (1980)
and subsequently updated and modified (e.g., Treisman, 1998). According
to the theory, we need to distinguish between object features (e.g., colour;
size; line orientation) and the objects themselves. There are two processing
stages:

(1) Basic visual features are processed rapidly and pre-attentively in par-
allel across the visual scene.
(2) Stage (1) is followed by a slower serial process with focused atten-
tion providing the “glue” to form objects from the available features
(e.g., an object that is round and has an orange colour is perceived
as an orange). In the absence of focused attention, features from
different objects may be combined randomly producing an illusory
conjunction.

It follows from the above assumptions that targets defined by a single feature
(e.g., a blue letter or an S) should be detected rapidly and in parallel. In
contrast, targets defined by a conjunction or combination of features (e.g., a
green letter T) should require focused attention and so should be slower to
detect. Treisman and Gelade (1980) tested these predictions using both types
of targets; the display size was 1–30 items and a target was present or absent.
As predicted, response was rapid and there was very little effect of
KEY TERM
display size when the target was defined by a single feature: these findings
Illusory conjunction
Mistakenly combining
suggest parallel processing (see Figure 5.11). Response was slower and was
features from two strongly influenced by display size when the target was defined by a con-
different stimuli to junction of features: these findings suggest there was serial processing.
perceive an object that is According to the theory, lack of focused attention can produce illusory
not present. conjunctions based on random combinations of features. Friedman-Hill
 Attention and performance 203

Figure 5.11
Performance speed on
a detection task as a
function of target definition
(conjunctive vs single
feature) and display size.
Adapted from Treisman and
Gelade (1980).

et al. (1995) studied a brain-damaged patient (RM) having problems with

the accurate location of visual stimuli. This patient produced many illu-
sory conjunctions combining the shape of one stimulus with the colour of
another.

Limitations
What are the theory’s limitations? First, Duncan and Humphreys (1989,
1992) identified two factors not included within feature integration theory:

(1) When distractors are very similar to each other, visual search is faster
because it is easier to identify them as distractors.
(2) The number of distractors has a strong effect on search time to
detect even targets defined by a single feature when targets resemble
distractors.

Second, Treisman and Gelade (1980) estimated the search time with con-
junctive targets was approximately 60 ms per item and argued this repre-
sented the time taken for focal attention to process each item. However,
research with other paradigms indicates it takes approximately 250 ms for
attention indexed by eye movements to move from one location to another.
Thus, it is improbable focal attention plays the key role assumed within the
theory.
Third, the theory assumes visual search is often item-by-item. However,
the information contained within most visual scenes cannot be divided up
into “items” and so the theory is of limited applicability. Such considera-
tions led Hulleman and Olivers (2017) to produce an article entitled “The
impending demise of the item in visual search”.
Fourth, visual search involves parallel processing much more than
implied by the theory. For example, Thornton and Gilden (2007) used
29 different visual tasks and found 72% apparently involved parallel
204 Visual perception and attention

processing. We can explain such findings by assuming that each eye fixa-
tion permits considerable parallel processing using information available in
peripheral vision (discussed below, see pp. 206–208).
Fifth, the theory assumes that the early stages of visual search are
entirely feature-based. However, recent research using event-related poten-
tials indicates that object-based processing can occur much faster than pre-
dicted by feature integration theory (e.g., Berggren & Eimer, 2018).
Sixth, the theory assumes visual search is essentially random. This
assumption is wrong with respect to the real world – we typically use our
knowledge of where a target object is likely to be located when searching
for it (see below).

Dual-path model
In most of the research discussed so far, the target appeared at a random
location within the visual display. This is radically different from the real
world. Suppose you are outside looking for your missing cat. Your visual
search would be very selective – you would ignore the sky and focus mostly
on the ground (and perhaps the trees). Thus, your search would involve
top-down processes based on your knowledge of where cats are most likely
to be found.
Ehinger et al. (2009) studied top-down processes in visual search by
recording eye fixations of observers searching for a person in 900 real-
world outdoor scenes. Observers typically fixated plausible locations (e.g.,
pavements) and ignored implausible ones (e.g., sky; trees; see Figure 5.12).
Observers also fixated locations differing considerably from neighbouring
locations and areas containing visual features resembling those of a human
figure.
How can we reconcile Ehinger et al.’s (2009) findings with those
discussed earlier? Wolfe et al. (2011) proposed a dual-path model (see
Figure 5.13). There is a selective pathway of limited capacity (indicated
by the bottleneck) with objects being selected individually for recognition.

Figure 5.12
The first three eye fixations made by observers searching for pedestrians. As can be
seen, the great majority of their fixations were on regions in which pedestrians would
most likely be found. Observers’ fixations were much more like each other in the left-
hand photo than in the right-hand one, because there were fewer likely regions in the
left-hand one.
From Ehinger et al. (2009). Reprinted with permission from Taylor & Francis.
 Attention and performance 205

Figure 5.13
A two-pathway model of
y
wa visual search. The selective
a th pathway is capacity limited
e p
c tiv and can bind stimulus
ele features and recognise
ns
Early vision No objects. The non-selective
pathway processes the gist
of scenes. Selective and
non-selective processing
occur in parallel to produce
Features effective visual search.
Color
Orientation From Wolfe et al. (2011).
Size Reprinted with permission from
Depth Elsevier.
Select

Motion
Etc.
ive pa
thway

Binding and
recognition

This pathway has been the focus of most research until recently. There is
also a non-selective pathway in which the “gist” of a scene is processed.
Such processing can then guide processing within the selective pathway
(represented by the arrow labelled “guidance”). This pathway allows us to
utilise our stored environmental knowledge and so is of great value in the
real world.

Findings
Wolfe et al. (2011) compared visual searches for objects presented within a
scene setting or at random locations. As predicted, search rate per item was
much faster in the scene setting (10 ms vs 40 ms, respectively). Võ and Wolfe
(2012) explained that finding in terms of “functional set size” – searching in
scenes is efficient because most regions can be ignored. As predicted, Võ
and Wolfe found 80% of each scene was rarely fixated.
Kaiser and Cichy (2018) presented observers with objects typically
located in the upper (e.g., aeroplane; hat) or lower (e.g., carpet; shoe) visual
field. These objects were presented in their typical or atypical location
(e.g., hat in the lower visual field). Observers had to indicate whether an
object presented very briefly was located in the upper or lower visual field.
Observers’ performance was better when objects appeared in their typical
location because of their extensive knowledge of where objects are gener-
ally located.
Chukoskie et al. (2013) found observers can easily learn where
targets are located. An invisible target was presented at random loca-
tions on a blank screen and observers were provided with feedback. There
was a strong learning effect – fixations rapidly shifted from being fairly
206 Visual perception and attention

KEY TERM random to being focused on the area within which the target might be
present.
Fovea
A small area within the Ehinger et al.’s (2009) findings (discussed earlier, see p. 204) suggested
retina in the centre in the that scene gist or context can be used to enhance the efficiency of visual
field of vision where visual search. Katti et al. (2017) presented scenes very briefly (83 ms) followed by
acuity is greatest. a mask. Observers were given the task of detecting a person or a car and
performed very accurately (over 90%) and rapidly. Katti et al. confirmed
that scene gist or context influenced performance. However, performance
was influenced more strongly by features of the target object – the more
key features of an object were visible, the faster it was detected.
What is the take-home message from the above study? The efficiency
of visual search with real-world scenes is more complex than implied by
Ehinger et al. (2009). More specifically, observers may rapidly fixate on the
area close to a target person because they are using scene gist or because
they rapidly process features of the person (e.g., wearing clothes).

Evaluation
Our knowledge of likely (and unlikely) locations for any given object in a
scene influences visual search in the real world. This is fully acknowledged
in the dual-path model. There is also support for the notion that scene
knowledge facilitates visual search by reducing functional set size.
What are the model’s limitations? First, how we use gist knowledge
of a scene very rapidly to reduce the search area remains unclear. Second,
there is insufficient focus on the learning processes that can greatly facil-
itate visual search – the effects of such processes can be seen in the very
rapid and accurate detection of target information by experts in several
domains (see Chapter 11).
Third, it is important not to exaggerate the importance of scene gist or
context in influencing the efficiency of visual search. Features of the target
object can influence visual search more than scene gist (Katti et al., 2017).
Fourth, the assumption that items are processed individually within
the selective pathway is typically mistaken. As we will see shortly, visual
search often depends on parallel processes within peripheral vision and
such processes are not considered within the model.

Attention vs perception: texture tiling model

Several theories (e.g., Treisman & Gelade, 1980) have assumed that indi-
vidual items are the crucial units in visual search. Such theories have also
often assumed that slow visual search depends mostly on the limitations of
focused attention. A plausible implication of these assumptions is that slow
visual search depends mostly on foveal vision (the fovea is a small area of
maximal visual acuity in the retina).
Both the above assumptions have been challenged recently. At the
risk of oversimplification, full understanding of visual search requires less
emphasis on attention and more on perception. According to Rosenholtz
(2016), peripheral (non-foveal) vision is of crucial importance. Acuity
decreases as we move away from the fovea to the periphery of vision, but
much less than often assumed. You can demonstrate this by holding out
 Attention and performance 207

your thumb and fixating the nail. Foveal vision only covers the nail so the KEY TERM
great majority of what you can see is in peripheral vision.
Visual crowding
We can also compare the value of foveal and peripheral vision by con- The inability to recognise
sidering individuals with impaired eyesight. Those with severely impaired objects in peripheral
peripheral vision (e.g., due to glaucoma) had greater problems with mobil- vision due to the presence
ity (e.g., number of falls; ability to drive) than those who lack foveal vision of neighbouring objects.
(due to macular degeneration) (Rosenholtz, 2016). Individuals with severely
impaired central or foveal vision performed almost as well as healthy con-
trols at detecting target objects in coloured scenes (75% vs 79%, respec-
tively) (Thibaut et al., 2018).
If visual search depends heavily on peripheral vision, what predictions
can we make? First, if each fixation provides observers with a considerable
amount of information about several objects, visual search will typically
involve parallel rather than serial processing. Second, we need to consider
limitations of peripheral vision (e.g., visual acuity is less in peripheral
than foveal vision). However, a more important limitation concerns visual
crowding – a reduced ability to recognise objects or other stimuli because
of irrelevant neighbouring objects or stimuli (clutter). Visual crowding
impairs peripheral vision to a much greater extent than foveal vision.
Rosenholtz et al. (2012) proposed the texture tiling model based on
the assumption peripheral vision is of crucial importance in visual search.
More specifically, processing in peripheral vision can cause adjacent stimuli
to tile (join together) to form an apparent target, thus increasing the diffi-
culty of visual search. Below we consider findings relevant to this model.

Findings
As mentioned earlier (p. 203), Thornton and Gilden (2007) found almost
three-quarters of the visual tasks they studied involved parallel processing.
This is entirely consistent with the emphasis on parallel processing in the
model.
Direct evidence for the importance of peripheral vision to visual search
was reported by Young and Hulleman (2013). They manipulated the visible
area around the fixation point making it small, medium or large. As pre-
dicted by the model, visual search performance was worst when the visible
area was small (so only one item could be processed per fixation). Overall,
visual search was almost parallel when the visible area was large but serial
when it was small.
Chang and Rosenholtz (2016) used various search tasks. According to
feature integration theory, both tasks shown in Figure 5.14 should be com-
parably hard because the target and distractors share features. In contrast,
the texture tiling model predicts the task on the right should be harder
because adjacent distractors seen in peripheral vision can more easily tile
(join together) to form an apparent T. The findings from these tasks (and
several others) supported the texture tiling model but were inconsistent
with feature integration theory.
Finally, Hulleman and Olivers (2017) produced a model of visual
search consistent with the texture tiling model. According to this model,
each eye fixation lasts 250 ms, during which information from foveal and
peripheral vision is extracted in parallel. They also assumed that the area
208 Visual perception and attention

Figure 5.14
The target (T) is easier to (a) Easier search (b) Harder search
find in the display on the
left than the one on the
right.
From Chang and Rosenholtz
(2016). Find the T

around the fixation point within which a target can generally be detected
is smaller when the visual search task is difficult (e.g., because target dis-
criminability is low).
A key prediction from Hulleman and Olivers’ (2017) model is that the
main reason why search times are longer with more difficult search tasks is
because more eye fixations are required than with easier tasks. A computer
simulation based on these assumptions produced search times very similar
to those obtained in experimental studies.

Evaluation
What are the strengths of the texture tiling model? First, the information
available in peripheral vision is much more important in visual search than
assumed previously. The model explains how observers make use of the
information available in peripheral vision.
Second, the model explains why parallel processing is so prevalent
in visual search – it reflects directly parallel processing within peripheral
vision. Third, there is accumulating evidence that search times are gener-
ally directly related to the number of eye fixations.
Fourth, an approach based on eye fixations and peripheral vision can
potentially explain findings from all visual search paradigms, including
complex visual scenes and item displays. Such an approach thus has more
general applicability than feature integration theory.
What are the model’s limitations? First, as Chang and Rosenholtz
(2016) admitted, it needs further development to account fully for visual
search performance. For example, it does not predict search times with
precision. In addition, it does not specify the criteria used by observers to
decide no target is present.
Second, visual search is typically much faster for experts than non-­
experts in their domain of expertise (e.g., medical experts examining mam-
mograms) (see Chapter 11). The texture tiling model does not identify
clearly the processes allowing experts to make very efficient use of periph-
eral information.

CROSS-MODAL EFFECTS
Nearly all the research discussed so far is limited in that the visual (or audi-
tory) modality was studied on its own. We might try to justify this approach
by assuming attentional processes in each sensory modality operate
 Attention and performance 209

independently from those in other modalities. However, that assumption is KEY TERMS
incorrect. In the real world, we often coordinate information from two or
Cross-modal attention
more sense modalities at the same time (cross-modal attention). An example The coordination of
is lip reading, where we use visual information about a speaker’s lip move- attention across two or
ments to facilitate our understanding of what they are saying (see Chapter 9). more modalities (e.g.,
Suppose we present participants with two streams of lights (as was vision and audition).
done by Eimer and Schröger, 1998), with one stream being presented to the Ventriloquism effect
left and the other to the right. At the same time, we present participants The mistaken perception
with two streams of sounds (one to each side). In one condition, partici- that sounds are
coming from their
pants detect deviant visual events (e.g., longer than usual stimuli) presented
apparent source (as in
to one side only. In the other condition, participants detect deviant audi- ventriloquism).
tory events in only one stream.
Event-related potentials were recorded to assess the allocation of atten-
tion. Unsurprisingly, Eimer and Schröger (1998) found ERPs to deviant
stimuli in the relevant modality were greater to stimuli presented on the
to-be-attended side than the to-be-ignored side. Thus, participants allo-
cated attention as instructed.
Of more interest is what happened to the allocation of attention in
the irrelevant modality. Suppose participants detected visual targets on the
left side. In that case, ERPs to deviant auditory stimuli were greater on
the left side than the right. This is a cross-modal effect: the voluntary or
endogenous allocation of visual attention also affected the allocation of
auditory attention. Similarly, when participants detected auditory targets
on one side, ERPs to deviant visual stimuli on the same side were greater
than ERPs to those on the opposite side. Thus, the allocation of auditory
attention also influenced the allocation of visual attention.

Ventriloquism effect
What happens when there is a conflict between simultaneous visual and
auditory stimuli? We will focus on the ventriloquism effect in which
sounds are misperceived as coming from their apparent visual source.
Ventriloquists (at least good ones!) speak without moving their lips while
manipulating a dummy’s mouth movements. It seems as if the dummy is
speaking. Something similar happens at the movies. The actors’ lips move
on the screen but their voices come from loudspeakers beside the screen.
Nevertheless, we hear those voices coming from their mouths.
Certain conditions must be satisfied for the ventriloquism effect to
occur (Recanzone & Sutter, 2008). First, the visual and auditory stimuli
must occur close together in time. Second, the sound must match expecta-
tions created by the visual stimulus (e.g., high-pitched sound coming from
a small object). Third, the sources of the visual and auditory stimuli should
be close together spatially. More generally, the ventriloquism effect reflects
the unity assumption (the assumption that two or more sensory cues come
from the same object: Chen & Spence, 2017).
The ventriloquism effect exemplifies visual dominance (visual informa-
tion dominating perception). Further evidence comes from the Colavita
effect (Colavita, 1974): participants instructed to respond to all stimuli
respond more often to visual than simultaneous auditory stimuli (Spence
et al., 2011).
210 Visual perception and attention

KEY TERM When during processing is visual spatial information integrated with
auditory information? Shrem et al. (2017) found that misleading visual
Temporal ventriloquism
effect information about the location of an auditory stimulus influenced the pro-
Misperception of the cessing of the auditory stimulus approximately 200 ms after stimulus onset.
timing of a visual stimulus The finding that this effect is still present even when participants are aware
when an auditory stimulus of the spatial discrepancy between the visual and auditory input suggests it
is presented close to it in
occurs relatively “automatically”.
time.
However, the ventriloquism effect is smaller when participants had
previously heard syllables spoken in a fearful voice (Maiworm et al., 2012).
This suggests the effect is not entirely “automatic” but is reduced when the
relevance of the auditory channel is increased.
Why does vision capture sound in the ventriloquism effect? The visual
modality typically provides more precise information about spatial loca-
tion. However, when visual stimuli are severely blurred and poorly local-
ised, sound captures vision (Alais & Burr, 2004). Thus, we combine visual
and auditory information effectively by attaching more weight to the more
informative sense modality.

Temporal ventriloquism
The above explanation for the ventriloquist illusion is a development of
the modality appropriateness and precision hypothesis (Welch & Warren,
1980). According to this hypothesis, when conflicting information is pre-
sented in two or more modalities, the modality having the greatest acuity
generally dominates. This hypothesis predicts the existence of another illu-
sion. The auditory modality is typically more precise than the visual modal-
ity at discriminating temporal relations. As a result, judgements about
the temporal onset of visual stimuli might be biased by auditory stimuli
presented very shortly beforehand or afterwards. This is the temporal
­ventriloquism effect.
Research on temporal ventriloquism
was reviewed by Chen and Spence (2017).
A simple example is when the apparent
onset of a flash is shifted towards an abrupt
sound presented slightly asynchronously (see
Figure 5.15). Other research has found that
the apparent duration of visual stimuli can be
distorted by asynchronous auditory stimuli.
We need to consider the temporal ven-
triloquism effect in the context of the unity
assumption. This is the assumption that “two
or more uni-sensory cues belong together
(i.e., that they come from the same object or
event)” (Chen & Spence, 2017, p. 1). Chen
and Spence discussed findings showing that
Figure 5.15 the unity assumption generally (but not
An example of temporal ventriloquism in which the apparent
always) enhances the temporal ventriloquism
time of onset of a flash is shifted towards that of a sound
presented at a slightly different timing from the flash. effect.
From Chen and Vroomen (2013). Reprinted with permission from
Orchard-Mills et al. (2016) extended
Springer. research by using two visual stimuli (one
 Attention and performance 211

IN THE REAL WORLD: WARNING SIGNALS PROMOTE SAFE DRIVING

Front-to-rear-end collisions cause 25% of road accidents with driver inattention the most common
cause (Spence, 2012). Thus, it is important to devise effective warning signals to enhance driver
attention and reduce collisions. Warning signals might be especially useful if they were informative
(i.e., indicating the nature of the danger). However, informative warning signals requiring time-con-
suming cognitive processing might be counterproductive.
Ho and Spence (2005) considered drivers’ reaction times when braking to avoid a car in front or
accelerating to avoid a speeding car behind. An auditory warning signal (car horn) came from the
same direction as the critical visual event on 80% or 50% of trials. Braking times were faster when
the sound and critical visual event were from the same direction. The greater beneficial effects of
auditory signals when predictive rather than non-predictive suggests the involvement of endoge-
nous spatial attention (controlled by the individual’s intentions). Auditory stimuli also influenced
visual attention even when non-predictive: this probably involved exogenous spatial attention
(“automatic” allocation of attention).
Gray (2011) studied braking times to avoid a collision with the car in front when drivers heard
auditory warning signals increasing in intensity as the time to collision reduced. These signals are
known as looming sounds. The most effective condition was the one where the rate of increase
in the intensity of the auditory signal was the fastest because it implied the time to collision was
the least. Lahmer et al. (2018) found evidence that looming sounds are effective because they are
consistent with the visual experience of an approaching collision.
Vibrotactile signals produce the perception of vibration through touch. Gray et al. (2014) studied
the effects of such signals on speed of braking to avoid a collision. Signals were presented at three
sites on the abdomen arranged vertically. In the most effective condition, successive signals moved
towards the driver’s head at an increasing rate reflecting the speed they were approaching the
car in front. Braking time was 250 ms faster in this condition than a no-warning control condition,
probably because it was highly informative.
Ahtamad et al. (2016) compared the effectiveness of three vibrotactile warning signals delivered
to the back on braking times to avoid a collision with the car in front: (1) expanding (centre of back
followed by areas to left and right); (2) contracting (areas to left and right followed by the centre
of the back); (3) static (centre of the back + areas to left and right at the same time). The dynamic
vibrotactile conditions (1 and 2) produced comparable braking reaction times that were faster than
those in the static condition (3).
In a second experiment, Ahtamad et al. (2016) compared the expanding vibrotactile condition
against a linear motion condition (vibrotactile stimulation to the hands followed by the shoulders).
Emergency braking reaction times were faster in the linear motion condition (approximately 585 ms
vs 640 ms) because drivers found it easier to interpret the warning signals in that condition.
In sum, the various auditory and vibrotactile warning signals discussed above typically reduce
braking reaction times by approximately 40 ms. That sounds modest. However, it can easily be the
difference between colliding with the car in front or avoiding it and so could potentially save many
lives. At present, however, we lack a theoretical framework within which to understand precisely
why some warning signals are more effective than others.

above and the other below fixation) and two auditory stimuli (low- and
high-pitch). When the visual and auditory stimuli were congruent (e.g.,
visual stimulus above fixation and auditory stimulus high-pitch), the tem-
poral ventriloquism effect was found. However, this effect was eliminated
when the visual and auditory stimuli were incongruent, which prevented
binding of information across the two senses.

9781138482210_COGNITIVE_PSYCHOLOGY_PART_1.indd 211 28/02/20 6:44 PM

212 Visual perception and attention

KEY TERMS Overall evaluation

Endogenous spatial
attention What are the limitations of research on cross-modal effects? First, as just
Attention to a stimulus mentioned, our theoretical understanding lags behind the accumulation of
controlled by intentions empirical findings. Second, much research has involved complex artificial
or goal-directed tasks far removed from naturalistic conditions. Third, individual differ-
mechanisms.
ences have generally been ignored. However, individual differences (e.g.,
Exogenous spatial preference for auditory or visual stimuli) influence cross-modal effects (van
attention Atteveldt et al., 2014).
Attention to a given
spatial location
determined by DIVIDED ATTENTION: DUAL-TASK PERFORMANCE
“automatic” processes.
In this section, we consider factors influencing how well we can perform
Multi-tasking
two tasks at the same time. In our hectic 24/7 lives, we increasingly try to
Performing two or more
tasks at the same time by do two things at once (multi-tasking) (e.g., sending text messages while
switching rapidly between walking down the street). More specifically, multi-tasking “refers to the
them. ability to co-ordinate the completion of several tasks to achieve an overall
goal” (MacPherson, 2018, p. 314). It can involve performing two tasks at
the same time or switching between two tasks. There is controversy as to
whether massive amounts of multi-tasking have beneficial or detrimental
effects on attention and cognitive control (see Box).
What determines how well we can perform two tasks at once?
Similarity (e.g., in terms of modality) is one important factor. Treisman
and Davies (1973) found two monitoring tasks interfered with each other
Case study: much more when the stimuli on both tasks were in the same modality
Multi-tasking efficiency (visual or auditory).
Two tasks can also be similar in response modality. McLeod (1977)
had participants perform a continuous tracking task with manual respond-
ing together with a tone-identification task. Some participants responded
vocally to the tones whereas others responded with the hand not involved
in tracking. Tracking performance was worse with high response similarity
(manual responses on both tasks) than with low response similarity.
Practice is the most important factor determining how well two tasks
can be performed together. The saying “Practice makes perfect” was
apparently supported by Spelke et al. (1976). Two students (Diane and
John) received 5 hours of training a week for 4 months on various tasks.
Their first task involved reading short stories for comprehension while
writing down words from dictation, which they initially found very hard.
After 6 weeks of training, however, they could read as rapidly and with
as much comprehension when writing to dictation as when only reading.
With further training, Diane and John learned to write down the names
of the categories to which the dictated words belonged while maintaining
normal reading speed and comprehension.
Spelke et al.’s (1976) findings are hard to interpret for various reasons.
First, Spelke et al. focused on accuracy measures, which are typically less
sensitive to dual-task interference than speed measures. Second, Diane and
John’s attentional focus was relatively uncontrolled, and so they may have
alternated attention between tasks rather than attending to both at the
same time. More controlled research on the effects of practice on dual-task
performance is discussed later.
 Attention and performance 213

IN THE REAL WORLD: MULTI-TASKING

What are the effects of frequent multi-tasking in our everyday lives? Two main answers have been
proposed. First, heavy multi-tasking may impair cognitive control because it leads individuals to
allocate their attentional resources too widely. This is the scattered attention hypothesis (van der
Schuur et al., 2015).
Second, heavy multi-tasking may enhance some control processes (e.g., task switching) because
of prolonged practice in processing multiple streams of information. This is the trained attention
hypothesis (van der Schuur et al., 2015). The relevant evidence is very inconsistent – “positive,
negative, and null effects have all been reported” (Uncapher & Wagner, 2018, p. 9894).
Ophir et al. (2009) used a questionnaire (the Media Multitasking Index) to identify levels of
multi-tasking. Heavy multi-taskers were more distractible. In a review, van der Schuur et al. (2015)
found findings supported the scattered attention hypothesis (e.g., heavy multi-taskers had impaired
sustained attention).
Moisala et al. (2016) found heavy multi-taskers were more adversely affected than light
­multi-taskers by distracting stimuli while performing speech–listening and reading tasks. During
distraction, the heavy multi-taskers had greater activity than the light multi-taskers in the right pre-
frontal cortex (associated with attentional control). This suggests heavy multi-taskers have greater
problems than previously believed – their performance is impaired even though they try harder to
exert top-down attentional control.
Uncapher and Wagner (2018) found in a review that most research indicated negative effects
of heavy multi-tasking on tasks involving working memory, long-term memory, sustained atten-
tion and relational reasoning. These negative effects are likely to be due to attentional lapses.
Of relevance, there are several studies where media multi-tasking was positively associated with
self-reported everyday attentional failures. In addition, heavy multi-taskers often report high impul-
sivity – such individuals often make rapid decisions based on very limited evidence.
Most studies have only found an association between media multi-tasking and measures of
attention and performance. This makes it hard to establish causality – it is possible individuals with
certain patterns of attention choose to engage in extensive multi-tasking. Evidence suggesting
that media multi-tasking can cause attention problems was reported by Baumgartner et al. (2018).
They found that high media multi-tasking at one point in time predicted attention problems several
months later.

Serial vs parallel processing

When individuals perform two tasks together, they might use serial or par-
allel processing. Serial processing involves switching attention backwards
and forwards between two tasks with only one task being processed at any
given moment. In contrast, parallel processing involves processing both
tasks at the same time.
There has been much theoretical controversy on the issue of serial vs
parallel processing in dual-task conditions (Koch et al., 2018). Of impor-
tance, processing can be mostly parallel or mostly serial. Lehle et al. (2009)
trained participants to use serial or parallel processing when perform-
ing two tasks together. Those using serial processing performed better.
However, they found the tasks more effortful because they had to inhibit
processing of one task while performing the other one.
Lehle and Hübner (2009) also instructed participants to perform two
tasks together in a serial or parallel fashion. Those using parallel processing

9781138482210_COGNITIVE_PSYCHOLOGY_PART_1.indd 213 28/02/20 6:44 PM

 214 Visual perception and attention

performed much worse. Fischer and Plessow (2015) reviewed dual-task

research and concluded: “While serial task processing appears to be the
most efficient [dual-task] processing strategy, participants are able to adopt
parallel processing. Moreover, parallel processing can even outperform
serial processing under certain conditions” (p. 8).
Brüning and Manzey (2018) confirmed serial processing is not always
more efficient than parallel processing. Participants performed many alter-
nate trials on two different tasks but could see the stimulus for the next
trial ahead of time. Participants engaging in parallel processing (process-
ing the stimulus for trial n+1 during trial n) performed better than those
using only serial processing (not processing the trial n+1 stimulus ahead of
time). Parallel processing reduced the costs incurred when task switching.
Individuals high in working memory capacity (see Glossary) were more
likely to use parallel processing, perhaps because of their superior atten-
tional control.

IN THE REAL WORLD: CAN WE THINK AND DRIVE?

Car driving is the riskiest activity engaged in by tens of millions of adults. Over 50 countries have
laws restricting the use of mobile or cell phones by drivers to increase car safety. Are such restric-
tions necessary? The short answer is “Yes” – drivers using a mobile phone are several times more
likely to be involved in a car accident (Nurullah, 2015). This is so even though drivers try to reduce
the risks by driving slightly more slowly (reducing speed by 5–6 mph) than usual shortly after initi-
ating a mobile-phone call (Farmer et al., 2015).
Caird et al. (2008) in a review of studies using simulated driving tasks reported that reaction
times to events (e.g., onset of brake lights on the car in front) increased by 250 ms with mobile-
phone use and were greater when drivers were talking rather than listening. This 250 ms increase
in reaction time translates into travelling an extra 18 feet (5.5 metres) before stopping for a motor-
ist doing 50 mph (80 kph). This could be the difference between stopping just short of a child or
killing that child.
Strayer and Drews (2007) studied the above slowing effect using event-related potentials while
drivers responded rapidly to the onset of brake lights on the car in front. The magnitude of the
P300 (a positive wave associated with attention) was reduced by 50% in mobile-phone users.
Strayer et al. (2011) considered a real-life driving situation. Drivers were observed to see whether
they obeyed a law requiring them to stop at a road junction. Of drivers not using a mobile phone,
79% obeyed the law compared to only 25% of mobile-phone users.

Theoretical considerations
Why do so many drivers endanger people’s lives by using mobile phones? Most believe they
can drive safely while using a mobile phone whereas other drivers cannot (Sanbonmatsu et al.,
2016b). Their misplaced confidence depends on limited monitoring of their driving performance:
drivers using a mobile phone make more driving errors but do not remember making more errors
(Sanbonmatsu et al., 2016a).
Why does mobile-phone use impair driving performance? Strayer and Fisher (2016) in their
SPIDER model identified five cognitive processes that are adversely affected when drivers’ ­attention
is diverted from driving (e.g., by mobile-phone use):

9781138482210_COGNITIVE_PSYCHOLOGY_PART_1.indd 214 28/02/20 6:44 PM

 Attention and performance 215

(1) There is less effective visual scanning of the environment for potential threats. Distracted drivers
are more inclined to focus attention on the centre of the road and less inclined to scan objects
in the periphery and their side mirrors (Strayer & Fisher, 2016).
(2) The ability to predict where threats might occur is impaired. Distracted drivers are much less
likely to make anticipatory glances towards the location of a potential hazard (e.g., obstructed
view of a pedestrian crossing) (Taylor et al., 2015).
(3) There is reduced ability to identify visible threats, a phenomenon known as inattentional blind-
ness (see Glossary; and Chapter 4). In a study by Strayer and Drews (2007), 30 objects (e.g.,
pedestrians; advertising hoardings) were clearly visible to drivers. However, those using a
mobile phone subsequently recognised far fewer objects they had fixated than those not using
a mobile phone (under 25% vs 50%, respectively).
(4) It is harder to decide what action is necessary in a threatening situation. Cooper et al. (2009)
found drivers were 11% more likely to make unsafe lane changes when using a mobile
phone.
(5) It becomes harder to execute the appropriate action. Reaction times are slowed (Caird et al.,
2008, discussed above, p. 214).

The SPIDER model is oversimplified in several ways. First, various different activities are associated
with mobile-phone use. Simmons et al. (2016) found in a meta-analytic review that the risk of safety-­
critical events was increased by activities requiring drivers to take their eyes off the road (e.g., locat-
ing a phone; dialling; texting). However, talking on a mobile phone did not increase risk.
Second, driving-irrelevant cognitive activities do not always impair all aspects of driving per-
formance. Engstrom et al. (2017, p. 734) proposed their cognitive control hypothesis: “Cognitive
load selectively impairs driving sub-tasks that rely on cognitive control but leaves automatic perfor-
mance unaffected.” For example, driving-irrelevant activities involving cognitive load (e.g., mobile-
phone use) typically have no adverse effect on well-practised driving skills, such as lane keeping
and braking when getting close to the vehicle in front (Engstrom et al., 2017).
Third, individuals using mobile phones while driving are unrepresentative of drivers in general
(e.g., they tend to be relatively young and to engage in more risk-taking activities: Precht et al.,
2017). Thus, we must consider individual differences in personality and risk taking when interpret-
ing accidents associated with mobile-phone use.
Fourth, the SPIDER model implies that performance cannot be improved by adding a secondary
task. However, driving performance in monotonous conditions is sometimes better when drivers
listen to the radio at the same time (see Engstrom et al., 2017). Listening to the radio can reduce
the mind-wandering that occurs when someone drives in monotonous conditions. Drivers indicat-
ing their immediate thoughts during their daily commute reported mind-wandering 63% of the
time and active focus on driving only 15%–20% of the time (Burdett et al., 2018).

Multiple resource theory

Wickens (1984, 2008) argued in his multiple resource model that the pro-
cessing system consists of several independent processing resources or
mechanisms. The model includes four major dimensions (see Figure 5.16):

(1) Processing stages: there are successive stages of perception, cognition

(e.g., working memory) and responding.
(2) Processing codes: perception, cognition and responding can use
spatial and/or verbal codes; action can involve speech (vocal verbal)
or manual/spatial responses.
216 Visual perception and attention

Figure 5.16
Wickens’s four-dimensional
multiple resource model.
The details are described in
the text.
From Wickens (2008). © 2008.
Reprinted by permission of
SAGE Publications.

(3) Modalities: perception can involve visual and/or auditory resources.

(4) Visual channels: visual processing can be focal (high acuity) or ambient
(peripheral).

Here is the model’s crucial prediction: “To the extent that two tasks use dif-
ferent levels along each of the three dimensions [excluding (4) above], time-­
sharing [dual-task performance] will be better” (Wickens, 2008, p. 450). Thus,
tasks requiring different resources can be performed together more success-
fully than those requiring the same resources. Wickens’s approach bears
some resemblance to Baddeley’s (e.g., 2012) working memory model (see
Chapter 6). According to that model, two tasks can be performed together
successfully provided they use different components or processing resources.

Findings
Research discussed earlier (Treisman & Davies, 1973; McLeod, 1977)
showing the negative effects of stimulus and response similarity on per-
formance are entirely consistent with the theory. Lu et al. (2013) reviewed
research where an ongoing visual-motor task (e.g., car driving) was
­performed together with an interrupting task in the visual, auditory or
tactile (touch) modality. As predicted, non-visual interrupting tasks (espe-
cially those in the tactile modality) were processed more effectively than
visual ones and there were no adverse effects on the visual-motor task.
According to the model, there should be only limited dual-task inter-
ference between two visual tasks if one requires focal or foveal vision,
whereas the other requires ambient or peripheral vision. Tsang and Chan
(2018) obtained support for this prediction in a study in which participants
tracked a moving target in focal vision while responding to a spatial task
in ambient or peripheral vision.
Dual-task performance is often more impaired than predicted by
the theory. For example, consider a study by Robbins et al. (1996; see
 Attention and performance 217

Chapter 6). The main task was selecting chess moves and we will focus on
the condition where the task performed at the same time was generating
random letters. These two tasks involve different processing codes (spatial
vs verbal, respectively) and they also involve different response types
(manual vs vocal, respectively). Nevertheless, generating random letters
caused substantial interference on the chess task.

Evaluation
The main assumptions of the theory have largely been supported by the
experimental evidence. In other words, dual-task performance is generally
less impaired when two tasks differ with respect to modalities, processing
codes or visual channels than when they do not.
What are the model’s limitations?

(1) Successful dual-task performance often requires higher-level processes

of coordinating and organising the demands of the two tasks (see
later section on cognitive neuroscience, pp. 220–222). However, these
processes are de-emphasised within the theory.
(2) The theory’s assumption there is a sequence of processing stages (per-
ception; cognition; responding) is too rigid given the flexible nature of
much dual-task processing (Koch et al., 2018). The numerous forms
of cognitive processing intervening between perception and respond-
ing are not discussed in detail.
(3) It is implied within the theory that negative or interfering effects of
performing two tasks together would be constantly present. However,
Steinborn and Huestegge (2017) found dual-task conditions led only
to occasional performance breakdown due to attention failures.

Threaded cognition
Salvucci and Taatgen (2008, 2011) proposed a model of threaded cogni-
tion in which streams of thought are represented as threads of processing.
For example, processing two tasks might involve two separate threads. The
central theoretical assumptions are as follows:

Multiple threads or goals can be active at the same time, and as long as
there is no overlap in the cognitive resources needed by these threads,
there is no multi-tasking interference. When threads require the same
resource at the same time, one thread must wait and its performance
will be adversely affected.
(Salvucci & Taatgen, 2011, p. 228)

This is because all resources have limited capacity.

Taatgen (2011) discussed the threaded cognition model (see
Figure 5.17). Several cognitive resources can be the source of competition
between two tasks. These include visual perception, declarative memory,
task control and focal working memory or problem state. Nijboer et al.
(2016a) discussed similarities between this model and Baddeley’s working
memory model (see Chapter 6). Three components of the model relate to
218 Visual perception and attention

working memory: (1) problem state (atten-

tional focus); (2) declarative memory (acti-
vated short-term memory); and (3) subvocal
rehearsal (resembling the phonological loop;
see Chapter 6).
Each thread or task controls resources
in a greedy, polite way – threads claim
resources greedily when required but release
them politely when no longer needed. These
aspects of the model lead to one of its most
original assumptions – several goals (each
associated with a given thread) can be active
simultaneously.
The model resembles Wickens’s multiple
resource model: both models assume there
are several independent processing resources.
However, only the threaded cognition model
led to a computational model making spe-
cific predictions. In addition, the threaded
Figure 5.17
cognition model identifies the brain areas
Threaded cognition theory. We possess several cognitive
resources (e.g., declarative memory, task control, visual
associated with each processing resource (see
perception). These resources can be used in parallel but each Figure 5.17).
resource can only work on one task at a time. Our ability to
perform two tasks at the same time (e.g., driving and dialling,
subtraction and typing) depends on the precise ways in which
Findings
cognitive resources need to be used. The theory also identifies According to the model, any given cogni-
some of the brain areas associated with cognitive resources.
tive resource (e.g., visual perception; focal
From Taatgen (2011). With permission of the author.
working memory) can be used by only one
process at any given time. Nijboer et al. (2013)
tested this assumption using multi-column subtraction as the primary task
with participants responding using a keypad. Easy and hard conditions dif-
fered in whether digits were carried over (“borrowed”) from one column to
the next:

(1: easy) (2: hard)

336789495  3649772514
–224578381 –1852983463

The model predicts focal working memory is required only in the hard con-
dition. Subtraction was combined with a secondary task: a tracking task
involving visual and manual resources or a tone-counting task involving
working memory.
Nijboer et al. (2013) predicted performance on the easy subtraction
task should be worse when combined with the tracking task because both
compete for visual and manual resources. In contrast, performance on
the hard subtraction task should be worse when combined with the tone-­
counting task because there are large disruptive effects when two tasks
compete for working memory resources. The findings were as predicted.
Borst et al. (2013) found there was far less impairment of hard sub-
traction performance by a secondary task requiring working memory when
 Attention and performance 219

participants saw a visual sign explicitly indicating that “borrowing” was

needed. This supports the model’s assumption that dual-task performance
can be enhanced by appropriate environmental support.
According to the threaded cognition model, we often cope with the
demands of combining two tasks by switching flexibly between them to
maximise performance. Support was reported by Farmer et al. (2018).
Participants performed a typing task and a tracking task at the same time.
The relative value of the two tasks was varied by manipulating the number
of points lost for poor tracking performance. Participants rapidly learned
to adjust their strategies over time to increase the overall number of points
they gained.
Katidioti and Taatgen (2014) found task switching is not always
optimal. Participants performed two tasks together: (1) an email task in
which information needed to be looked up; (2) chat messages containing
questions to be answered. When there was a delay on the email task, most
participants switched to the chat task. This happened even when this was
suboptimal because it caused participants to forget information in the
email task.
How can we explain the above findings? According to Katidioti and
Taatgen (2014, p. 734), “The results . . . agree with threaded cognition’s
‘greedy’ theory . . . which states that people will switch to a task that is
waiting as soon as the resources for it are available.” Huijser et al. (2018)
obtained further evidence of “greediness”. When there were brief blank
periods during the performance of a cognitively demanding task, partic-
ipants often had task-irrelevant thoughts (e.g., mind-wandering) even
though these thoughts impaired task performance.
Katidioti and Taatgen (2014) also discovered substantial individual
differences in task switching – some participants never switched to the chat
task when delays occurred on the email task. Such individual differences
cannot be explained by the theory.
As mentioned earlier, a recent version of threaded cognition theory
discussed by Nijboer et al. (2016a) identifies three components of working
(i.e., problem state or focus of attention; declarative memory or activated
short-term memory; and subvocal rehearsal). Nijboer et al. had partici-
pants perform two working memory tasks at the same time; these tasks
varied in the extent to which they required the same working memory com-
ponents. They obtained measures of performance and also used neuroim-
aging under dual-task and single-task conditions.
What did Nijboer et al. (2016a) find? First, dual-task interference
could be predicted from the extent to which the two tasks involved the
same working memory components. Second, dual-task interference could
also be predicted from the extent of overlap in brain activation of the two
tasks in single-task conditions. In sum, dual-task interference depended
on competition for specific resources (i.e., working memory components)
rather than general resources (e.g., central executive).

Evaluation
The model has proved successful in various ways. First, several important
cognitive resources have been identified. Second, the model identifies brain

9781138482210_COGNITIVE_PSYCHOLOGY_PART_1.indd 219 28/02/20 6:44 PM

220 Visual perception and attention

areas associated with various cognitive resources. This has led to compu-
tational modelling testing the model’s predictions using neuroimaging and
behavioural findings. Thus, the model accounts for dual-task performance
without assuming the existence of a central executive or other executive
process (often vaguely defined in other theories). Fourth, the theory predicts
factors determining switching between two tasks being performed together.
Fifth, individuals often have fewer problems performing two simultaneous
tasks than generally assumed.
What are the model’s limitations? First, it predicts that “Practising two
tasks concurrently [together] results in the same performance as perform-
ing the two tasks independently” (Salvucci & Taatgen, 2008, p. 127). This
de-emphasises the importance of processes coordinating and managing two
tasks performed together (see next section). Second, excluding processes
resembling Baddeley’s central executive is controversial and may well
prove inadvisable. Third, most tests of the model have involved the simul-
taneous performance of two relatively simple tasks and its applicability to
more complex tasks remains unclear. Fourth, the theory does not provide
a full explanation for individual differences in the extent of task switching
(e.g., Katidioti & Taatgen, 2014).

Cognitive neuroscience
The cognitive neuroscience approach is increasingly used to test theoretical
models and enhance our understanding of processes underlying dual-task
performance. Its value is that neuroimaging provides “an additional data
source for contrasting between alternative models” (Palmeri et al., 2017,
p. 61). More generally, behavioural findings indicate the extent to which
dual-task conditions impair task performance but are often relatively unin-
formative about the precise reasons for such impairment.
Suppose we compare patterns of brain activation while participants
perform tasks x and y singly or together. Three basic patterns are shown
in Figure 5.18:

(1) Underadditive activation: reduced activation in one or more brain

areas in the dual-task condition occurs because of resource competi-
tion between the tasks.

Figure 5.18 (a) Underadditive activation (b) Additive activation (c) Overadditive activation
(a) Underadditive activation; Component task 1 Component task 1 Component task 1
(b) additive activation;
(c) overadditive activation.
White indicates task 1 Time Time Time
Component task 2 Component task 2 Component task 2
activation; grey indicates
task 2 activation; and
black indicates activation Time Time Time
only present in dual-task Dual-task Dual-task Dual-task

conditions.
From Nijboer et al., 2014.
Time Time Time
Reprinted with permission of
Elsevier.
 Attention and performance 221

(2) Additive activation: brain activation in the dual-task condition is KEY TERM
simply the sum of the two single-task activations because access to
Underadditivity
resources is integrated efficiently between the two tasks. The finding that brain
(3) Overadditive activation: brain activation in one or more brain areas is activation when tasks
present in the dual-task condition but not the single-task conditions. A and B are performed
This occurs when dual-task conditions require executive processes that at the same time is less
than the sum of the brain
are absent (or less important) with single tasks. These executive pro-
activation when tasks
cesses include the coordination of task demands, attentional control A and B are performed
and dual-task management generally. We would expect such executive separately.
processes to be associated mostly with activation in prefrontal cortex.

Findings
We start with an example of underadditive activation. Just et al. (2001)
used two very different tasks (auditory sentence comprehension and mental
rotation of 3-D figures) performed together or singly. Performance on both
tasks was impaired under dual-task conditions compared to single-task
conditions. Under dual-task conditions, brain activation in language pro-
cessing areas decreased by 53% and reduced by 29% in areas associated
with mental rotation. These findings suggest fewer task-relevant processing
resources were available when both tasks were performed together.
Schweizer et al. (2013) also reported underadditivity. Participants
performed a driving task on its own or with a distracting secondary task
(answering spoken questions). Driving performance was unaffected by
the secondary task. However, driving with distraction reduced activa-
tion in posterior brain areas associated with spatial and visual processing
(underadditivity). It also produced increased activation in the prefrontal
cortex (overadditivity; see Figure 5.19) probably because driving with dis-
traction requires increased attentional or cognitive control within the pre-
frontal cortex.
Figure 5.19
Dual-task performance is often associated with overadditivity due to Effects of an audio
increased activity within the prefrontal cortex (especially the lateral pre- distraction task on brain
frontal cortex) during dual-task performance (see Strobach et al., 2018, for activity associated with a
a review). However, most such findings do not show that this increased straight driving task. There
prefrontal activation is actually required for dual-task performance. were significant increases
More direct evidence that prefrontal areas associated with attentional in activation within the
ventrolateral prefrontal
or cognitive control are causally involved in enhancing dual-task perfor-
cortex and the auditory
mance was reported by Filmer et al. (2017) and Strobach et al. (2018). cortex (in orange). There
Filmer et al. (2017) studied the effects of transcranial direct current stim- was decreased activation
ulation (tDCS; see Glossary) applied to areas of the prefrontal cortex in occipital-visual areas (in
associated with cognitive control. Anodal tDCS during training enhanced blue).
cognitive control and subsequent dual-task performance. From Schweizer et al. (2013).
 222 Visual perception and attention

Strobach et al. (2018) reported similar findings. Anodal tDCS applied

to the lateral prefrontal cortex led to enhanced dual-task performance.
In another condition, cathodal tDCS to the same area of the prefrontal
cortex impaired dual-task performance. These findings were as predicted
given that anodal and cathodal tDCS often have opposite effects on per-
formance. These findings indicate that the lateral prefrontal cortex causally
influences dual-task performance.
Additional evidence of the importance of the lateral prefrontal cortex
was reported by Wen et al. (2018). Individuals with high connectivity (con-
nectedness) within that brain area showed superior dual-task performance
to those with low connectivity.
Finally, patterns of brain activation can help to explain practice effects
on dual-task performance. Garner and Dux (2015) found much fronto-­
parietal activation (associated with cognitive control) when two tasks were
performed singly or together. Extensive training greatly reduced dual-task
interference and also produced increasing differentiation in the pattern of
fronto-parietal activation associated with the two tasks. Participants showing
the greatest reduction in dual-task interference tended to have the greatest
increase in differentiation. Thus, using practice to increase differences in pro-
cessing and associated brain processing between tasks can be very effective.

Evaluation
Brain activity in dual-task conditions often differs from the sum of brain
activity of the same two tasks performed singly. Dual-task activity can
exhibit underadditivity or overadditivity. The findings are theoretically
important because they indicate performance of dual tasks can involve
much more cognitive control and other processes than single tasks. Garner
and Dux’s (2015) findings demonstrate that enhanced dual-task perfor-
mance with practice can depend on increased differentiation between the
two tasks with respect to processing and brain activation.
What are the limitations of the cognitive neuroscience approach?
First, increased (or decreased) activity in a given brain area in dual-task
conditions is not necessarily very informative. For example, Dux et al.
(2009) found dual-task performance improved over time because prac-
tice increased the speed of information processing in the prefrontal cortex
rather than because it changed activation within that region. Second, it is
often unclear whether patterns of brain activation are directly relevant to
task processing rather than reflecting non-task processing.
Third, findings in this area are rather inconsistent (Strobach et al., 2018)
and we lack a comprehensive theory to account for these inconsistencies.
Plausible reasons for these apparent inconsistencies are the great variety of
task combinations used in dual-task studies and individual differences in
task proficiency among participants (Watanabe & Funahashi, 2018).

Psychological refractory period: cognitive bottleneck?

Much of the research discussed so far was limited because the task combi-
nations used made it hard to assess in detail the processes used by partic-
ipants. For example, the data collected were often insufficient to indicate

9781138482210_COGNITIVE_PSYCHOLOGY_PART_1.indd 222 28/02/20 6:44 PM

 Attention and performance 223

the frequency with which participants switched their attentional focus from KEY TERMS
one task to the other. This led researchers to use much simpler tasks so that
Psychological refractory
they had “better experimental control over the timing of the component period (PRP) effect
task processes” (Koch et al., 2018, p. 575). The slowing of the
The dominant paradigm in recent research is as follows. There are two response to the second
stimuli (e.g., two lights) and two responses (e.g., button presses), one asso- of two stimuli when
presented close together
ciated with each stimulus. Participants respond to each stimulus as rapidly
in time.
as possible. When the two stimuli are presented at the same time (dual-task
condition), performance is typically worse on both tasks than when each Stimulus onset
­asynchrony (SOA)
task is presented separately (single-task conditions).
Time interval between the
When the second stimulus is presented shortly after the first, there start of two stimuli.
is typically a marked slowing of the response to the second stimulus:
the ­psychological refractory period (PRP) effect. This effect is robust –
Ruthruff et al. (2009) obtained a large PRP effect even when participants
were given strong incentives to eliminate it.
The PRP effect has direct real-world relevance. Hibberd et al. (2013)
studied the effects of a simple in-vehicle task on braking performance when
the vehicle in front braked and slowed down. There was a classic PRP
effect – braking time was slowest when the in-vehicle task was presented
immediately before the vehicle in front braked.
How can we explain the PRP effect? It is often argued task perfor-
mance involves three successive stages: (1) perceptual; (2) central response
selection; and (3) response execution. According to the bottleneck model
(e.g., Pashler, 1994),

The response selection stage of the second task cannot begin until the
response selection stage of the first task has finished, although the
other stages . . . can proceed in parallel . . . according to this model,
the PRP effect is a consequence of the waiting time of the second task
because of a bottleneck at the response selection stage.
(Mittelstädt & Miller, 2017, p. 89)

The bottleneck model explains several findings. For example, consider the
effects of varying the time between the start of the first and second stimuli
(stimulus onset asynchrony (SOA)). According to the model, processing on
the first task should slow down second-task processing much more when
the SOA is small than when it is larger. The predicted finding is generally
obtained (Mittelstädt & Miller, 2017).
The bottleneck model remains the most influential explanation of the
PRP effect (and other dual-task costs). However, resource models (e.g.,
Navon & Miller, 2002) are also influential. According to these models,
limited processing capacity can be shared between two tasks so both are
processed simultaneously. Of crucial importance, sharing is possible even
during the response selection process. A consequence of sharing processing
capacity across task is that each task is processed more slowly than if per-
formed on its own.
Many findings can be explained by both models. However, resource
models are more flexible than bottleneck models. Why is this? Resource
models assume the division of processing resources between two tasks
varies freely to promote efficient performance.

9781138482210_COGNITIVE_PSYCHOLOGY_PART_1.indd 223 28/02/20 6:44 PM

224 Visual perception and attention

KEY TERM Another factor influencing the PRP effect is crosstalk (the two tasks
interfere directly with each other). This mostly occurs when the stimuli
Crosstalk
In dual-task conditions, and/or responses on the two tasks are similar. A classic example of cross-
the direct interference talk is when you try to rub your stomach in circles with one hand while
between the tasks that is patting your head with the other hand (try it!).
sometimes found. Finally, note that participants in most studies receive only modest
amounts of practice in performing two tasks at the same time. As a con-
sequence, the PRP effect may occur at least in part because participants
receive insufficient practice to eliminate it.

Findings
According to the bottleneck model, we would expect to find a PRP effect
even when easy tasks are used and/or participants receive prolonged prac-
tice. Contrary evidence was reported by Schumacher et al. (2001). They
used two tasks: (1) say “one”, “two” or “three” to low-, medium- and
high-pitched tones, respectively; (2) press response keys corresponding to
the position of a disc on a computer screen. These tasks were performed
together for over 2,000 trials, by which time some participants performed
them as well together as singly.
Strobach et al. (2013) conducted a study very similar to that of
Schumacher et al. (2001). Participants took part in over 5,000 trials involv-
ing single-task or dual-task conditions. However, dual-task costs were not
eliminated after extensive practice: dual-task costs on the auditory task
reduced from 185 to 60 ms and those on the visual task from 83 to 20 ms
(see Figure 5.20). How did dual-task practice benefit performance? Practice
speeded up the central response selection stage in both tasks.
Why did the findings differ in the two studies discussed above? In both
studies, participants were rewarded for fast responding on single-task and
dual-task trials. However, the way the reward system was set up in the
Schumacher et al. study may have led participants to exert more effort in
dual-task than single-task trials. This potential bias was absent from the

Figure 5.20
Reaction times for correct
responses only over eight
experimental sessions
under dual-task (auditory
and visual tasks) and single-
task (auditory or visual task)
conditions.
From Strobach et al. (2013).
Reprinted with permission of
Springer.
 Attention and performance 225

Strobach et al. study. This difference in reward structure could explain the KEY TERM
greater dual-task costs in the Strobach et al. study.
Backward crosstalk
Hesselmann et al. (2011) studied the PRP effect using event-related effect
potentials. The slowing of responses on the second task was closely Aspects of Task 2
matched by slowing in the onset of the P300 (an ERP component reflecting influence response
response selection). However, there was no slowing of earlier ERP compo- selection and
performance speed on
nents reflecting perceptual processing. Thus, as predicted by the bottleneck
Task 1 in studies on the
model, the PRP effect depended on response selection rather than percep- psychological refractory
tual processes. period (PRP) effect.
According to the resource model approach, individuals choose whether
to use serial or parallel processing on PRP tasks. Miller et al. (2009) argued
that serial processing generally leads to superior performance compared
with parallel processing. However, parallel processing should theoretically
be superior when the stimuli associated with the two tasks are mostly pre-
sented close in time. As predicted, there was a shift from predominantly
serial processing towards parallel processing when that was the case.
Miller et al. (2009) used very simple tasks and it is likely parallel pro-
cessing is most likely to be used with such tasks. Han and Marois (2013)
used two tasks, one of which was relatively difficult. Participants used
serial processing even when parallel processing was encouraged by finan-
cial rewards.
Finally, we consider the theoretically important backward crosstalk
effect: “characteristics of Task 2 of 2 subsequently performed tasks influ-
ence Task 1 performance” (Janczyk et al., 2018, p. 261). Hommel (1998)
obtained this effect. Participants responded to Task 1 by making a left or
right key-press and to Task 2 by saying “left” or “right”. Task 1 responses
were faster when the two responses were compatible (e.g., press right key
+ say “right”) than when they were incompatible (e.g., press right key +
say “left”). Evidence for the backward crosstalk effect was also reported by
Janczyk et al. (2018).
Why is the backward crosstalk effect theoretically important? It indi-
cates that aspects of response selection processing on Task 2 occur before
response selection processing on Task 1 has finished. This effect is incom-
patible with the bottleneck model, which assumes response selection on
Task 1 is completed prior to any response selection on Task 2. In other
words, this model assumes there is serial processing at the response selec-
tion stage. In contrast, the backward crosstalk effect is compatible with the
resource model approach.

Summary and conclusions

The findings from most research on the psychological refractory period
effect are consistent with the bottleneck model. As predicted, this effect is
typically larger when the second task follows very soon after the first task.
In addition, even prolonged practice rarely eliminates the psychological
refractory period effect suggesting that central response selection processes
typically occur serially.
The bottleneck model assumes processing is less flexible than is often
the case. For example, the existence of the backward crosstalk effect is
inconsistent with the bottleneck model but consistent with the resource

9781138482210_COGNITIVE_PSYCHOLOGY_PART_1.indd 225 28/02/20 6:44 PM

226 Visual perception and attention

model approach. Fischer et al. (2018) also found evidence for much flex-
ibility. There was less interference between the two tasks when finan-
cial rewards were offered because participants devoted more processing
resources to protecting the first task from interference. However, the
resource model approach has the disadvantage compared to the bottleneck
model that its predictions are less precise, making it harder to submit to
detailed empirical testing.
Finally, as Koch et al. (2017, p. 575) pointed out, the bottleneck
model “can be applied (with huge success) mainly for conditions in which
two tasks are performed strictly sequentially”. This is often the case with
research on the psychological refractory period effect but is much less
applicable to more complex dual-task situations.

“AUTOMATIC” PROCESSING
We have seen in studies of divided attention that practice often causes a dra-
matic improvement in performance. This improvement has been explained
by assuming some processes become automatic through prolonged prac-
tice. For example, the huge amount of practice we have had with reading
words has led to the assumption that familiar words are read “automati-
cally”. Below we consider various definitions of “automaticity”. We also
consider different approaches to explaining the development of automatic
processing.

Traditional approach: Shiffrin and Schneider (1977)

Shiffrin and Schneider (1977) and Schneider and Shiffrin (1977) distin-
guished between controlled and automatic processes:

●● Controlled processes are of limited capacity, require attention and can

be used flexibly in changing circumstances.
●● Automatic processes suffer no capacity limitations, do not require
attention and are very hard to modify once learned.

In Schneider and Shiffrin’s (1977) research, participants memorised letters

(the memory set) followed by a visual display containing letters. They
then decided rapidly whether any item in the visual display was the same
as any item in the memory set. The crucial manipulation was the type of
mapping. With consistent mapping, only consonants were used as members
of the memory set and only numbers were used as distractors in the visual
display (or vice versa). Thus, a participant given only consonants to mem-
orise would know any consonant detected in the visual display was in the
memory set. With varied mapping, numbers and consonants were both used
to form the memory set and to provide distractors in the visual display.
The mapping manipulation had dramatic effects (see Figure 5.21). The
numbers of items in the memory set and visual display greatly affected
decision speed only with varied mapping. According to Schneider and
Shiffrin (1977), varied mapping involved serial comparisons between each
item in the memory set and each item in the visual display until a match
was achieved or every comparison had been made. In contrast, consistent
 Attention and performance 227

Figure 5.21
Response times on a
decision task as a function
of memory-set size, display-
set size and consistent vs
varied mapping.
Response times on a
decision task as a function
of memory-set size, display-
set size and consistent
vs varied mapping. Data
from Shiffrin and Schneider
(1977).
American Psychological
Association.

mapping involved automatic processes operating independently and in par-

allel. These automatic processes have evolved through prolonged practice
in distinguishing between letters and numbers.
In a second experiment, Shiffrin and Schneider (1977) used consistent
mapping with the consonants B to L forming one set and Q to Z the other
set. As before, items from only one set always formed the memory set
with all the distractors in the visual display being selected from the other
set. Performance improved greatly over 2,100 trials, reflecting increased
automaticity. After that, there were 2,100 trials with the reverse consist-
ent mapping (swapping over the memory and visual display sets). With
this reversal, it took nearly 1,000 trials before performance recovered to its
level at the start of the experiment!
Evidence that there may be limited (or no conscious awareness in
the consistent mapping condition was reported by Jansma et al. (2001)).
Increasing automaticity (indexed by increased performance speed) was
accompanied by reduced activation in areas associated with conscious
awareness (e.g., dorsolateral prefrontal cortex).
In sum, automatic processes function rapidly and in parallel but are
inflexible (second part of the second experiment). Controlled processes are
flexible and versatile but operate relatively slowly and in a serial fashion.

Limitations
What are the limitations with this approach? First, the distinction between
automatic and controlled processes is oversimplified (discussed below).
Second, Shiffrin and Schneider (1977) argued automatic processes operate
in parallel and place no demands on attentional capacity and so decision
speed should be unrelated to the number of items. However, decision
228 Visual perception and attention

speed was slower when the memory set and visual display both contained
several items (see Figure 5.21). Third, the theory is descriptive rather than
­explanatory – it does not explain how serial controlled processing turns into
parallel automatic processing.

Definitions of automaticity
Shiffrin and Schneider (1977) assumed there is a clear-cut distinction
between automatic and controlled processes. More specifically, automatic
processes possess several features (e.g., inflexibility; very efficient because
they have no capacity limitations; occurring in the absence of attention). In
essence, it is assumed there is perfect coherence or consistency among the
features (i.e., they are all found together).
Moors and De Houwer (2006) and Moors (2016) identified four key
features associated with automaticity:

(1) unconscious: lack of conscious awareness of at least one of the fol-

lowing: “the input, the output, and the transition from one to the
other” (Moors, 2016, p. 265);
(2) efficient: using very little attentional capacity;
(3) fast;
(4) goal-unrelated or goal-uncontrolled: at least one of the following is
missing: “the goal is absent, the desired state does not occur, or the
causal relation [between the goal and the occurrence of the desired
state] is absent” (Moors, 2016, p. 265).

Why might these four features (or the similar ones identified by Shiffrin
and Schneider (1977)) often be found together? Instance theory (Logan,
1988; Logan et al., 1999) provides an influential answer. It is assumed
task practice leads to storage of information in long-term memory which
facilitates subsequent performance on that task. In essence, “Automaticity
is memory retrieval: performance is automatic when it is based on a
­single-step direct-access retrieval of past solutions from memory” (Logan,
1988, p. 493). For example, if you were given the problem “24 × 7 = ???”
numerous times, you would retrieve the answer (168) “automatically”
without performing any mathematical calculations.
Instance theory makes coherent sense of several characteristics of
automaticity. Automatic processes are fast because they require only the
retrieval of past solutions from long-term memory. They make few demands
on attentional resources because the retrieval of heavily o ­ver-learned
­information is relatively effortless. Finally, there is no conscious awareness
of automatic processes because no significant processes intervene between
stimulus presentation and retrieval of the correct response.
In spite of its strengths, instance theory is limited (see Moors, 2016).
First, the theory implies the key features of automaticity will typically all
be found together. However, this is not the case (see below). Second, it
is assumed practice leads to automatic retrieval of solutions with learners
having no control over such retrieval. However, Wilkins and Rawson (2011)
found evidence learners can exercise top-down control over retrieval: when
the instructions emphasised accuracy, there was less evidence of retrieval
 Attention and performance 229

than when they emphasised speed. Thus, the use of retrieval after practice
is not fully automatic.
Melnikoff and Bargh (2018) argued that the central problem with the
traditional approach is that no research has shown the four features asso-
ciated with “automaticity” occurring together. As they pointed out, “No
attempt has been made to estimate the probability of a process being inten-
tional given that is conscious versus unconscious, or the probability of a
process being controllable given that it is efficient versus inefficient, and so
forth” (p. 282).

Decompositional approach: Moors (2016)

Moors and De Houwer (2006) and Moors (2016) argued that previous
theoretical approaches are greatly oversimplified. Instead, they favoured
a decompositional approach. According to this approach, the features
of automaticity are clearly separable and are by no means always found
together: “It is dangerous to draw inferences about the presence or absence
of one feature on the basis of the presence or absence of another” (Moors &
de Houwer, 2006, p. 320).
Moors and De Houwer (2006) also argued there is no firm dividing line
between automaticity and non-automaticity. The features are continuous
rather than all-or-none (e.g., a process can be fairly fast or slow; it can be
partially conscious). As a result, most processes involve a blend of automa-
ticity and non-automaticity. This approach is rather imprecise because few
processes are 100% automatic or non-automatic. However, we can make
relative statements (e.g., process X is more/less automatic than process Y).
Moors (2016) claimed the relationships between factors such as goals,
attention and consciousness are much more complex than claimed within
traditional approaches to “automaticity”. This led her to develop a new the-
oretical account (see Figure 5.22). A key assumption is that all information

Prior stimulus factors Prior stimulus representation factors

• Frequency • Existence of stimulus representation in LTM
• Recency • Strength of trace to stimulus representation in LTM
• Stimulus quality: duration, intensity ~ Availability of stimulus representation in LTM
• Quality of stimulus representation in WM
Prior stimulus × person factors
• Selection history
• Reward history
Conscious
Attention processing

2nd threshold
Current stimulus representation factors Unconscious
Current stimulus factors Attention
• Quality of stimulus representation: duration, processing
• Stimulus quality: duration, intensity
• Un/expectedness intensity, distinctiveness
• Goal in/congruence ~ Accessibility of stimulus representation for 1st threshold
• Novelty/familiarity processing

Figure 5.22
Factors that are hypothesised to influence representational quality within Moors’ (2016) theoretical approach.
From Moors (2016).
 230 Visual perception and attention

processes require an input of sufficient representational quality (defined by

the “intensity, duration, and distinctiveness of a representation”, Moors,
2016, p. 273).
What factors determine representational quality?

(1) current stimulus factors, including the extent to which a stimulus is

expected or unexpected, familiar or novel, and goal congruent or
incongruent;
(2) prior stimulus factors (e.g., the frequency and recency with which the
current stimulus has been encountered);
(3) prior stimulus representation factors based on relevant information
stored within long-term memory;
(4) attention, which enhances or amplifies the impact of current stimulus
factors and prior stimulus representation factors on the current stim-
ulus representation.

According to this theoretical account, the above factors influence repre­

sentational quality additively so that a high level of one factor can compen-
sate for a low level of another factor. For example, selective attention or
relevant information in long-term memory can compensate for brief stim-
ulus presentations. The main impact of consciousness occurs later than
for other factors (e.g., attention and goal congruence). More specifically,
representational quality must reach the first threshold to permit uncon-
scious processing but a more stringent second threshold to permit conscious
processing.

Findings
According to Moors’ (2016) theoretical framework, there is a flexible rela-
tionship between controlled and conscious processing. This contrasts with
Schneider and Shiffrin’s (1977) assumption that executive control is always
associated with conscious processing.
Diao et al. (2016) reported findings consistent with Moors’ prediction.
They used a Go/No-Go task where participants made a simple response
(Go trials) or withheld it (No-Go trials). High-value or low-value financial
rewards were available for successful performance. Task stimuli were pre-
sented above or below the level of conscious awareness.
What did Diao et al. (2016) find? Performance was better on high-­
reward than low-reward trials even when task processing was uncon-
scious. In addition, participants showed superior unconscious inhibitory
control (assessed by event-related potentials) on high-reward trials. Thus,
one feature of automaticity (unconscious processing) was present whereas
another feature (goal-uncontrolled) was not.
Huber-Huber and Ansorge (2018) also reported problems for the tra-
ditional approach. Participants received target words indicating an upward
or downward direction (e.g., above; below). Prior to the target word, a
prime word also indicating an upward or downward direction was pre-
sented below the level of conscious awareness. Response times to the target
words were slower when there was a conflict between the meanings of the
prime and target words than when they were congruent in meaning. As in

9781138482210_COGNITIVE_PSYCHOLOGY_PART_1.indd 230 28/02/20 6:44 PM

 Attention and performance 231

the study by Diao et al. (2016), unconscious processing was combined with
control, a combination that is inconsistent with the traditional approach.

Evaluation
The theoretical approach to automaticity proposed by Moors (2016) has
several strengths. First, the assumption that various features associated with
automaticity often correlate poorly with each other is clearly superior to
the earlier notion that these features exhibit perfect coherence. Second, her
assumption that processes vary in the extent to which they are “automatic”
is much more realistic than the simplistic division of processes into auto-
matic and non-automatic. Third, the approach is more comprehensive than
previous ones because it considers more factors relevant to “automaticity”.
What are the limitations with Moors’ (2016) approach? First, numer-
ous factors are assumed to influence representational quality (and thus the
extent to which processes are automatic) (see Figure 5.22). It would thus
require large-scale experimental research to assess the ways all these factors
interact. Second, the approach provides only a partial explanation of the Interactive exercise:
underlying mechanisms causing the various factors to influence representa- Definitions of attention
tional quality.

CHAPTER SUMMARY
• Focused auditory attention. When two auditory messages
are presented at the same time, there is less processing of the
unattended than the attended message. Nevertheless, unattended
messages often receive some semantic processing. The restricted
processing of unattended messages may reflect a bottleneck at
various stages of processing. However, theories assuming the
existence of a bottleneck de-emphasise the flexibility of selective
auditory attention. Attending to one voice among several (the
cocktail party problem) is a challenging task. Human listeners use
top-down and bottom-up processes to select one voice. Top-
down processes include the use of various control processes
(e.g., focused attention; inhibitory processes) and learning about
structural consistencies present in the to-be-attended voice.

• Focused visual attention. Visual attention can resemble a

spotlight or zoom lens. In addition, the phenomenon of split
attention suggests visual attention can also resemble multiple
spotlights. However, accounts based on spotlights or a zoom
lens typically fail to specify the underlying mechanisms. Visual
attention can be object-based, space-based or feature-based, and
it is often object-based and space-based at the same time. Visual
attention is flexible and is influenced by factors such as individual
differences.
According to Lavie’s load theory, we are more susceptible
to distraction when our current task involves low perceptual load
and/or high cognitive load. There is much support for this theory.