Content uploaded by Mark Chignell
Author content
All content in this area was uploaded by Mark Chignell on May 30, 2016
Content may be subject to copyright.
The Impact of Central Executive Function Loadings on
Driving-Related Performance
Sachi Mizobuchi1, Mark Chignell1, Junko Suzuki2, Ko Koga2, Kazunari Nawa2
1University of Toronto/ Vocalage Inc.
5 King’s College Rd., Toronto
Ontario, M5S 3G8, Canada
sachi.mizobuchi@utoronto.ca,
chignell@mie.utoronto.ca
2Toyota InfoTechnology Center Co., Ltd.
6-6-20 Akasaka, Minato-ku
Tokyo 107-0052, Japan
{ju-suzuki, ko-koga, nawa}@jp.toyota-itc.com
ABSTRACT
The study reported in this paper investigated the impact of
individual ability, with respect to central executive (CE) functions,
on performance of two driving-related tasks when distracted by
CE loading secondary tasks. The two driving-related tasks used
were visual target detection, and a one-dimensional pedal-tracking
task designed to be an analogue of a vehicle following task. The
three CE tasks were each designed to load mainly on just one of
three different CE functions (inhibition, shifting, and updating,
respectively) in both audio and visual conditions. An additional
single key press secondary task was used to assess the impact of a
non-CE loading secondary task. We hypothesized that people with
a higher level of ability for a given CE function would do better,
relative to those with lower ability, on the driving-related task
when it was accompanied by a secondary task that loaded the
corresponding CE function. 102 people participated in the
screening portion of the study, and 34 of these participants were
then selected to participate in the main experiment. We found that
the impact of CE abilities on dual task performance is more
complex than a simple tradeoff model would predict. Shifting
ability generally improved primary task performance in the dual
tasks, and inhibition ability tended to improve performance in the
target detection task, while updating ability tended to improve
performance in the pedal-tracking task.
Keywords
Driver distraction, Mental/cognitive workload, Central Executive,
Individual differences, Inhibition, Shifting, Updating, In-Vehicle
information device, User interface
1. INTRODUCTION
The long-term goal of this research is to identify interaction
design requirements for minimizing the distracting effect of in-
vehicle information systems on drivers. We focus in particular on
the distraction and workload caused by tasks that place a load on
the central executive (CE) functions. CE is one of the main
components in the dominant model of working memory (e.g.,[1]).
Extensive research has associated CE function activity with the
prefrontal cortical region of the brain (e.g.,[16]).
Driver distraction caused by interaction with in-car information
systems involves multi-tasking situations that likely comprise
several types of workload, including perceptual (visual and audio),
manual and cognitive workload involving central executive
functions [18]. In the research reported here we focused on
cognitive workload and investigated its effects on driving-related
performance. We were particularly interested in using individual
differences in cognitive ability (CE functions) as a way to identify
when higher levels of CE function ability are needed to maintain
adequate driving performance in the presence of distracting
secondary tasks.
2. RELATED RESEARCH
2.1 Fractionated CE functions and Individual
Differences
There has been considerable discussion around the issue of
whether CE functioning should be understood as a unified system
or as a fractionated system. The fractionated system view has
mainly been supported by studies on individual difference in
cognitive ability, which have been conducted with a variety of
populations, such as normal young adults [17][9], normal elderly
adults [13], brain-damaged adults [3], and children with
neurocognitive pathologies [11]. These studies typically employed
a battery of widely used executive tasks like the Wisconsin card
sort test (WCST) and the n-back test, and examined how well
these tasks correlated with one another by performing
correlation/regression analysis and exploratory factor analysis
(EFA). Many of these studies have shown low (not statistically
significant) inter-correlations among different executive tasks,
consistent with the fractionated system view.
Observations from neuropsychology have also supported the
fractionated view of CE functioning. For example, Logie et al.
examined the basis for a multiplicity of CE functions, showing
that the function for multitasking could be selectively impaired in
Alzheimer’s disease (AD) patients group [12].
In the fractionated CE view, a variety of ways to classify
executive functions have been proposed such as "mental set
shifting", "inhibition", "flexibility", "updating", “monitoring”,
"planning", and "dual-tasking". In this research, we decided to
start our exploratory study from the following three functions:
"inhibition", "shifting", and "updating" based on Miyake et al’s
characterization [17], since the three functions, or analogous ones,
are often seen in other classification systems (e.g., [19]). In
addition, each of these three functions have been associated with
tasks that can be used to measure the level of ability that a person
has with respect to that function.
Some researchers have argued that these three functions differ in
their degree of independence. For instance, Szmalec et al. argued
that updating ability as measured by the n-back task includes an
aspect of conflict solving that is related to inhibition [25]. While it
is probably difficult, if not impossible, to develop “pure” tasks
that load on only one CE function (since tasks will generally have
Copyright held by author(s)
AutomotiveUI'12, October 17-19, Portsmouth, NH, USA.
Adjunct Proceedings
shared perceptual, selective attention, and response selection
components), it should still be possible to assess the impact of CE
functions using “impure” tasks that tend to have high loadings on
one CE function relative to the others.
2.2 Multiple resource model and multitasking
performance
Wickens proposed a multiple resource model to describe cognitive
(mental) workload [27]. In the multiple resource model there
separable attentional resources (for example, visual and auditory
in modalities, spatial and verbal in codes). In this research, we are
interested in whether or not the multiple resource model should be
extended to include the impact of different CE functions, and we
are also interested in the impact of individual differences on those
CE functions.
To measure cognitive workload, a dual-task procedure is often
used. In the literature review on the use of secondary tasks in the
assessment of workload, Ogden et al. found that there is no single
best task or class of tasks for the measurement of workload [20]
(also see [6], for a collection of chapters on the different
approaches to measuring mental workload). Given the strong
evidence for a multiplicity of CE functions, it is natural to ask
what role, if any, they should have in models of mental workload
and cognitive distraction. This question was the motivation for the
research reported below.
2.3 CE functions and driving performance
While many researchers have tried to measure the levels of driver
distraction caused by different secondary tasks (eg. [7][15][24]),
relatively little research has focused on understanding the types of
cognitive workload and their effects.
However, Baumann et al. investigated the effect of CE load on
driving performance (as assessed by time to collision and driving
speed) [2]. They used simulated driving where participants were
required to avoid obstacles while performing either an auditory
monitoring task that should not load on the comprehension
functions of the CE, or a running memory task that should heavily
load on the CE involving comprehension and prediction function
of situation awareness. They found that participants received less
benefit from being provided with a warning signal when they had
to perform the running memory task. The researchers concluded
that the CE function is strongly involved in the construction of
situation awareness.
Mäntylä et al. also examined the relationship between CE function
and driving performance [14]. In their experiment, high school
students completed a simulated driving task and six experimental
tasks that tapped the three CE functions of inhibition, shifting, and
updating. Their results showed that updating ability was a
significant predictor of performance on a Lane Change Task
(LCT) while doing simulated driving.
3. OUR RESEARCH INTEREST AND
HYPOTHESIS
In the present study, we assumed that workload experienced by
individuals is defined as the interaction between individual
differences and task requirements. Figure 1 represents our
approach, where workload is attributable to task loadings on the
three CE functions of inhibition, shifting, and updating. The
model makes the following three assumptions:
- Individual differences exist in the capacity of each CE function;
- Different tasks load the CE to varying extents as a function of
both the nature of the task itself and the unique effect of the task
on each individual.
- The cognitive workload that a person experiences while
performing a task is determined by the direction and degree of
mismatch between the person’s abilities and the task requirements
with respect to the CE functions.
If validated, this model provides for the profiling of tasks by
measuring the CE ability of individuals and the subsequent
cognitive workload they experience while performing those tasks.
Tasks with unacceptably high loads on particular CE functions
could then be identified and redesigned so as to reduce those loads
to acceptable levels. Figure 1 summarizes this approach whereby
cognitive demands by a task (pink bars on the left) combine with
individual cognitive ability (green bars in the middle) and result in
workload predictions across each of the three CE functions (blue
circles on the right). For instance, participant 1 has low shifting
ability and the task requires high shifting ability, meaning
Participant 1’s shifting workload is high for that task.
Figure 1. A model of cognitive workload based on the
individual differences and different task requirements
In a driving task it seems that all three CE functions (inhibition,
shifting, and updating) are required. Updating is likely required
for keeping track of the position of one’s vehicle relative to other
vehicles in the road, and of keeping track of one’s current location
for navigation purposes. Inhibition would seem to be required to
detect and respond to external events such as changing traffic
signals or people who move into the path of one’s vehicle. Finally,
since driving often involves multi-tasking, shifting ability will be
involving in managing the process of switching between the main
driving task and other tasks. However, people with wide ranging
cognitive abilities appear to perform well at driving. Thus it seems
that the traditional driving task is not overloading CE function
ability for most people.
What happens, though, when novel in-vehicle information
technologies create demanding new secondary tasks? Could it be
that some CE functions become overloaded with a corresponding
decrement in driving performance and a decrease in safety? In
order to test whether this is a far-fetched concern or not, we
designed the experiment below to examine the impact of CE
function ability on driving performance when CE loading tasks
are involved. Should concerns about CE loading prove to be
justified, the methodology used in this paper (assessing the impact
of CE abilities on driving in the context of CE loading secondary
tasks) may help to find problems relating to CE function loading
in in-vehicle interfaces.
We developed the following hypotheses to test in the experiment:
1. Effect of Cognitive workload
An individual's ability on a particular CE function will be
significantly related to primary task (driving-related task)
performance when also performing a secondary task loading on
that CE function. To test this we used a series of regression
analyses with performance on the primary task (either pedal
tracking or target detection) as the criterion, and measured levels
of individual ability (inhibition, updating, and shifting) as the
S U I
S U I
SUI
SUI
Participant 3
Participant 2
Participant 1
Participant 4
SUI
Individual cognitive abilityCognitive demands
by task(s)
CWLSUI
CWL
SUI
C
SUI
Experienced workload
U
SI
predictor variables. Forward and backward stepwise analyses
were run with one pair of analyses for each combination of
primary task complexity (block 1 was lower complexity, block 2
was higher complexity), modality (auditory or visual) and type of
loading on the secondary task (inhibition, shifting, or updating).
We expected that the overall pattern of results in the regression
analyses would tell us the extent to which CE function ability was
driving primary task performance, and in what contexts. If a
particular CE function ability was only a significant predictor of
primary task performance when the secondary task loaded on that
same CE function, then that would show that the secondary task
was harming primary task performance because of its loading on
that function.
Alternatively, if the presence of the secondary task in itself was
making loading on the primary task more critical, then we might
expect to see different CE abilities affecting performance on the
two primary tasks, regardless of which CE function loaded the
secondary task. Specifically, we would predict in this case that
performance on the pedal tracking task would be significantly
related to updating ability, while performance on the target
detection task would be significantly related to inhibition ability.
In addition, to the extent that shifting is related to task switching
we would expect it to be significantly related to primary task
performance in all the dual task conditions.
2. Effect of Perceptual workload
Overall experienced workload is a combination of visual (or
perceptual), manual and cognitive workloads. Since both the main
and CE tasks involve visual processing, we would expect overall
workload to be higher when the information in the secondary task
is presented visually due to the resulting high load on visual
attentional resources (cf. [27]). However, it could also be
hypothesized that mental workload could rise, rather than fall
when an audio secondary task was used due to the fact that
auditory information tends to require more storage in working
memory.
4. SCREENING TEST
Prior to the experiment, we conducted a screening test to select
people who cover a range of different cognitive profiles with
respect to the three CE functions considered in this research.
4.1 Method
4.1.1 Participants
102 people participated in the screening test. Participants were
recruited through recruiting firms, emails to distribution lists and
from notices posted on University of Toronto campus bulletin
boards. The participants consists of 53 males and 49 females, aged
from 16 to 64 years old (M=42.3, SD=13.4). All of the
participants were English speakers living in the Toronto area with
normal vision and hearing.
4.1.2 Tasks
We selected three cognitive tests to measure each participant’s CE
ability based on Miyake et al.’s findings concerning the mappings
between tasks and CE functions [17].
(1) Stroop test (Inhibition): Six color words (‘black’, ‘white’,
‘yellow’, ‘orange’, ‘purple’, and ‘green’) were presented in one of
the six same font colors individually and at random. There were
36 possible word-font color combinations. On each trial, three
color names (response alternatives) were presented in black at the
bottom of the display. The participant's task was to respond with
the color in which the stimulus word was written, by pressing a
corresponding key. The three response alternatives were mapped
to the left arrow key, down arrow key, and right arrow key,
respectively.
(2) Color monitoring test
（
Updating task): Participants were
shown blue, yellow and red circles (8cm in diameter) one at a
time for 500ms in randomized order with an inter-stimulus
interval of 2500ms. The task was to respond when the third
instance of each circle color was presented (e.g., after seeing the
third blue circle, or the third yellow circle), which required
participants to monitor and keep track of the number of times each
color had been presented. For example, if the sequence was ‘‘blue,
red, yellow, yellow, red, blue, yellow, blue, red’’ then the
participant should have responded to the third blue, yellow and
red circle (italicized). In order for momentary mental lapses to
have less impact on task performance, the circle count for each
color was automatically reset to 0 if the participant made a key
press for that color, and participants were informed of this feature
before starting the task. Prior to completing the trial blocks,
participants received a practice session, which continued until
they made 3 correct responses.
(3) Wisconsin Card Sort Test (WCST; Shifting task): In this task,
four stimulus cards were presented to participants. The objects on
the cards could differ in color, quantity, and shape. The
participants were then given an additional card and were asked to
choose which one of the four original cards conformed to the
same category as the additional card. As the classification rule
was not provided to the participants, they had to guess the rule.
They did this based on the pattern of feedback provided to them
(“correct” or “incorrect”), after they chose one of the four cards to
match with the additional card. In this experiment, the
classification rule changed after 10 correct responses under the
rule. The task was finished when a participant completed 8
different rules or 128 trials, whichever came earlier. We used the
number of perseveration errors as the performance measure based
on previous research [17].
4.1.3 Results
Data from six of the participants was removed from the analysis
because of problems in collecting their data (e.g., failing to follow
instructions). The skill levels of the remaining 96 participants
were then assigned into three categories on each of the three
executive functions (inhibition, updating, and shifting) using the
following method. Measures obtained on the experimental tasks
that corresponded to each of the three executive functions were
characterized as low (-1), medium (0) and high (1) by segmenting
the standardized (z-) scores obtained on each measure across the
entire sample of participants. A z-score of less than -1 was
interpreted as low ability (relative to the rest of the sample), a z-
score between -1 and 1 was interpreted as medium ability, and a
z-score of greater than +1 was interpreted as high ability. This
created three variables that represented the three skill levels (high,
medium, and low).
Table 1 CE ability patterns
Group Inhibition Shi fting Updating
screening
test
average 0 0 0 33
high inhibiti on 1 0 0 6
low inhibi tion -1 0 0 2
high shifting 0 1 0 12
low shifting 0 -1 0 5
high updating 0 0 1 4
low updating 0 0 -1 4
-1 -1 0 ( 4)
0-1 1 (1)
1-1 1 (1)
-1 -1 1 ( 1)
Total 96
mixed
30
7
2
main experime nt
CE ability
34
Number of participants
7
5
0
7
3
3
Table 1 shows the results of CE ability patterns and the number of
participants who are classified into the patterns.
5. EXPERIMENT
5.1 Method
5.1.1 Participant
Thirty-four people were selected from the screening sample for
the main experiment . The people selected represented a variety of
different profiles in terms of shifting, updating, and inhibition
ability (however, one limitation was that neither of the two "low
inhibition" people could participate in the main test). The people
who participated in the main experiment consisted of 20 males
and 14 females, aged from 17 to 64 years old (M=42.9, SD=13.2).
The numbers of each cognitive pattern are shown in Table 1.
5.1.2 Primary tasks (driving-related tasks)
We selected two types of tasks, which we assume to be related to
fundamental aspects of driving.
(1) Detect-respond task
This target detection task was designed to simulate a situation to
detect a particular road sign or sudden obstacles while driving and
to respond to it. The task consists of two blocks. In the first block
(easy block), either a red, down-pointing triangle (r=20 pixels) or
a gray circle was presented on the main display for 1500 ms
(Figure 2). Participants were instructed to tap a pedal with their
right foot immediately after (and not until) they saw a red triangle.
The study used six different inter-trial time intervals (750, 1250,
1750, 2250, 2750, 3250 ms) between the end of one trial (when
the participant pushed the foot pedal to make his or her response)
and the display of the stimulus for the next trial. Both the length
of inter-trial intervals and the position that objects appeared in
were varied randomly across trials. There was a 1/2 chance on
each trial of getting either stimulus (a red triangle or gray circle)
and a 1/6 chance of being assigned a specific inter-trial interval.
In the second (more difficult) block, stimuli included a red
triangle, red circle, gray triangle or gray circle (i.e., there were
three distractors, and the target was defined by the conjunction of
two features).
Figure 2 Detect-respond task
(2) Pedal tracking task
This task was designed to simulate performance of keeping inter-
vehicle distance. We chose this task based on a pedal tracking task
used by [26]. Our implementation of the task was designed to
function like an inter-vehicle distance keeping task. A target
rectangle in blue (corresponds to a car in front) and a frame-
shaped area in yellow were displayed on the main display (Figure
3). The participants' goal was to keep the outer edge of the target
rectangle inside the yellow area by controlling a foot pedal.
To simulate adjust inter-vehicle distance controlling an
acceleration pedal, The size (side length) of the target rectangle
(D) was defined by the equation (1).
(1)
Initially D0 was equal to half the width of the acceptable area
(yellow area), V0 equaled 0 km/h and dt was 0.1sec. Sf
represented the fluctuating signal while Lt was a percentage of the
first order lag of the throttle opening. D was the second-order
integral of the difference between the fluctuating signal
(corresponding to the acceleration of the car in front) and the
control signal (corresponding to the acceleration of one’s own car;
the first order lag of the throttle opening %). The fluctuation
signal was generated from a mixture of four sine waves.
Figure 3 Pedal tracking task
5.1.2.1 Secondary tasks (CE tasks)
We developed inhibition, shifting, and updating tasks in both
visual and audio conditions. We also prepared a simple key press
task as a control condition (Table 2).
(1) Visual Inhibition (VI)
We used the Stroop task from the screening test as a visual
inhibition task. Accuracy and RT were used as the performance
measures.
(2) Audio Inhibition (AS)
We used a modified auditory Stroop task based on previous
research [5][19][22]. Three different words ("High", "Day",
"Low") were presented individually in two pitches (High pitch:
290Hz, Low pitch: 110 Hz) with semi-random word-pitch
combinations (the numbers of each combination of word-pitch
was balanced). The participant's task was to indicate the pitch by
pressing a corresponding key (low = left arrow, high = right
arrow).
(3) Visual Shifting (VS)
We wanted to have equivalent shifting tasks in both visual and
audio conditions. However, since it was difficult to utilize the
WCST in an audio condition, we developed a new task that
required rule shifting.
A single digit number (the target number, varying between 1 and
8) was presented on a display with three single digit numbers (the
option numbers, between 0 and 9) underneath it. The option
numbers represented (a) the sum of the target number plus 1, (b)
the decrement of the target number minus 1, and (c) the same
number as the target number (i.e., plus 0). Participants were
expected to apply one of the rules (+1, -1, 0) to the target number,
and then indicate the result by pressing a key that corresponded
with the position of the desired option number (the right arrow for
the option displayed on the right, the left arrow for the option on
the left, and the down arrow for the option presented in the
middle. The horizontal ordering of potential responses (-1, 0, +1)
Easy (Block 1) Hard (Block 2)
Target rectangle fluctuated based on a
mixed sign wave signal.
Goal: control the pedal to keep the target
rectangle inside the yellow frame.
•Tap the foot pedal→The target
rectangle expands.
•Release the foot pedal→The target
rectangle shrinks.
Easy (Block 1) Hard (Block 2)
Frame-shaped area (Yellow)
(Red)
(Red)
Task difficulties: Easy
condition (block 1) had a thick
frame whereas hard condition
(block 2) had a thin frame
Target rectangle (blue)
too large too small
Feedback: The target rectangle
turned to red when it becomes
too large/small
presented along the bottom of the screen changed randomly
between trials. At the start of the task participants were told to
simply guess the rule. After the system provided subsequent
feedback as to whether the rule they applied was the correct
(expected) one or not (a red "X" for incorrect responses),
participants were instructed to find the expected rule as quickly as
possible and to apply the same rule until it changed. After eight
consecutive correct responses, the program changed the rule.
(4) Audio Shifting (AS)
The procedure in this task was equivalent to the Visual Shifting
task except that all the stimuli were presented in audio; A single
digit number (1-8; the target number) was presented in a high-
pitched voice (290Hz) followed by three single digit number (0-9)
in low-pitch voice (110Hz) as options. Feedback to an incorrect
response was given using a beep sound. The three option numbers
corresponded to the left, down and right arrow keys, in that order.
(5) Visual Updating (VU)
The procedure in this task was similar to that used in the updating
task during the screening study. However, in this condition we
used two colored circles instead of 3, and the participants were
instructed to respond to the second blue and second yellow circle.
This visual version of the task was equivalent to the Audio
Updating task except that all stimuli were presented visually.
(6) Audio Updating (AU)
For this task we used the modified procedure based on Miyake et
al. [17] which was modeled on the Mental Counters task
developed by Larson et al. [10]. Participants were presented with
high-pitched tones (880Hz) and low-pitched tones (220Hz) for
500ms, with an inter-stimulus interval of 2500ms. This procedure
was essentially a repetition of the visual updating tasks used in
screening and in the VU condition except that two tones were
used in place of two colored circles: Participants responded to the
second occurrence of any given tone.
(7) Simple key press task (SK, control condition)
In this task, one of the words "Left", "Down" or "Right" was
presented on the secondary display. The participant's task was to
press the key that corresponded to the word (Left = left arrow,
Down = down arrow, right = right arrow). This task was designed
to require roughly equivalent visual and manual workload to the
other CE tasks, so that the effect of cognitive workload could be
assessed.
Table 2 CE task conditions
5.1.3 Apparatus
The main and CE task programs were run on the same computer,
and were shown on the main display and secondary display
correspondingly. Experimental equipment was set up as shown in
Figure 4. Table 3 shows information concerning the
manufacturers and models of the equipment.
Figure 4 Experiment settings
To measure the participants' eye gaze information, a two-camera
Remote Eye-Gaze Estimation (REGT) system (VISION 2020-RB,
El-MAR Inc., [3]) was used. The system consists of two cameras
(1624 x 1224 pixels) and four infrared light-emitting diodes
(LEDs) mounted to either side of the camera (Figure 4).
Table 3 Equipment used in the experiment
5.1.4 Procedure
All experimental sessions were conducted at the University of
Toronto from January through March 2012. Participants
participated in the experiment individually.
The participants performed one of the two driving-related tasks as
a single task. They then performed the driving-related task and CE
tasks as dual tasks. The CE tasks consisted of the six CE function
conditions (AI, VI, AS, VS, AU, VU) and a Simple Key press
(SK) condition. The order of the CE task was varied among
participants in order to avoid order effects. After a 5-10 minute
break following the first driving-related task condition (performed
as a dual task with each of the CE tasks), participants then
performed the second driving-related task condition with the same
seven CE tasks. Each participant was exposed to all 14 dual task
conditions (one CE task at a time), with approximately two
minutes' worth of trials per condition. Ordering of conditions was
counterbalanced between participants. Participants were instructed
to respond as quickly and accurately as they could, and to allocate
their attention in such a manner as to perform as well as they
could on both of the tasks. Participants were paid for their
participation and signed a consent form before participating, in
accordance with a research protocol that was approved by the
University of Toronto Ethics Review Board.
5.2 Results
5.2.1 The effect of CE ability on the primary task
performance under a particular CE loading condition
(1) Detect-respond task
We calculated the median correct response times by participant
and condition, and compared them between different CE ability
groups. Medians, rather than means, were used as measures of
central location as they are robust to the effects of positive skew
that typically occur in distributions of response time measures,
and that was also present in our data. Regression analyses were
Inhibition
Shifting
Updating
Simple Key press
(Control)
Visual
(VI) Stroop task
(VS) Number
calculation rule
task
(VU) Color
monitoring task
(SK) Pressleft, down
or right arrow key
corresponding to the
instructed direction.
Audio
(AI) Auditory
stroop task
“High”
“Low””Day ”in
high/low pitch
(AS) Auditory
number
calculation rule
task
(AU) Tone
monitoring task
Black
2
nd
Blue
2
nd
Yello
w
2
nd
Blue
2
nd
Yello
w
↓
↓
↓
↓
1
st
Blue
1
st
Blue
1
st
Blue
1
st
1
st
RESET
RESET
RESET
RESET
Left
43O
51cm
37.2cm
65cm
Cameras
IR lights
Main display
(22”)
Secondary display
(16”)
Arrow keys
Foot pedal
24cm
Manufacturer/model
Acer/ 23"/58cm Wide LCD Monitor, S231HL
Dell/ E177FPf TFT, E177FPf
Logitech/ Driving Force GT
OS Microsoft/ Windows XP Professional
Motherboard Gigabyte Technology Co., Ltd./ X58A-UD3R
Main display (23'' LCD)
Secondary display (17'' TFT)
Foot pedal
PC
then carried out to assess the degree to which high CE workload
affected detect-respond response times in the presence of CE
loading tasks. Both backward and forward methods of entry were
used on all the regression analyses reported below. However,
since the results for forward and backward entry were similar in
all cases, only the backward entry results are reported below.
For analysis of the detect-respond task data, seven backward entry
stepwise regression analyses were carried out with response time
on the detect-respond task as the dependent measure. One analysis
was carried out for the SK condition, and three analyses each were
carried out for the inhibition and shifting conditions respectively.
These three analyses consisted of one for block one, one for block
two, and one where the slowing in detect-respond time between
block1 and block2 was used as the dependent measure. The
predictor variables in each of these analyses were six measures of
CE ability measured in the screening test. The variables and
results are summarized in Table 4.
For the block 1 in AI condition, the best fitting model (p<.01)
contained one predictor variable (inhibition correct RT) that
explained 25% (r=.503) of the variance in detect-respond
response time. The best fitting model for the block 2 (p<.05) was
also inhibition correct RT as a single predictor, in this case
explaining 16% of the variance (r=.401). No model was found
that predicted the slowing (due to the added difficulty of two extra
distractors in the detect-respond task) between block 1 and 2.
For the block 1 in AS condition, the best fitting model (p<.05)
contained inhibition correct RT and WCST perseveration errors,
which jointly explained 22% (r=.466) of the variance in detect-
respond response time. The best fitting model for the block 2 in
AS (p<.05) was a single measure of shifting ability, WCST rules
completed, in this case explaining 30% of the variance (r=.544).
No model was found that predicted the slowing between block 1
and block 2 of the detect respond task.
For the block 1 in AU condition, the best fitting model (p<.05)
contained inhibition correct RT, inhibition accuracy and WCST
perseveration errors, which jointly explained 28% (r=.524) of the
variance in detect-respond response time. There was no
significant predictive model for the block 2 data. However, a
single variable model involving inhibition accuracy significantly
predicted (p<.05) the slowing between block 1 and block 2 of the
detect respond task, explaining 14% of the variance (r=.367).
For the SK task the best fitting model (p<.05) again contained
only the inhibition correct RT measure, which explained 13%
(r=.366) of the variance in detect-respond response time.
For the block 1 in VI task the best fitting model (p<.001)
contained four predictor variables (the two measures of inhibition
CE ability plus the two measures of shifting CE ability) that
jointly explained 85% (r=.919) of the variance in detect-respond
response time. The best fitting model for the block 2 in VI
(p<.001) included three predictors (the two measures of inhibition
CE ability, plus the number of WCST perseveration errors), which
explained 54% of the variance (r=.734). No model was found
that predicted the slowing between block 1 and block 2 with the
visual inhibition CE task.
For the block 1 in VS task the best fitting model (p<.001)
contained three predictor variables representing each of the three
CE abilities (inhibition correct RT, updating accuracy, and WCST
perseveration errors) that explained 58% (r=.76) of the variance
in detect-respond response time. The best fitting model for the
block 2 in VS (p<.005) contained the same three predictors as had
been found for block 1, in this case explaining 39% of the
variance (r=.62). No model was found that predicted the slowing
between block 1 and block 2.
For the VU task none of the predictive models were statistically
significant either for block one or for block two data. As in most
of the other CE task conditions, no model was found that
significantly predicted the slowing between block 1 and block 2.
The preceding results are summarized in the first two columns of
Table 4, where each cell represents a regression analysis. Note
that no significant predictors were found for the block 2 updating
analyses.
Table 4 Significant Predictors of the Primary Task
Performance by type of CE task CE Function loading
(2) Pedal tracking task
We calculated the mean error rates (the proportion of the time that
the target rectangle was out of the yellow allowable area) by
participant and condition, and investigated which CE functions
can be predictor of the error rates. As with the detect-respond task
we ran a series of backward stepwise regression analyses
examining which of the CE abilities predicted pedal tracking
accuracy in the presence of the different CE task conditions. The
results are summarized in the right-most column of Table 4. Note
that the pedal tracking regression analyses were carried out with
pooled data across both blocks because no significant differences
were found between the blocks.
For tracking accuracy during the AI task as a secondary task,
updating total accuracy was the only significant predictor
(p<.005) explaining 30% of the variance (r=.547).
In the AS condition, tracking accuracy was predicted (p<.001) by
a combination of the three CE abilities (inhibition accuracy,
WCST completed, and updating total accuracy), with 48% of the
variance explained (r=.689).
In the AU condition, only two of the CE abilities were represented
in the best fitting (p<.005) model (WCST completed rules, and
updating total accuracy), accounting for 38% of the variance in
pedal tracking accuracy (r=.613).
In the SK condition, updating total accuracy (p<.001) was the sole
predictor in the selected model explaining 34% of the variance
(r=.583).
In the VI condition, updating total accuracy significantly
predicted pedal tracking accuracy (p<.001), explaining 40% of the
variance ( r=.632).
In the VS condition, updating total accuracy and WCST jointly
predicted the pedal tracking accuracy (p<.005), accounting for
38% of the variance (r=.614).
Finally, in the VU condition, updating total accuracy and WCST
jointly predicted (p<.001 43% of the variance in pedal tracking
accuracy (r=.658).
Primary task
Pedal Tracking
Secondary task
Block 1 Block 2
(No signifi cant
differe nces
between blocks)
Inhibiti on I I U
Shifting I, S S I, S, U
Updating I, S S, U
Inhibiti on I, S I, S U
Shifting I, S, U I, S, U S, U
Updating S, U
Detect-Respond Task
Auditory
Visual
●dependent variable s
detect-respond task: reponse time (RT)
pedal tracking task: error rates (the proportion of the time that the target
rectangle was out of the yellow allow able area)
●predictor variables (based on the screening test results)
inhibiti on (I) : correct RT and accuracy in Stroop test
shifting (S) : prseveration error rate and the number of rules completed in WCST
updating (U) : correct RT and accuracy in color monitoring test
5.2.2 The effect of modality and task types
Based on the eye tracking data, we calculated the proportion of
the time that participants viewed the main vs. the secondary task
display (main display gaze rate). Figure 6 shows the main display
gaze rate and detect-respond task performance (difference of
correct response time between single and dual task;
diffMedianHitRT).
In general, the main display gaze rate was higher in the audio vs.
visual secondary task conditions (i.e., there was less visual
distraction in the audio conditions). However, the VU condition
showed a higher gaze rate on the main display (lower visual
distraction) as compared to the other visual conditions. This may
be because the task could be performed using peripheral vision, as
was reported by some of the participants.
Figure 6. Main display gaze rate and the driving-related task
performance (detect-respond task)
As shown in Figure 6, driving-related task performance was not
impaired as much with the audio secondary tasks. The VI and VS
conditions were visually distracting and results in a slowing of
primary task performance. On the other hand, the VU secondary
task resulted in less visual distraction and relatively little slowing
in primary task performance relative to the single task condition.
One other feature in Figure 6 is that the slowing in primary task
performance with the AS secondary task was much higher than
the other audio conditions and almost comparable with the VI and
VS slowing. This suggests that the AS task was more cognitively
distracting than the other audio secondary tasks.
5.3 Discussion
5.3.1 Individual CE abilities and driving-related task
performance while performing the particular CE task
as a secondary task
The results of the regression analysis did not support our
hypothesis that the match between CE Function loading on a CE
task and individual ability on that CE function predicts task
performance on the driving-related task in a dual task setting. In
other words, we could not find evidence that CE ability can be a
predictor of the driving-related task performance under a
particular CE loading condition.
On the contrary, we found (unexpectedly) a fairly consistent
relationship between CE ability and primary task performance,
regardless of which CE function was loaded by the secondary
task.. The results showed that better inhibition and shifting
abilities helped people perform better in the target detection task
whereas in the pedal-tracking task, it was higher shifting and
updating ability that led to better performance (Table 4).
These results suggest that shifting ability aids in managing dual
tasks, whereas inhibition is required to deal with distraction in the
detect-respond task. Additionally updating appears necessary to
deal with distraction in the pedal tracking task.
The overall pattern of results showed a strong influence of
cognitive distraction on the target detection and pedal tracking
tasks, as indicated by the fact that people who had higher abilities
on specific CE functions were able to perform better on the
driving-related task when it was carried out in a dual task setting.
Why wasn’t the relationship between ability on a particular CE
function and primary task performance affected by a secondary
task that loaded on that function? Two possible explanations are
suggested below.
1. CE functioning changed in the dual task setting
Some previous research studies have argued that dual-task
coordination is a separate component of CE function. When a task
is performed in a dual-task setting, the function of dual-task
coordination may become the most relevant, and strongly loaded,
function. This might explain why shifting ability significantly
predicted primary task performance in almost of all the dual task
settings that we examined.
2. The secondary tasks we designed might not have been pure
enough measures of the CE functions that we were trying to
characterize.
In order to provide a equivalent rule-shifting task in both visual
and audio conditions, we created a new shifting task based on the
WCST. However, we observed that participants often needed to
retain or retrieve the target number while performing this task.
Thus it is likely that the task did not purely load on shifting
function, but also loaded on memory retention/retrieval. Similar
concerns might be raised about the updating and inhibition
secondary tasks that were used in this study.
Thus it is possible that expected secondary task CE function
loadings did not interact with ability on those function because
participants were loaded on several CE functions during the dual-
task, and not only the CE function targeted by the secondary task.
5.3.2 The effect of visual and cognitive workload
Comparison between visual and audio conditions showed that
participants generally looked away from the main display for
longer periods of time in visual vs. audio conditions, except for
the VU condition, where participants seemed to use their
peripheral vision. However, comparison between SK and other
audio conditions showed that audio tasks with high cognitive
loads had been slowed down more. This suggests that "audio UI
with high cognitive load could be more distracting than visual UI
with low cognitive load” . In this research, we used a single test to
estimate each CE ability. However, due to the well-known test
impurity problem (e.g. [21]), no test completely represents a
particular CE ability. Thus it is recommended for further research
to use multiple test batteries for each CE function to assess the
common factors among the tasks.
6. CONCLUSIONS
We had expected that cognitive ability would affect the ability to
perform a driving-related task in the presence of secondary tasks
that loaded on activities requiring CE abilities and this was found
to be generally true. With some exceptions people with higher
cognitive abilities tended to have better performance, both on
pedal tracking, and on the detect-respond task, in the presence of
the CE loading secondary tasks. However, this affect of CE ability
was more notable in the easier conditions of the detect-respond
task, and generally did not affect the slowing that occurred when
the detect-respond task was made more difficult in the second
block through the addition of more distractors. Thus it appears
that the benefits of higher cognitive ability are stronger in simpler
versions of a primary task when performed in the presence of
50
100
150
200
250
300
350
400
450
20
40
60
80
100
AI
AS
AU
SK
VI
VS
VU
Monitor1
DiffMedianHitRT(ms)
Monitor1WatchRate
detect_HitRT
Monitor1WatchRate
distracting tasks that load cognitive abilities. We also found
evidence of a fair amount of interplay between the CE functions
highlighted by the previous research [17] with performance in the
presence of a distracting task representing one function sometimes
being predicted by a combination of the CE abilities, or in some
cases by a different CE ability.
Overall these results indicate that there is a role for detailed
evaluation of CE abilities and their impact on distracted driving.
However, the present results suggest that CE abilities play a larger
role in simpler versions of primary tasks and that the mapping
between the impact of CE abilities and the CE functions required
in the distracting CE task is not a simple one.
7. ACKNOWLEDGMENTS
We would like to thank Chorong Lee, Andrea Jovanovic, Rie
Toriyama, Ryan Kealey and Phil Lam who helped to set up and
run the experiments, and Pierre Duez who helped in the software
development required in this study. We would also like to thank
Professor Moshe Eizenman, Kai Fok and Sahar Javaherhaghighi
for their help in use of the eye tracking system, and David Canella
for his help in revising this paper.
8. REFERENCES
[1] Baddeley, A.D. (1986) Working Memory. Oxford:
Clarendon Press.
[2] Baumann, M. R. K, Petzoldt, T., Hogema, J. & Krems, J. F.
(2008). The effect of cognitive tasks on predicting events
in traffic. In Proceedings of the European Conference on
Human Centred Design for Intelligent Transport Systems.
Lyon, France, 2008, 3-11.
[3] Burgess, P. W. (1997). Theory and methodology in executive
function research. In P. Rabbitt (Ed.), Methodology of frontal
and executive function. Hove, UK: Psychology Press.81–116.
[4] Guestrin, E. D. & Eizenman, M. (2007). Remote point-of-
gaze estimation with free head movements requiring a single-
point calibration, Proceedings of Engineering in Medicine
and Biology Society 2007, 4556 - 4560.
[5] Hamers, J. E, & Lambert, W. E. (1972). Bilingual
interdependencies in auditory perception. Journal of Verbal
Learning and Verbal Behavior, 11, 303-310.
[6] Hancock, P. & Meshkati, N. (Eds.), Human Mental
Workload. Amsterdam: North Holland.
[7] Hurts, K., Angell, L. S., and Perez, M.A. (2011). The
Distracted Driver: Mechanisms, Models, and
Measurement. In P. De Lucia (Ed.), Reviews of Human
Factors and Ergonomics, Vol. 7, 3-57.
[8] Kass, S. J., Cole, K. S., & Stanny, C. J. (2007). Effects of
distraction and experience on situation awareness and
simulated driving. Transportation Research Part F:
Traffic Psychology and Behaviour, 10 (4), 321-329.
[9] Lehto, J. (1996). Are Executive Function Tests Dependent on
Working Memory Capacity? The Quarterly Journal of
Experimental Psychology, Section A, 49:1, 29-50
[10] Larson, G. E., Merritt,C. R. & Williams, S. E., Information
processing and intelligence: Some implications of task,
Intelligence, Volume 12(2), 131–147
[11] Levin, H. S., Fletcher, J. M., Kufera, J. A., Harward, H.,
Lilly, M. A., Mendelsohn, D., Bruce, D., & Eisenberg, H. M.
(1996). Dimensions of cognition measured by the Tower of
London and other cognitive tasks in head-injured children
and adolescents. Developmental Neuropsychology, 12, 17–
34.
[12] Logie, R. H., Cocchini, G., Della Sala, S., & Baddeley, A.
(2004). Is there a specific executive capacity for dual task co-
ordination? Evidence from Alzheimer’s disease.
Neuropsychology, 18, 504–513.
[13] Lowe, C., & Rabbitt, P. (1997). Cognitive models of aging
and frontal lobe deficits. In P. Rabbitt (Ed.), Methodology of
frontal and executive functions, 39–59. Hove, UK:
Psychology Press.
[14] Mäntylä, T., Karlsson, M., & Marklund, M. (2009).
Executive control functions in simulated driving. Applied
Neuropsychology, 16, 11-18.
[15] Matthews, R., Legg, S., and Charlton S. (2003).
Distraction, driving, multiple-task, workload,
performance, Accident analysis and prevention, 35 (4),
451-457.
[16] Miller E.K., & Cohen J.D. (2001). An integrative theory of
prefrontal cortex function. Annual Review of Neuroscience,
24 (1): 167–202.
[17] Miyake, A., Friedman, N. P., Emerson, M. J., Witzki, A. H.
& Howerter A. (2000). The unity and diversity of executive
functions and their contributions to complex ‘‘Frontal Lobe’’
tasks: A latent variable analysis. Cognitive Psychology, 41,
49–100.
[18] Mizobuchi, S., Chignell, M., Suzuki, J., Koga, K. & Nawa, K.
(2011) Central Executive Functions Likely Mediate the
Impact of Device Operation When Driving, Proceeding of
the third conference on automotive user interface and
interactive vehicular applications 2011, 129-136.
[19] Morgan, A. (1989) An auditory Stroop effect for pitch,
loudness, and time, Brain and Language, 36 (4), 592-603.
[20] Ogden, G.D., Levine, I.M., & Eisner, E.J., (1979).
Measurement of workload by secondary tasks, Human
Factors, 21, 529-548.
[21] Rabbitt, P. (1997). Introduction: Methodologies and models
in the study of executive function. In P. Rabbitt (Ed.),
Methodology of frontal and executive function, 1–38. Hove,
UK: Psychology Press.
[22] Roberts K. L. & Hall D. A. (2008). Examining a supramodal
network for conflict processing: a systematic review and
novel functional magnetic resonance imaging data for related
visual and auditory Stroop tasks. Journal of Cognitive
Neuroscience. 20, 1063–1078
[23] Strayer, D. L., Cooper, J. M., & Drews, F. A. (2004). What
do drivers fail to see when conversing on a cell phone?
Proceedings of the 48nd Annual Meeting of the Human
Factors and Ergonomics Society, 2213-2217.
[24] Strayer, D. L., & Drews, F. A. (2007). Cell-phone-induced
driver distraction. Current Directions in Psychological
Science, 16, 128-131.
[25] Szmalec, A., Demanet, J., Vandierendonck, A., and
Verbruggen, F. (2009). Investigating the role of conflict
resolution in memory updating by means of the one-back
choice RT task. Psychological Research Psychologische
Forschung 73:3, 390-406.
[26] Uno, H., and Nakamura, Y. (2010) Study on surrogate
laboratory techniques to assess driver workload induced from
voice-input interface. Proceedings of the JASE annual
congress (Fall), No.99-10, 1-6. (in Japanese)
[27] Wickens, C.D. (2002): Multiple resources and performance
prediction, Theoretical Issues in Ergonomics Science, 3:2,
159-177.
