10/12/2007
Date:___________________
Gregory J. Funke
I, _________________________________________________________,
hereby submit this work as part of the requirements for the degree of:
Doctor of Philosophy
in:
Psychology
It is entitled:
The Effects of Automation and Workload on Driver Performance,
Subjective Workload, and Mood
Gerald Matthews
Chair: _______________________________
Joel S. Warm
_______________________________
Michael Riley
_______________________________
_______________________________
_______________________________
THE EFFECTS OF AUTOMATION AND WORKLOAD ON DRIVER PERFORMANCE,
DOCTOR OF PHILOSOPHY
2007
by
Gregory J. Funke
Committee Chair:
Gerald Matthews, Ph.D.
ABSTRACT
The effects of vehicle automation, drive difficulty, workload transitions, and subjective state on drivers’
performance efficiency and mood were assessed. This combination of driving variables is likely to play a
key role in automotive safety in the future as in-vehicle technologies increase. Two levels of drive
difficulty (straight, curved) were combined factorially with three levels of vehicle-automation (manual,
condition). In the straight roadway condition, road curvature was absent from the drive. The road in the
curved condition was a continuous set of ‘s-curves,’ which required participants to make constant steering
which controlled drivers’ lateral position and longitudinal speed. Participants in the intermittent-
automation condition drove in a mix of manual and automated vehicle control, which required frequent
control transitions. Participants in the manual condition completed the drive without automated vehicle
control. Performance during the experiment was assessed on several indices, including a monitoring task
which required participants to detect pedestrian hazards. Participants completed the Driver Stress
Inventory (DSI; Matthews, Desmond, Joyner, & Carcary, 1997) a measure of stress vulnerability in a
driving context, the Dundee Stress State Questionnaire (DSSQ, Matthews, et al.., 1999; 2002), a measure
of subjective mood state, and the NASA-Task Load Index (TLX; Hart & Staveland, 1988), a measure of
the perceived mental workload associated with a task. Results of the experiment indicated that curved
roadways impaired driver performance, but did not influence workload. Automation facilitated
performance, but the effect was transient and only observed in the continuous-automation condition.
Automation did not reduce driver workload; perceived workload was actually elevated in the intermittent-
automation condition. Workload transition effects were found to relate to impaired lateral vehicle control,
and increased driver errors. Several DSI factors and DSSQ task engagement were found to be predictive
of post-task subjective state and several indices of task performance. Overall, results indicated that
vehicle automation may facilitate driver performance, but transitions between manual and automated
vehicle control may pose substantial safety risks. Human factors implications and solutions are discussed.
i
TABLE OF CONTENTS
REFERENCES ______________________________________________________________ 68
LIST OF TABLES
1. Summary of vehicle control for each vehicle-automation condition in each period ____ 34
2. Mean pre-test and post-test standard scores for DSSQ worry, distress, and
3. Correlations between DSSQ pre- and post-task scores and DSI factors _____________ 50
4. Summary statistics for regressions of post-task DSSQ state factors onto pre-task
LIST OF FIGURES
1. A scene from the straight roadway condition of the experimental drive _____________ 23
2. A scene from the curved roadway condition of the experimental drive _____________ 24
7. Mean SD of steering wheel rate as a function of roadway condition and period ______ 31
10. Mean reaction time in the curved roadway condition as a function of vehicle-
12. Mean SD of steering wheel rate as a function of transition segment and vehicle-
CHAPTER 1
Introduction
Vehicle Automation
In-vehicle technologies are becoming increasingly complex as new devices and systems are
added to modern automobiles (Ashley, 2001). In 2001, Walker, Stanton, and Young reported that several
major automobile manufacturers planned on incorporating systems designed to monitor headway error,
track and avoid collisions, and adaptively maintain a set cruise speed into their production vehicles by
2015. However, recently revealed specifications concerning these same manufacturers’ 2008 models
indicated that, though these features are costly, each of the technological goals described by Walker et al.
(2001) have been realized, and are now commercially available. For example, BMW, Mercedes, Toyota,
and General Motors now offer vehicles equipped with adaptive cruise control (ACC; Kranz, 2006). ACC
units utilize a headway monitoring system to detect slower moving vehicles within the driver’s lane, and
dynamically modify vehicle speed to establish safe following distances and matching speeds (Kranz,
2006). Similarly, several major automobile manufacturers now offer vehicle models featuring integrated
lane departure warning systems (LDWS; Kranz, 2006). These systems differ slightly from each other, but
all utilize some sort of monitoring technology that activates a warning (auditory, visual, vibro-tactile, or a
combination thereof) if drivers begin to exit their lane without signaling (Houser, Pierowicz, &
Fuglewicz, 2005).
mature in sophistication, some researchers estimate that the next step will be to incorporate fully
automated vehicle control, meaning that all driving tasks would be relegated to a machine system
(Walker, Stanton, & Young, 2001), though the final form this automation may take is still being debated.
One possibility is that of an intelligent vehicle highway system (IVHS; Hancock & Parasuraman, 1992;
Owens, Helmers, & Sivak, 1993). Proposed IVHS systems aim to automate most aspects of normal
vehicle operation, including collision avoidance and navigation, while a driver is traveling on specially
2
designed or retrofitted roadways. However, IVHS systems will require large capital investments to
establish the infrastructure necessary for their successful operation (e.g., sensors to download and upload
navigation information to each vehicle on the roadway, etc.), and the resultant systems may also require
‘road-traffic controllers,’ similar to air-traffic controllers, to monitor and direct the resulting automated
traffic patterns.
Other proposed systems would automate vehicles more globally (Walker et al., 2001). For
example, it may be possible to achieve automated vehicle control by combining the functionality of a
headway monitor, a LDWS, and a navigation system. These integrated commercial automation systems
could be offered as an optional ‘automated driving’ package to new vehicle purchasers, thereby placing
the economic burden associated with the automation on the interested consumer. Such automated systems
could be activated (a) at the operator’s command; (b) when the operator’s driving performance fell below
some threshold value previously determined by the technology manufacturer (Matthews, 2001); or (c) at
all times, regardless of the situation or driver preference (Walker et al., 2001).
Of the two proposed systems described above, Congressional legislation has provided limited
support for each. The Intermodal Surface Transportation Efficiency Act of 1991 (H.R. 2950, 1991)
established the Automated Highway System (AHS) program and provided $660 million for projects from
fiscal years 1992-1997. The goal of the AHS program was to "develop an automated highway and vehicle
prototype from which future fully automated intelligent vehicle-highway systems [could] be developed"
(H.R. 2950, 1991). However, more recent legislatorial overtures favor consumer-based automation over
larger, publicly funded projects. For example, H.R. 4931 (2004) was designed to “encourage and
accelerate the nationwide production, retail sale, and consumer use of new commercial and consumer
motor vehicles with intelligent vehicle technology systems.” Though H.R. 4931 was not passed by the
108th Congress (it expired after the midterm elections of 2006), it would have made a tax credit available
to purchasers of “qualified intelligent systems property,” examples of which included navigation systems,
lane departure warnings, automatic crash notification systems, and other similar devices (H.R. 4931,
2004).
3
Automation defined. Given the rapid innovation and development of commercially available in-
vehicle technologies coupled with tacit legislative support, it has become increasingly important to
understand the nature and consequences of vehicle automation for driver performance. Parasuraman and
Riley (1997) have operationally defined automation as “the execution by a machine agent (usually a
computer) of a function that was previously carried out by a human.” According to this definition, what
constitutes automation and what constitutes machine operation will change over time. Tasks that are
completely and permanently allocated to a machine are classified as machine operations, while tasks that
can be accomplished by a human or machine agent are classified as automated. An example of a new
vehicle technology that qualifies as automation under Parasuraman and Riley’s (1997) definition is the
The LS460 can be purchased with an optional ‘advanced parking guidance system’ (APGS), an
automated system that allows the vehicle to parallel park itself, with some assistance from the driver. To
operate the system, the driver must first pull up in front of an open parking space, as they would if they
were going to parallel park manually, and put the car in reverse. Next, the driver activates the automated
system. A series of sensors mounted on the front and rear bumpers of the vehicle estimate the size of the
parking space. From this information, a likely desired parking location is derived. A touch-screen located
in the dashboard displays the current road scene (captured by a rear-mounted camera) and the estimated
parking location to the driver. The driver may then manually adjust the final parking location using a
series of icons located on the touch-screen. Once the final location is established to the driver’s
satisfaction, the vehicle will automatically steer itself into that location, though the driver is still required
The APGS has been reported to work quite well (Vanderwerp, 2006; Wiesenfelder, 2006), though
it is not perfect. Reported problems include (a) the system does not currently feature an alert to drivers if
the desired parking location is too small to accommodate the vehicle (Vanderwerp, 2006), (b) the
automated system may back into or clip other parked vehicles while executing the parallel parking
maneuver (Vanderwerp, 2006; Wiesenfelder, 2006), and (c) establishing the final parking location using
4
the touch-screen system may take 30 seconds or more to accomplish (Vanderwerp, 2006; Wiesenfelder,
2006). However, since this is the first system of its kind, it is likely that future models will address these
technical difficulties.
Benefits and costs of automation. The overriding impetus for the addition of automated
technologies to vehicles seems to be a desire to decrease driver workload and improve driver safety
(Stanton & Marsden, 1996; Walker et al., 2001). Automobile manufacturers are attempting to achieve this
goal chiefly by augmenting drivers’ perceptual capabilities (e.g. preventative collision detection by
headway monitors) or by automating aspects of driving (e.g. automated route-finding using a GPS
system). To some extent, driving may be characterized as requiring dual-task performance as drivers must
maintain lateral and longitudinal vehicle control (a complex tracking task) while scanning the roadway for
potential hazards (a complex vigilance task; Young & Stanton, 2007). Automating aspects of vehicle
control may therefore ease driver workload by reducing the number of concurrent tasks the driver must
perform. Though it may be difficult to argue for a direct causal link between driver workload and
increased accident risk, a substantial body of research supports a relationship between mental workload
and poor driver performance (e.g., Kantowitz & Simsek, 2001), which may, in turn, be related to
increased risk. Therefore, the goal of incorporating automated systems into automobiles would appear to
be a worthwhile one.
According to Wickens and Hollands (2000), the reasons for automating a task may encompass
four broad categories. Tasks may be automated because 1) people cannot perform the task because of
inherent limitations, 2) people may perform the task, but performance is poor or involves high workload,
to perform the function. However, whether or not a task is automated is often determined by technological
and economic feasibility, rather than by careful analysis of the consequences (Parasuraman & Riley,
1997). This approach tends to define operators’ roles in terms of the automation, often relegating them to
positions of supervising and monitoring the automated system (Parasuraman & Riley, 1997), a change
First, automation may not actually reduce operator workload, but instead transfer it to new tasks,
such as gathering information about the state of the automated system and monitoring it for errors
(Parasuraman, Mouloua, Molloy, & Hilburn, 1996; Reinartz & Gruppe, 1993; Stanton & Marsden, 1996;
Stanton & Young, 2000). Additionally, deliberations over when to employ an automated system may
impose cognitive workload over and above that already required by a task (Parasuraman & Riley, 1997).
Research on automation in aviation indicates that it has worked relatively well for aspects of aviation
where pilot workload demands were already quite low, such as routine in-flight operations, leading to
increased crew boredom (Stanton & Marsden, 1996), and perhaps more seriously, mental underload (e.g.,
Hancock & Parasuraman, 1992; Young & Stanton, 2002a, 2007). However, research with pilots also
reveals that many of them feel that aviation automation has actually increased their workload, particularly
during high-workload phases of flight (i.e., during descent and final approach; Wiener, 1985, 1989). The
negative impact of such an increase in workload may be most apparent in time-critical situations, such as
an emergency malfunction of the automated system (Young & Stanton, 2002a, 2007).
A second problem associated with automation is inappropriate operator trust in the system.
Previous research on operator trust has indicated that reliable automation engenders trust (Lee & Moray,
1992) and that monitoring may be inversely related to system reliability (May, Molloy, & Parasuraman,
1993). As a system becomes more reliable, operators tend to become complacent, spending less time and
effort monitoring the automated system for errors. The net result is that operators may become lax in their
Conversely, automation that is less than perfectly reliable may cause operators to overly distrust
the automated system, leading them to disuse or neglect such systems even under circumstances that
clearly favor automated task performance (Parasuraman & Riley, 1997; Stanton & Marsden, 1996; Young
& Stanton, 2007). In addition, operators’ perceptions of the types of errors made by an automated system
may also negatively impact trust. Conspicuous automation errors (Dzindolet, Peterson, Pomranky, Pierce,
& Beck, 2003) and errors that are diagnosed as ‘simple’ or easily avoided under manual control by human
operators (Madhavan, Wiegmann, & Lacson, 2006) may result in under-reliance on the automation as an
6
aid, and inflate operators’ confidence in their skills and abilities to successfully complete a task manually
(Madhavan et al., 2006). However, some research indicates that operators may be relatively insensitive to
differences in automation reliability (Dzindolet, Pierce, Beck, Dawe, & Anderson, 2001), suggesting that
operators’ potential loss of situational awareness (SA; Endsley, 1995; Stanton & Young, 2000; Wickens
& Hollands, 2000). SA refers to operators’ mental representations of a task, and it encompasses factors
that are both internal and external to the task (Endsley, 1995). In the driving context, internal factors
include the state of the automated system, the route selected by the driver, whether or not the driver is
fatigued or stressed, etc. External factors include local traffic patterns, geographical location, current road
conditions, etc. Research indicates that unreliable automation may impair driver SA (Ma & Kaber, 2007).
A driver who trusts an automated system may perceive little need to monitor the system, and thereby lose
situational awareness of the driving context. Even drivers who do detect an automation error may not be
prepared to deal with the problem appropriately. Loss of situational awareness may impair the drivers’
judgments about the driving situation, leading them to make critical errors.
A final concern is the tendency for human operators’ skills and knowledge concerning manual
task performance to degrade with disuse (Parasuraman, 2000; Stanton & Marsden, 1996; Young &
Stanton, 2007). For example, Stanton and Marsden (1996) reported that poor initial training coupled with
operator skill degradation due to disuse have been linked to several aviation incidents. At issue is that
highly reliable, continuous task execution by an automated system may reduce opportunities for manual
control through automatic task execution (perhaps without operator consent or knowledge), and because
highly reliable systems may subtly encourage operators to bypass opportunities to practice manual
Within the past decade, substantive efforts have been made by researchers to understand the
effects of vehicle automation on driver performance and perceived mental workload. Mental workload
has been previously proposed as an inferred construct that mediates between task difficulty, operator skill,
and observed task performance (Moray, 1979). The net result of this research indicates that the utility of
automation may be dependent on the ‘type’ of automation employed. Several authors (e.g., Lee, 2006;
Sheridan, 1997) have differentiated various ‘levels’ of automation, depending on the extent to which the
machine controls system operations, and the degree to which initiation and cessation of automation is
under machine or human control. There have also been efforts to categorize ‘types’ of automation (e.g.,
Parasuraman, Sheridan, & Wickens, 2000). In vehicle driving, functions that may be amenable to
automation include those related to acquiring and analyzing information (e.g., headway monitoring and
collision warning devices) and those that control vehicle trajectory (e.g., adaptive cruise control,
automated parking). Higher-level automation of vehicle operations, such that aspects of vehicle control
are transferred from the driver to the machine, as is envisioned for IHVS and future commercial
automation endeavors, appears to offer the greatest potential for benefits of workload reduction, but also
Empirical research concerning the relation of automated vehicle trajectory to driver workload has
produced conflicting results. For example, several studies have found that automated control of vehicle
speed may reduce driver workload (e.g., Stanton & Young, 2005; Stanton, Young, & McCaulder, 1997;
Young & Stanton, 2004); others have not (e.g., Funke, Matthews, Warm, & Emo, 2007; Nilsson, 1995;
Stanton, Young, Walker, Turner, & Randle, 2001). Similarly, some studies have found that automated
control of lateral position and longitudinal speed may reduce workload (e.g., Stanton et al., 2001; Young
& Stanton, 2002b; Young & Stanton, 2004); others have found it to have no effect (e.g., Desmond,
Recent research by Young and Stanton (2004, 2007) may help to explicate these seemingly
contradictory findings. They suggest that two factors, driving environment and driver skill, may moderate
Influence of driving environment. The first factor, driving environment, refers to external sources
of increased task demands encountered while driving, such as weather, traffic patterns, and road
geometry. Road geometry refers to physical aspects of the roadway (i.e., radius of curvature, lane width,
shoulder width, etc.; Neuman, 1992). Research indicates that driving environment may substantively
influence driver workload, with more difficult environmental circumstances resulting in increased driver
workload and impaired driver performance (see Kantowitz & Simsek, 2001, for a partial review of the
literature). Young and Stanton (2004) argue that the workload reduction benefits from automating lateral
and longitudinal vehicle control may be contingent upon the specific roadway conditions encountered,
with greater benefits accrued under more demanding circumstances. Therefore, automated vehicle
controls, lateral or longitudinal, may not significantly reduce driver workload in comparison to manual
driving on roads (real or simulated) that feature simple or undemanding roadway conditions. On the other
hand, automated vehicle controls may be expected to reduce driver workload in more demanding driving
environments (e.g., ‘stop and go’ traffic, narrow lanes, tight turns, etc.). Subsequent empirical research
Influence of driver skill. The second factor, driver skill, is related to drivers’ proficiencies at
driving tasks, and to the development of automaticity (e.g., Anderson, 1995; Matthews, Davies,
Westerman, & Stammers, 2000). Theories of skill development often focus on performance differences
associated with novice and expert task performance (e.g., Anderson, 1995; Fitts, 1962). One frequently
cited difference is that novice performance is cognitively demanding, entailing substantial workload and
undemanding, with little associated effort or workload (Schneider & Shiffrin, 1977; Shiffrin & Schneider,
1977). Young and Stanton (2007) suggest that, as drivers’ skills approach automaticity, drivers will
experience reduced task workload associated with normal vehicle operations, and consequently,
9
automated vehicle control will produce fewer workload reduction benefits. However, automation may still
be expected to reduce workload for expert drivers under novel or extremely demanding circumstances
(Young & Stanton, 2007). Additionally, Young and Stanton (2007) acknowledge that some components
of driving, such as hazard monitoring, may be too variable for automaticity to develop, and so these
elements continue to require conscious, effortful control (e.g., Harms, 1991; Groeger & Clegg, 1997).
substantially influence driver workload or performance, except perhaps for the most inexperienced of
drivers. However, under difficult and demanding roadway conditions, automation may be expected to
reduce driver workload and potentially improve performance on aspects of driving unrelated to vehicle
control, such as hazard monitoring, though more experienced drivers may receive fewer benefits.
As mentioned previously, a general concern within the automation literature is that automation
may negatively impact human operators’ skills and knowledge concerning manual task performance due
to disuse. However, the prospect of skill degradation is perhaps less likely given that the two automated
vehicle control schemes outlined previously (IVHS and commercial automation) require a mixture of
manual and automated vehicle control. Somewhat ironically, the required combination of controls may
entail a different set of negative consequences for drivers due to workload transition effects, the
potentiality of which has not yet been widely acknowledged within the driving literature. Workload
transition effects may occur when operators experience abrupt shifts between high and low task demands,
or vice versa, and must subsequently adapt to those changes. Within the ‘real-world,’ unexpected and
rapid shifts in task demands may be relatively common. For example, morning commuters can experience
relatively normal driving punctuated by periods of increased demand around highway entrance or exit
ramps, traffic snarls, and traffic accidents. Due to their prospectively ubiquitous nature, the National
Research Council Committee on Human Factors has identified transitions in task demand as an essential
Research concerning the effects of workload transitions typically has found that abrupt decreases
in task demands may impair subsequent task performance (e.g., Cox-Fuenzalida, Beeler, & Sohl, 2006;
Gluckman, Warm, Dember, & Rosa, 1993; Matthews & Desmond, 2002; Ungar et al., 2005); similarly,
abrupt increases in task demands may also impair task performance (e.g., Cox-Fuenzalida, 2007; Cox-
Fuenzalida, Beeler, & Sohl, 2006; Hart & Bartolussi, 1984). However, within the research literature a few
exceptions to these trends have been noted. For example, Helton et al. (2004) found that on a vigilance
task, participants who were switched from a high signal salience to a low signal salience, and vice versa,
and Molloy (1996) found that participants correctly detected a greater number of automation failures
One potential explanation for the observed differences concerning workload transitions may be
the unit of time selected by researchers for comparisons. For example, Gluckman et al. (1993) found that
participants shifted from dual- to single-task monitoring performed more poorly than non-shifted control
participants performing a single monitoring task. However, the observed differences between the two
groups dissipated after shifted participants continued to perform the single task for an additional ten
minutes. Somewhat similarly, Scallen, Hancock, and Duley (1995) found that performance on two
secondary tasks (engine monitoring and fuel management) was significantly disrupted by transitions
between manual and automated control of a tracking task in the 15 seconds immediately following a
transition, but comparisons of longer, five-minute blocks did not show these effects. Together, these
studies indicate that workload transition effects may be relatively brief, and that researchers must be
actively judicious when selecting a unit of time for comparison across experimental groups.
Two explanations have typically been offered for the observed negative effects of workload
Resource theory. Attentional-resource theories (e.g., Norman & Bobrow, 1975) posit that
information processing and task performance are dependent on the availability of system resources.
Typically these resources are believed to exist in a fixed quantity, though there is some evidence to
suggest that this quantity may vary with task demands and other factors (see Young & Stanton, 2002a;
Young & Stanton, 2002b for reviews). These resources are believed facilitate information processing. It
has been previously suggested that workload represents the proportion of resources required to meet the
demands of a task (e.g., Welford, 1978). According to resource theory, as task demands increase, more
effort is required to perform a task and as a result, more resources are expended in the performance of the
task and workload increases. If the available resources are not sufficient to meet the demands of the
current situation, skilled operators may adjust their strategy to compensate (Singleton, 1989), or barring
following a decrease in workload may be due to an increased expenditure of resources during high
workload task performance. Post-transition, shifted operators may have fewer resources available to meet
task demands relative to non-shifted operators performing a low workload task, resulting in comparatively
inferior performance. A shift upward or downward in task demands may also consume resources if rapid
changes in task demands require operators to recruit or reallocate resources for task performance (Huey &
Wickens, 1993), a process which may take time and effort to accomplish (Tsang & Wickens, 1988).
Following a transition, operators may have to make substantial alterations (in attention, strategy, task
directed effort, etc.) while continuing task performance. Changes in task performance following a
workload transition may then be due to decrements in available resources, or to temporary resource
shortfalls as operators adapt to new task demands. Performance may subsequently improve as initial
resource deficits are overcome by cognitive compensatory processes (Gluckman et al., 1993), as operators
A resource-based model may also be appropriate for understanding driver performance given the
literature linking task workload and driver performance mentioned previously. For example, Harms
(1986) found that driver workload ratings on 100-meter segments of roadway covaried with the number
of reported traffic accidents on those segments. In addition, Kantowitz and Simsek (2001) report that
performing a secondary task while driving may negatively impact several indices of driver performance,
such as lateral position, headway, and velocity. Similarly, Horrey and Wickens (2004) found that drivers’
lane keeping and speed regulation were negatively affected by a switch from single-task manual driving
to a dual-task driving situation (driving while completing a digit-dialling task). These findings tend to
support a resource-based model in that dual-task interference is often cited as evidence of resource-
limited performance.
From a resource-based perspective, the relationship between vehicle automation and workload
transitions may be somewhat complicated. Shifts between manual and automated vehicle control, and
vice versa, may negatively impact driver performance due to workload transition effects (though the
potential duration of interference has not yet been identified in driving). This disruption may also
potentially influence drivers’ experience of workload, particularly if drivers are cognizant of the
interference and must actively compensate for it. Conversely, vehicle automation, to the extent that it
actually reduces driver workload, may insulate drivers from resource shortfalls, thereby minimizing the
effects of workload transitions. Therefore, resource theory suggests that driver performance should be
best under conditions that minimize workload and exposure to workload transitions (i.e., on straight roads
with vehicle automation and few control transitions) and worst under conditions that impose continuous
match effort to environmental task demands (Hancock & Warm, 1989; Hockey, 1993; Matthews &
Desmond, 2002). Such performance impairment may be due to a lowering of performance standards
performance may be due to a misperception of task demands. For example, following a sudden change in
weather, drivers may under-estimate the difficulty of subsequent driving conditions, leading them to
apply less effort to the task. However, when perceived workload is high, drivers must mobilize additional
effort to maintain safety and performance, and this additional effort may focus attention on the task and
mitigate the effects of workload transitions. Additionally, performance recovery following workload
transitions may be due to a subsequent revision in assessment of task demands as a result of continued
exposure to post-shift conditions (Ungar et al., 2005). In support of an effort-regulation model of driver
performance, Matthews and Desmond (2002) found that fatigue effects on driving performance were
more pronounced when workload was low (driving a straight road) than when workload was high (driving
curves). Similarly, Matthews (1996) found that stress and worry effects on driver performance were
found in studies of critical vehicle automation failures. These studies typically begin with drivers
experiencing some form of automated vehicle control (lateral, longitudinal, or both). At some
unannounced point within the drive, the automation unexpectedly fails. Following this, drivers must
detect the failure and then act appropriately to bring the vehicle back under control. A frequent finding of
such research is that a substantial proportion of drivers are unsuccessful at this task, resulting in collisions
with other traffic or roadway objects (e.g., de Waard, van der Hulst, Hoedemaeker, & Brookhuis, 1999;
Nilsson, 1995; Stanton et al., 1997, 2001). Over-trust in the automation and complacency may have led
participants in these studies to misperceive workload demands associated with monitoring the automated
system for failures, leaving them under-prepared to detect the error. Subsequently, when the automation
failed, participants may not have applied adequate effort quickly enough to control the vehicle and
prevent a collision.
transitions may have an interactive effect on driver performance. To the extent that vehicle automation
14
encourages drivers to minimize task-directed effort, workload transition effects may be expected to be
exacerbated. In accord with the effort-regulation model, driver performance would be best under
conditions that require continuous task-directed effort (i.e., driving manually on curved roads), and
poorest under conditions that facilitate a misperception of the effort required and which minimize task-
directed effort (i.e., driving under mixed manual and automated vehicle control on straight roads).
Driver stress vulnerability. Research suggests that responses to demanding driving environments
may reflect both external factors and the driver's personality, as it relates to stress vulnerability (e.g.,
Funke et al., 2007; Matthews & Desmond, 2002; see also Matthews, 2001 for a review). Stress may be
seen as an interaction between external stressors and an individual’s cognitive and behavioral responses to
those stressors (Lazarus & Folkman, 1984). One measure designed to assess driver stress vulnerability is
the Driver Stress Inventory (DSI; Matthews, Desmond, Joyner, & Carcary, 1997). The DSI is a measure
of subjective emotional reactions to driving and behavior in demanding driving conditions, which has
been validated in experimental and on-road studies. It assesses personality traits associated with driving,
grouped into five factors labeled aggression, dislike of driving, fatigue proneness, hazard monitoring, and
thrill seeking. Similar driver trait dimensions have been found in Finnish (Lajunen & Summala, 1995)
and Japanese (Matthews, Tsuda, Xin, & Ozeki, 1999) samples, indicating the robustness of these traits
cross-culturally. According to Matthews (2002), aggression relates to anger while driving, and to goals
for maintaining mastery over other drivers. Dislike of driving corresponds to anxiety and stress associated
with the driving task and to fears about driving competence. Fatigue proneness relates to goals for
avoiding discomfort while driving. Hazard monitoring reflects a desire to preempt threat through vigilant
search for danger. Finally, thrill seeking relates to an interest in or enjoyment of danger while driving.
The five traits assessed by the DSI have been shown to be predictive of ‘real-world’ driving
performance. Aggression, thrill seeking, and less reliably, low hazard monitoring have all been linked to
self-reported accident involvement (Matthews, Dorn, & Glendon, 1991; Matthews et al., 1997; Matthews,
15
Tsuda, et al., 1999). Matthews (2002) reported that aggression, thrill seeking, and low dislike of driving
have been related to convictions for driving offenses, such as speeding, and to higher self-reported
violations of the law while driving. In addition, aggression, thrill seeking, dislike of driving, fatigue
proneness, and low hazard monitoring have all been linked to high rates of self-reported unintentional
State response to stress. It is also plausible that the effects of driver stress traits on performance
are in part mediated by state responses to the pressures of driving. For example, high dislike individuals
may suffer from impaired attention due to excessive worry that distracts the driver from the task at hand
(Matthews, 1996). However, investigating the mediating role of states as an influence on performance
requires a multidimensional perspective on subjective stress response, which adequately captures the
different ways in which stress may be experienced. The Dundee Stress State Questionnaire (DSSQ;
Matthews et al., 2002) was designed to assess multiple transient state factors associated with stress,
arousal, and fatigue. Factor analysis of the DSSQ has identified three broad stress dimensions of worry,
task engagement, and distress. The DSSQ has been successfully validated in a variety of studies,
including experiments focused on driver stress (see Matthews, Campbell, et al., 1999 for a review).
Collectively, these studies reveal that different stressors elicit qualitatively different patterns of stress
response, which may have differing impacts on performance. For example, states of driver fatigue elicit
task disengagement and distress, and loss of vehicle control produces elevated distress and worry
(Matthews, 2002).
To the extent that driver stress interferes with performance of driving tasks, perhaps through
inwardly focused attention to emotionally relevant stimuli, it may be expected to result in impaired driver
performance, and may potentially intensify the negative consequences of workload transitions.
Additionally, this effect may be exacerbated under conditions which facilitate a withdrawal of task-
General goals. The present study was designed to investigate the validity of the resource and
drive difficulty, workload transitions, and subjective state. It examined both external task factors believed
to influence workload and stress, and internal state variables that have previously demonstrated influence
on driver performance. Two factors were experimentally assessed. The first factor was a manipulation of
vehicle automation. Two conditions that featured fully automated vehicle control (both lateral and
longitudinal), designed to resemble proposed future vehicle automation systems and chosen to represent a
high level of automation, were compared with a control condition of manual vehicle control. One
automation condition featured minimal transitions between manual and automated vehicle control; the
other required participants to make repeated shifts in control. The second manipulated factor was of
driving difficulty, achieved by varying the curvature of the roadway (straight or curved). Participants in
all conditions completed a secondary hazard monitoring task during the experimental drive which
provided the opportunity to monitor driver performance under single- and dual-task conditions, and across
Driver performance and workload. It was predicted that automation would tend to improve
indices of driver performance and hazard monitoring detections, and reduce subjective ratings of
workload. Exposure to difficult driving circumstances (i.e., curved roadways) was predicted to impair
driver performance and hazard detection, and increase subjective workload. The two performance theories
described previously make differing predictions regarding the interactive effects of automation, drive
difficulty, and workload transitions on indices of driver performance and hazard monitoring. Resource
theory predicts that driver performance and hazard detection should be best under conditions that
minimize drive difficulty and exposure to workload transitions (i.e., on straight roads with vehicle
automation and few control transitions) and worst under conditions that impose continuous levels of high
task demand (i.e., on curved roads without automation). By contrast, effort-regulation predicts that driver
performance and hazard monitoring would be best under conditions that require continuous task-directed
17
effort (i.e., driving manually on curved roads), and poorest under conditions that facilitate a misperception
of the effort required and which minimize task-directed effort (i.e., driving under mixed manual and
Subjective state. An additional purpose for the study was to evaluate potential predictors of
individual differences in stress response and performance. It was anticipated that, as in previous studies,
the five factors assessed by the DSI will be predictive of post-task subjective state and indices of task
performance (Matthews et al., 1998). Consistent with previous research linking task engagement and
performance on difficult, attentionally demanding tasks (e.g., Matthews & Davies, 1998), it was predicted
that higher subjective task engagement should correlate with superior task performance. It was further
expected that worry would relate to performance impairment given that worry directs attention away from
CHAPTER 2
Method
Participants
Cincinnati (59 men, 61 women) served as participants in this experiment. They ranged in age from 18 to
29 (M = 19.76 years, SD = 2.22 years). Participants had a valid driver’s license for a mean duration of
3.78 years (SD = 2.18; Range = 1 – 13 years). Participants were asked to estimate the number of miles
they drive annually; 18.33% reported driving less than 5,000 miles per year, 38.33% reported driving
between 5,000 and 10,000 miles per year, 22.50% reported driving between 10,000 and 15,000 miles per
year, 11.67% reported driving between 15,000 and 20,000 miles per year, and 9.17% reported driving
more than 20,000 miles per year. Participants served in the study to receive research credit required for
their class. They were free to withdraw from the study at any time without penalty.
Experimental Design
condition. Participants were randomly assigned to conditions, with the caveat that equal proportions of
men and women were assigned to each condition. The experimentally manipulated factors were roadway
Changes in driver performance across the duration of the driving task were examined by separating the
drive into nine serial periods (pre-task baseline, periods 1-7, post-task baseline), and changes in driver
mood were examined from pre-task to post-task. Dependent variables employed in this study included
subjective measures of post-task state, assessed by the Dundee Stress State Questionnaire (DSSQ;
Matthews, Joyner, Gilliland, Campbell, Huggins, & Falconer, 1999), and workload, assessed by the
NASA Task Load Index (NASA-TLX; Hart & Staveland, 1988), as well as measures of lateral position,
longitudinal speed, steering wheel rate, and correct secondary task detections obtained through the driving
simulator.
19
Questionnaires
Driver Stress Inventory (DSI). The DSI (Matthews, Desmond, Joyner, & Carcary, 1997) is an
driving context and to evaluate the coping methods typically employed by that individual under stressful
driving conditions. The first section of the DSI is comprised of 12 items designed to evaluate driving
habits and history, including questions concerning length of time since obtaining a license, typical driving
habits, and frequency and severity of driving infractions. The second section is comprised of 41 items
designed to assess a driver on five dimensions of driver stress vulnerability; as described previously, these
five factors correspond to aggression, dislike of driving, hazard monitoring, thrill seeking, and fatigue
proneness. DSI scores are scaled so that they may range from 0-100.
Dundee Stress State Questionnaire (DSSQ). The DSSQ (Matthews et al., 2002) is a 96-item,
experimentally validated measure designed to assess transient states associated with stress, arousal, and
fatigue, and to reflect the multidimensionality of these states. Items on the DSSQ have been factor
arousal (nervousness-relaxation), hedonic tone (pleasantness of mood), intrinsic task motivation, success-
related motivation, self-focused attention (self-awareness, daydreaming, etc.), self esteem, concentration,
confidence and control, task-relevant cognitive interference (worry about task performance), and task-
irrelevant cognitive interference (worry about personal concerns). These correlated scales cluster into
three second-order ‘super factors’ associated with task engagement (energetic arousal, motivation, and
concentration), distress (tense arousal, hedonic tone, and confidence and control) and worry (self-focused
Factor scores for the three higher-order factors utilized in this experiment were calculated from
regression equations using weights derived from a previous study providing normative data (Matthews et
al., 2002). That is, each factor score was estimated as a weighted sum of the first-order scales, using scale
values that were standardized against the normative data. Factor scores are distributed with a mean of 0
20
and standard deviation of 1, so that the values calculated for a sample represent deviations from
NASA-Task Load Index (TLX). The TLX (Hart & Staveland, 1988) is a standard measure of
perceived mental workload that is widely used in human performance research (Wickens & Hollands,
2000). The TLX consists of six rating scales designed to reflect the degree of mental demand, physical
demand, temporal demand, performance, effort, and frustration associated with a task; scales range in
value from 0-100. The mean of the ratings given to all of the scales provides an overall or global index of
workload, while the rating for each subscale provides an index of the contribution of the dimensions
represented in that subscale to the workload of the task. Copies of all of the psychometric instruments
Driving Simulator
All simulated scenarios presented during this experiment were created using
System Technologies, Incorporated’s ‘STISIM Drive’ software (build version 20802). The simulation
program operated on a Dell Optiplex GX620 PC, equipped with a 3.79 GHz processor and 2 GB of RAM.
The simulation was presented to participants through a Westinghouse LVM-42w2 42-inch LCD monitor.
Participants interacted with the simulation by means of a Logitech MOMO Racing Force Feedback Wheel
(model 963282-0403), which included a steering wheel capable of providing realistic feedback by means
of a computer-controlled torque motor, gas and brake pedals, and an adjustable car seat.
Driving Tasks
Practice drive. Prior to experimental data collection, participants in all conditions completed a
15-minute practice drive. This drive was designed to familiarize them with the driving simulator, its
controls, and the driving tasks that they would engage in during data collection. The practice drive was
divided into three periods; each period provided participants the opportunity to practice different aspects
of the experimental drive. Throughout the practice drive, the posted speed limit was 35 mph. The
roadway consisted of two lanes separated by a double yellow line. Each lane was 12 feet wide, and was
21
bordered by a seven-foot wide road shoulder. Occasionally, the road would change from straight to
curved for short sections of the drive. No other traffic was included in the practice drive.
During the first period of the practice, participants drove normally (i.e., without any automation)
to familiarize themselves with controlling the simulated vehicle. This period of manual control lasted for
approximately six minutes. Following this, participants in the continuous-automation and intermittent-
automation conditions experienced a period of automatic vehicle control, during which the computer
regulated the lateral position and longitudinal speed of the vehicle. Automatic vehicle control was
accomplished using the ‘control vehicle’ command, a standard command of the STISIM driving simulator
programming language. These simulator functions activated and ceased at preprogrammed intervals
without input from participants (i.e., no decision or consent was required from participants). Entrance into
this second period of the practice drive was signaled to participants by the appearance of two green stars,
one on each side of the roadway. The stars were 5 feet in diameter, centered 2.5 feet from the edge of the
roadway, and hovered in midair. While automatically controlling participants’ vehicles, the simulation
maintained each vehicle in the center of the lane at a constant speed of 35 mph. Participants were
informed that while the simulation was controlling the vehicle, the steering wheel, accelerator, and brake
pedals would not function. Further, they were informed that the simulated automation was not perfect,
and that they would need to reestablish control of the vehicle if it malfunctioned. Participants in the
manual condition did not experience this period of automated driving, but instead continued driving
normally. The end of the automation period was signaled to participants by the appearance of a pair of red
stars, which were otherwise identical to the previously described green stars. The second period of the
The final phase of the practice drive was signaled to participants by a series of three orange
construction barrels, each approximately 3 feet in diameter, located on the right and left sides of the
roadway. The roadway during this period was similar to that of the second period, except that the
roadway was straight throughout. During this phase, participants were given the opportunity to practice a
hazard-monitoring task that they would also be required to perform during the experimental drive. In this
22
task, pedestrian pairs were stationed directly opposite one another on the shoulders of both sides of the
roadway. They were positioned five feet from the roadway and spaced approximately 200 feet apart. As is
illustrated in Figure 1, one pedestrian of a pair would occasionally begin to walk into the roadway. These
pedestrian icons were meant to simulate potential road hazards. The participant’s task was to detect the
moving pedestrians (the critical signals) and to indicate, using a pair of paddle shifters located on the
underside of the steering wheel, which side of the roadway a moving pedestrian was located on.
Participants were instructed to activate the right paddle for pedestrians on the right side of the road, or the
left paddle for pedestrians on the left side of the road. In all cases, moving pedestrians were programmed
to stop walking before they entered into the roadway (i.e., the task did not require participants to
maneuver around pedestrians in the roadway). Moving pedestrians walked at a rate of 0.2 feet per second.
Pilot testing indicated that participants’ detection rate of the moving pedestrians was approximately 65%,
suggesting that the task was relatively difficult. At a driving speed of 35 mph (the posted speed limit),
pedestrian pairs constituted a stimulus event rate of 15 pairs per minute. Participants were given 5.5
seconds to detect and respond to the moving pedestrians (5 seconds before the participant arrived at the
location of the pedestrian, and .5 seconds as they passed the pedestrian). The computer recorded the
participant’s reaction time and response to each moving pedestrian (correct, incorrect, and no response).
Participants first encountered 15 pedestrian pairs designed to demonstrate the appearance of the
moving pedestrians. As such, 12 of these pairs (80%) featured a critical signal. It was intended that this
high critical signal rate would allow them to observe the appearance of the moving pedestrians. Next,
participants drove a short section of roadway which did not feature pedestrian pairs. Lastly, they were
given a chance to practice the hazard monitoring task under conditions that were similar to those of the
experimental drive. The last set of pedestrians consisted of 60 pairs, of which eight were critical signals.
Critical signals appeared at a rate of 2 per minute, in equal numbers on the right and left sides of the
roadway. At the end of the second set of practice pedestrians, the simulator ended the practice drive. The
third period lasted approximately 5.5 minutes, bringing the duration of the practice drive to approximately
Figure 1. A scene from the straight roadway condition of the experimental drive. The pedestrian on the
right shoulder is walking into the roadway (i.e., the pedestrian is a critical signal).
Experimental drive. Following the practice drive, participants completed the experimental drive
in their assigned experimental condition. In all conditions, the posted speed limit was 35 mph and no
other roadway traffic was present. During the experimental drive, the ‘shape’ of the roadway was
determined by the roadway factor. The straight roadway condition featured no curvature. As is illustrated
in Figure 2, the curved roadway condition, on the other hand, featured no straight sections; it consisted of
continuous ‘s-curves,’ which required participants to make frequent course corrections to remain on the
roadway. In all other ways, the roadway of the experimental drive was identical to that of the practice
drive.
The experimental drive was separated into nine serial periods; the task requirements for each
period were determined by the vehicle-automation factor. In all conditions, the first and last periods of the
experimental drive were baseline measures of driver performance; each lasted approximately one minute.
Following the pre-task baseline were seven consecutive periods, each approximately seven minutes in
duration. In the manual vehicle-automation condition, participants completed each driving period
‘normally’ (i.e., the simulated vehicle operated similarly to a ‘real-world’ vehicle, without any automated
24
Figure 2. A scene from the curved roadway condition of the experimental drive. Neither pedestrian
depicted is a critical signal.
aides). In the continuous-automation condition, participants drove normally during both baseline
measurements, but the vehicle operated under computer control at all other times. In the intermittent-
automation condition, participants shifted between manual and automatic control of the vehicle.
Participants in this condition drove manually during both baselines, and during periods one, three, five,
and seven. During periods two, four, and six the vehicle operated under automatic control. A summary of
vehicle control for each vehicle-automation condition in each period is displayed in Table 1.
Table 1
Summary of vehicle control for each vehicle-automation condition in each period.
Period
Vehicle-
automation Pre-task Post-task
condition baseline 1 2 3 4 5 6 7 baseline
Manual M M M M M M M M M
Continuous-
M A A A A A A A M
automation
Intermittent-
M M A M A M A M M
automation
Note. M = manual vehicle control. A = automated vehicle control.
25
During periods one through seven of the experimental drive, participants in all conditions were
required to perform the previously described pedestrian monitoring task. Pedestrian pairs appeared at a
rate of 15 per minute. Ninety-eight critical signals (moving pedestrians) were included during the
experimental drive, occurring at a rate of two critical signals per minute; these signals occurred in equal
numbers on the right and left sides of the roadway. The duration of the experimental drive was
approximately 51 minutes.
General Procedure
Participants were tested in a windowless laboratory room. After arriving at the experiment, they
were required to complete an informed consent form and present a valid driver’s license. Participants with
restricted licenses were also required to wear their corrective lenses to participate. Those who did not
have a valid driver’s license or their corrective lenses at the time of the experiment were not permitted to
participate.
schedule. They were then asked to complete the DSI and the pre-task DSSQ. After completing the
questionnaires, participants were informed that they would begin the driving task. Participants sat
approximately 4 feet from the LCD monitor at a small table, upon which was the steering box, and below
which the accelerator and brake pedals were positioned. The height of the table and the distance between
the car seat and the table were both adjustable, thereby allowing participants to customize each until they
were comfortable. Participants then completed the practice drive, followed by the experimental drive.
Prior to each, the experimenter read participants a brief verbal description concerning the driving
simulator’s controls and the pedestrian hazard monitoring task (complete instructions for each condition
conditions were also read an additional passage concerning the automated vehicle control they would
After the termination of all driving tasks, participants completed the NASA-TLX and the post-
task DSSQ, with instructions to complete the questionnaire with specific reference to the test drive. The
CHAPTER 3
Results
During the experimental drive, the simulation software recorded driver performance data at a rate
of approximately ten samples per second. The program logged the mean and standard deviation (SD) of
each driver’s lateral position (in feet), longitudinal speed (in mph), and steering wheel rate (in radians per
second). A program was written using Microsoft Visual Basic .NET to compress the resultant raw data
into one minute intervals for further analysis. In addition, the simulation recorded the reaction time, the
number of correct secondary task detections (‘hits’), and the number of incorrect secondary task
detections (‘false alarms’) made by each participant. Across experimental conditions and periods,
however, approximately 40% of participants made no false alarms, and among the remaining participants
who did, the false alarm rate was very low (less than 2%). Therefore, examination of driver performance
on the secondary hazard monitoring task will be limited to measures of percentage correct detections and
reaction time, and will not include signal detection theory analyses of perceptual sensitivity and bias.
Driver Performance
To examine the driver performance data for changes across periods, driver performance indices
were compared across periods 1, 3, 5, and 7; these corresponded to periods of manual vehicle control in
the manual and intermittent-automation conditions. For each subsequently reported index of driver
performance (lateral position, longitudinal speed, steering wheel rate), the mean and SD were computed
for each participant in each period. The continuous-automation condition was excluded from these
analyses because lateral position and longitudinal speed were controlled by the simulator. Driver
performance indices were tested for statistical significance through separate 2 (roadway) × 2 (vehicle-
between-groups factors and period was a within-groups factor. In all subsequently reported analyses
involving repeated measures with more than two levels of the factor, Box’s epsilon was employed to
28
correct for violations of the sphericity assumption (Maxwell & Delaney, 2004). Complete summaries of
Mean lateral position. Analysis of the mean lateral position data did not reveal any statistically
Mean SD of lateral position. Analysis of the mean SD of lateral position indicated statistically
significant main effects for roadway, F (1, 76) = 112.75, p < .05, and period, F (2.51, 190.71) = 8.65, p <
.05. All other sources of variance in the analysis were not statistically significant (p > .05). As is
illustrated in Figure 3, participants displayed less variation in their lateral position in the straight roadway
condition, compared to the curved condition. In addition, participants in both conditions displayed greater
2
Mean SD of Lateral Position
1.5
(ft.)
0.5
Straight
Curved
0
1 3 5 7
Period
Figure 3. Mean SD of lateral position as a function of roadway condition and period. Error bars are
standard errors.
Mean longitudinal speed. Analysis of the mean longitudinal speed indicated a statistically
significant main effect of period, F (2.24, 170.32) = 13.55, p < .05, and a statistically significant roadway
× period interaction, F (2.24, 170.32) = 4.17, p < .05. All other sources of variance in the analysis were
The observed roadway × period interaction was further examined using separate repeated
measures ANOVAs, in which period was analyzed independently for each roadway condition. In these
29
and all subsequently reported post-hoc analyses, the Dunn-Sidak alpha correction was used to control
Type I error rates (Kirk, 1995). A statistically significant main effect of period, F (2.55, 99.57) = 17.99, p
< .05, was detected for the curved roadway condition; no such effect was detected for the straight
condition (p > .05). As can be seen in Figure 4, participants in the curved roadway condition initially
drove more slowly than participants in the straight condition. However, mean longitudinal speeds were
37
Mean Longitudinal Speed
36
35
34
(mph)
33
32
Straight
31
Curved
30
1 3 5 7
Period
Figure 4. Mean longitudinal speed as a function of roadway condition and period. Error bars are standard
errors.
Mean SD of longitudinal speed. Results of the analysis of the mean SD of longitudinal speed
indicated statistically significant main effects for roadway, F (1, 76) = 5.44, p < .05, and period, F (2.35,
178.76) = 4.76, p < .05. All other sources of variance in the analysis were not statistically significant (p >
.05). As is illustrated in Figure 5, participants displayed less variation in longitudinal speed in the straight
roadway condition compared to the curved condition. In addition, participants’ variation in longitudinal
Mean SD of Longitudinal
1.5
Speed (mph) 1
0.5
Straight
Curved
0
1 3 5 7
Period
Figure 5. Mean SD of longitudinal speed as a function of roadway condition and period. Error bars are
standard errors.
Mean steering wheel rate. Analysis of the mean steering wheel rate data indicated statistically
significant main effects of roadway, F (1, 76) = 790.71, p < .05, and period, F (1.95, 147.93) = 4.01, p <
.05. No other sources of variance in the analysis were significant (all p > .05). As can be seen in Figure 6,
participants made smaller steering wheel inputs in the straight roadway condition compared to the curved
condition. Additionally, participants made larger steering wheel inputs across periods.
0.2
Mean Steering Wheel Rate
Straight
Curved
0.15
(rad/s)
0.1
0.05
0
1 3 5 7
Period
Figure 6. Mean steering wheel rate as a function of roadway condition and period. Error bars are standard
errors.
31
Mean SD of steering wheel rate. Analysis of the mean SD of steering wheel rate data indicated
statistically significant main effects for roadway, F (1, 76) = 621.08, p < .05, and period, F (2.24, 170.02)
= 15.73, p < .05, and a statistically significant roadway × period interaction, F (2.24, 170.02) = 15.93, p <
.05.
To further explore the observed roadway × period interaction, separate repeated measures
ANOVAs were calculated in which period was analyzed independently for each roadway condition. A
statistically significant main effect for period was found for both the curved, F (2.85, 111.03) = 9.62, p <
.05, and straight roadway conditions, F (2.08, 81.05) = 16.93, p < .05. The mean SD of steering wheel
rate for each period and roadway condition is displayed in Figure 7. As is illustrated in the figure,
participants exhibited greater variability in steering across periods; however, this effect was more
Straight
0.3
Curved
Rate (rad/s)
0.2
0.1
0
1 3 5 7
Period
Figure 7. Mean SD of steering wheel rate (rad/s) as a function of roadway condition and period. Error
bars are standard errors.
Percentage correct detections. During the experimental drive, participants completed the
pedestrian monitoring task during periods 1 through 7. Fourteen critical signals were presented per period
minute, for a total of 98 critical signals in the experiment. Using participants’ number of correct
32
pedestrian detections, the percentage of correct pedestrian detections were calculated for each participant
ANOVA, in which roadway and vehicle-automation were between-groups factors and period was a
within-groups factor.
The analysis revealed statistically significant main effects for vehicle-automation condition, F (2,
114) = 3.19, p < .05, and period, F (5.14, 585.75) = 3.87, p < .05, and statistically significant roadway ×
period, F (5.14, 585.75) = 4.82, p < .05, and vehicle-automation × period interactions, F (10.28, 585.75) =
5.29, p < .05. All other sources of variance in the analysis were not statistically significant (p > .05).
To further investigate the roadway × period interaction, separate repeated measures ANOVAs
were calculated in which period was analyzed independently for the straight and curved roadway
conditions. A statistically significant main effect of period was detected in the straight roadway condition,
F (4.24, 250.57) = 5.91, p < .05; however, no such effect was detected in the curved condition (p > .05).
As is illustrated in Figure 8, participants in the straight roadway condition initially correctly detected a
greater percentage of the pedestrian hazards. Across periods, however, their detection performance
85
Mean Percentage Correct
80
75
Detections
70
65
60
Straight
55
Curved
50
1 2 3 4 5 6 7
Period
Figure 8. Mean percentage of correct pedestrian detections as a function of roadway condition and
period. Error bars are standard errors.
33
ANOVAs were calculated in which period was analyzed independently for each vehicle-automation
condition. A statistically significant main effect of period was detected for the continuous-automation
condition, F (3.97, 154.74) = 11.38, p < .05; however, no such effect was detected for the manual and
intermittent-automation conditions (both p > .05). The mean percentage of correct pedestrian detections
for each vehicle automation conditions is illustrated in Figure 9. As can be seen in the figure, participants
in the continuous-automation condition initially correctly detected a greater percentage of the pedestrian
hazards. Across periods, however, their performance decreased to levels comparable to that of the manual
100
Mean Percentage Correct
Manual
90 Cont.-Auto.
Inter.-Auto.
Detections
80
70
60
50
1 2 3 4 5 6 7
Period
Figure 9. Mean percentage of correct pedestrian detections as a function of vehicle-automation condition
and period. Error bars are standard errors.
Reaction time. Participants’ mean reaction times for correct pedestrian detections were calculated
for each period. However, 6 participants failed to make at least 1 correct detection in 1 or more periods of
the experimental drive; these participants were excluded from further analyses concerning reaction time
data. Mean reaction times in each condition were compared using a 2 (roadway) × 3 (vehicle-automation)
The analysis revealed statistically significant main effects for roadway, F (1, 108) = 3.92, p < .05,
and period, F (4.55, 490.99) = 5.43, p < .05, as well as statistically significant vehicle-automation ×
period, F (9.09, 490.99) = 2.25, p < .05, and roadway × vehicle-automation × period interactions, F (9.09,
490.99) = 2.12, p < .05. All other sources of variance in the analysis were not statistically significant (all
p > .05).
model ANOVAs were calculated in which vehicle-automation and period were analyzed independently
for each roadway condition. A statistically significant main effect of period, F (4.13, 223.14) = 3.75, p <
.05, and a statistically significant vehicle-automation × period interaction, F (8.26, 223.14) = 3.30, p <
.05, were detected for the curved roadway condition; no such effects were detected for the straight
condition (all p > .05). The vehicle-automation × period interaction in the curved roadway condition was
further examined by means of separate repeated measures ANOVAs in which period was analyzed
independently for each vehicle-automation condition. These analyses indicated a statistically significant
main effect of period in the continuous-automation condition, F (3.87, 73.60) = 12.19, p < .05; no period
effects were detected for the manual or intermittent-automation conditions (both p > .05). As is illustrated
in Figure 10, participants’ reaction time in the curved roadway continuous-automation condition slowed
across periods.
35
5.3
4.8
4.5
Manual
4.3 Cont.-Auto.
Inter.-Auto.
4.0
1 2 3 4 5 6 7
Period
Figure 10. Mean reaction time in the curved roadway condition as a function of vehicle-automation
condition and period. Error bars are standard errors.
As was previously presented in Table 1, during the experimental drive, participants in the
manual to automated vehicle control, and a subsequent transition from automated to manual vehicle
control. Participants in the continuous-automation condition drove manually during the pre-task baseline,
followed by automated vehicle control during periods 1 through 7, and finally manual control during the
post-task baseline. Participants in the intermittent-automation condition experienced three shifts from
manual to automated vehicle control (in transitioning from period 1 to period 2, from period 3 to period 4,
and from period 5 to period 6), and three shifts from automated to manual control (from period 2 to period
Automated vehicle control, achieved through the driving simulator, was perfect in this experiment
(i.e. the vehicle did not deviate from the center of the driver’s lane, and maintained a constant 35 mph).
As such, the focus of subsequently presented analyses in this section was on transitions from automated
vehicle control to manual vehicle control. The aim of these analyses was twofold. First, to test for changes
in manual driver performance across periods of automated control; this was accomplished by comparing
driver performance immediately prior to and following automated vehicle control. Second, to determine if
36
observed changes in driver performance were due to variations over time, or if they were due to exposure
to automated vehicle control. This was accomplished by comparing driver performance in the continuous-
and intermittent-automation conditions relative to performance in the manual condition during the same
periods of time; automation induced performance changes should be indicated by pre-post transition ×
vehicle-automation interactions.
To examine the effects of transition from automated to manual vehicle control on driver
performance, the 60 seconds immediately prior to a transition from manual control to automated control,
and the 60 seconds immediately following a transition from automated control to manual control were
selected for comparison. Based on the observations of Scallen, Hancock, and Duley (1995) concerning
the potentially short durations of automation-related transition effects, the 60 second interval following a
transition from automated to manual control was further subdivided into four 15-second serial segments
to facilitate examination of driver performance across the transition period. In these analyses, the effects
of the experimentally manipulated factors on indices of variation in driver performance (i.e. SD of lateral
position, SD of longitudinal speed, SD of steering wheel rate) were tested using separate 2 (roadway) × 3
automation were between-groups factors and transition segment was a within-groups factor. For these
analyses, segment 0 corresponded to the 60 seconds immediately prior to a transition from manual control
to automated control, and segments 1-4 corresponded to the 15-second intervals immediately following
In addition, during the experimental drive, the simulator recorded the number of moving
violations each driver made during each control transition. To examine these data for differences due to
the experimental manipulations, two categories of moving violations were established: lane deviations
and major violations. Lane deviations were defined as drivers exiting their lane either by crossing the
center yellow line into the opposite lane, or by crossing the white line onto the road’s shoulder. Major
violations were defined as drivers colliding with a pedestrian or drivers exiting the roadway (by crossing
out of their lane, across the road shoulder, and into the verge). These specific types of driver violations
37
were selected for analysis due to the potential severity of consequences for both the driver and others (e.g.
other motorists, pedestrians, etc.) in the event of their occurrence. Seven participants committed major
violations during transition periods (2 in the manual condition, 2 in the continuous-automation condition,
3 in the intermittent-automation condition). Following a major violation, the simulator repositioned the
driver’s vehicle to the center of their lane, and set their speed to zero. These changes took the driving
simulator several seconds to accomplish, during which, all driver performance indices were recorded as
zero by the simulator. As such, the data for participants who committed major violations were excluded
from all analyses concerning the transition periods in which those violations occurred.
Continuous-automation
Mean SD of lateral position. The analysis of the transition from automated to manual control
indicated statistically significant main effects of roadway condition, F (1, 74) = 25.77, p < .05, and
transition segment, F (1.80, 129.34) = 31.07, p < .05, and statistically significant roadway × vehicle-
automation, F (1, 74) = 5.86, p < .05, roadway × transition segment, F (1.80, 129.34) = 26.87, p < .05,
and roadway × vehicle-automation × transition segment interactions, F (1.80, 129.34) = 4.29, p < .05.
To further investigate the roadway × vehicle-automation × transition segment interaction, separate mixed-
model ANOVAs were calculated in which vehicle-automation and period were analyzed independently
for each roadway condition. A statistically significant main effect of segment was detected for the curved
roadway condition, F (1.51, 54.48) = 33.15, p < .05; no such effects were detected for the straight
condition (all p > .05). As the change in lateral variability across segments in the curved roadway
condition was observed in both the manual and intermittent-automation conditions, indicating that the
effect was not solely due to transition from automated to manual control, further post-hoc analyses were
discontinued.
Mean SD of longitudinal speed. Analysis of the transition from automated to manual control
revealed statistically significant main effects of roadway, F (1, 74) = 12.45, p < .05, and segment, F (2.47,
173.09) = 516.96, p < .05, and statistically significant roadway × transition segment, F (2.47, 173.09) =
38
3.18, p < .05, and vehicle-automation × transition segment interactions, F (2.47, 173.09) = 2.93, p < .05.
As the roadway × transition segment interaction was not dependent on vehicle-automation condition,
further post-hoc analyses of this effect were not pursued. However, the vehicle-automation × transition
segment interaction was examined by means of separate repeated measures ANOVAs, in which transition
segment was analyzed independently for each vehicle-automation condition. Statistically significant main
effects of transition segment were detected for the manual, F (2.35, 84.48) = 233.57, p < .05, and
continuous-automation conditions, F (2.75, 98.90) = 283.77, p < .05. As is illustrated in Figure 11,
participants in the manual and continuous-automation conditions displayed a less variation in longitudinal
speed during the post-task baseline compared to the pre-task baseline. However, the pattern of variation is
similar across vehicle-automation conditions, indicating that the observed effect is not due to the
5
Mean SD of Longitudinal
Manual
4 Cont.-Auto.
Speed (mph)
0
0 1 2 3 4
Segment
Figure 11. Mean SD of longitudinal speed as a function of vehicle-automation condition and transition
segment. Depicted are performance changes from pre-task baseline (segment 0) to post-task baseline
(segments 1 through 4). Error bars are standard errors.
Mean SD of steering wheel rate. Analysis of the transition from automated to manual control
indicated statistically significant main effects of roadway condition, F (1, 74) = 26.19, p < .05, and
transition segment, F (1.28, 59.58) = 30.14, p < .05, and statistically significant roadway × transition
segment, F (1.28, 59.58) = 24.36, p < .05, and vehicle-automation × transition segment interactions, F
39
(1.28, 59.58) = 4.92, p < .05. As the roadway × transition segment interaction was not dependent on
vehicle-automation condition, further post-hoc analyses of this effect were not pursued.
measures ANOVAs were calculated in which transition segment was analyzed independently for each
vehicle-automation condition. These analyses indicated statistically significant main effects of transition
segment for both the manual, F (1.90, 68.41) = 16.56, p < .05, and continuous-automation conditions, F
(1.13, 40.50) = 13.04, p < .05. The mean SD of steering wheel rate as a function of transition segment and
vehicle-automation condition is illustrated in Figure 12. As can be seen in the figure, participants in the
during the transition from automated to manual vehicle control). Moreover, the degree of variation was
greater in the continuous-automation condition compared to the manual condition, suggesting that the
difference was due to the shift from automated to manual vehicle control.
0.15
Mean SD of Steering Wheel
Manual
Cont.-Auto.
Rate (rad/s)
0.1
0.05
0
0 1 2 3 4
Transition Segment
Figure 12. Mean SD of steering wheel rate as a function of transition segment and vehicle-automation
condition. Depicted are performance changes from pre-task baseline (segment 0) to post-task baseline
(segments 1 through 4). Error bars are standard errors.
Lane deviations. A Kruskal-Wallis H-test was utilized to examine the lane deviations data for
differences in the frequency of lane deviations made during the transition from automated to manual
control due to the experimental manipulations. In this test, the factorial combinations of roadway
40
separate groups for analysis. The results of the analysis indicated a statistically significant difference in
the median number of lane deviations between conditions, χ2 (3, N = 80) = 24.66, p < .05. Post-hoc
Mann-Whitney U-tests did not reveal any statistically significant differences in the number of lane
deviations between the manual and continuous-automation conditions, for either roadway condition (both
p > .05). However, a trend within the data suggested that, in the curved roadway condition, the median
number of lane deviations was greater in the continuous-automation condition (median = 1) compared to
The lane deviations data were also analyzed for differences in the number of participants
committing these errors during the transition, using a Chi-square test in which vehicle-automation was
analyzed independently for each roadway condition. However, no statistically significant differences
between the manual and continuous-automation conditions were detected (p > .05).
Major violations. A Kruskal-Wallis H-test was again utilized to examine the major violations data
for differences in the frequency of such violations due to the experimental manipulations. However, no
statistically significant differences between the manual and intermittent-automation conditions were
The major violations data were also analyzed for differences in the number of participants
committing these errors during the transition, by means of separate Chi-square tests in which vehicle-
automation was analyzed independently for each roadway condition. However, no statistically significant
differences between the manual and intermittent-automation conditions were detected (all p > .05).
Intermittent-automation
transitions from automated vehicle control to manual control during the experimental drive: from period 2
to period 3, from period 4 to period 5, and from period 6 to period 7. Changes in driver performance due
mixed-model ANOVAs; each transition was examined using a separate ANOVA. In all such analyses,
roadway and vehicle-automation were between-groups factors and transition segment was a within-
groups factor.
Mean SD of lateral position. In the transition from period 2 to period 3, results of the analysis
indicated statistically significant main effects of roadway condition, F (1, 76) = 175.14, p < .05, vehicle-
automation condition, F (1, 76) = 10.89, p < .05, and transition segment, F (1.75, 132.69) = 9.92, p < .05,
and statistically significant roadway × vehicle-automation, F (1, 76) = 8.88, p < .05, roadway × transition
segment, F (1.75, 132.69) = 10.57, p < .05, vehicle-automation × transition segment, F (1.75, 132.69) =
5.27, p < .05, and transition segment × roadway × vehicle-automation interactions, F (1.75, 132.69) =
mixed-model ANOVAs were calculated in which vehicle-automation and period were analyzed
independently for each roadway condition. Analysis of the straight roadway condition revealed no
statistically significant differences between conditions (p > .05); analysis of the curved roadway
condition, however, indicated statistically significant main effects of transition segment, F (1.46, 55.51) =
11.64, p < .05, and vehicle-automation, F (1, 38) = 12.85, p < .05, and a statistically significant vehicle-
The transition segment × vehicle-automation interaction in the curved roadway condition was
further examined by means of separate repeated measures ANOVAs in which transition segment was
analyzed independently for each vehicle-automation condition. These analyses indicated statistically
significant main effects of transition segment for both the manual, F (3.12, 59.18) = 7.36, p < .05, and
intermittent-automation conditions, F (1.31, 24.91) = 8.95, p < .05. The mean SD of lateral position in the
illustrated in Figure 13. As can be seen in the figure, participants in the intermittent-automation condition
displayed a sharp increase in lateral variability in segment 1 (i.e., during the transition from automated to
manual vehicle control). The magnitude of disruption was greater in the intermittent-automation condition
42
compared to the manual condition, suggesting that the difference was due to the shift from automated to
2
(ft.)
0
0 1 2 3 4
Transition Segment
Figure 13. Mean SD of lateral position in the curved roadway condition as a function of transition segment
and vehicle-automation condition. Depicted are performance changes from period 1 (segment 0) to period
3 (segments 1 through 4). Error bars are standard errors.
Analysis of the transitions from period 4 to 5 and from period 6 to 7 revealed no statistically
Mean SD of longitudinal speed. Analyses of the mean SD of longitudinal speed data for the
transitions from period 2 to 3 and from period 4 to 5 revealed no statistically significant pre-post
transition × vehicle-automation interactions (all p > .05). Analysis of the transition from period 6 to
period 7, on the other hand, revealed a statistically significant main effect of transition segment, F (3.27,
248.23) = 15.43, p < .05), and a statistically significant vehicle-automation × transition segment
The observed transition segment × vehicle-automation interaction for the transition from period 6
to period 7 was further examined by means of separate repeated measures ANOVAs in which transition
segment was analyzed independently for each vehicle-automation condition. These analyses indicated
statistically significant main effects of transition segment for both the manual, F (2.65, 103.33) = 12.30, p
< .05, and intermittent-automation conditions, F (3.17, 123.70) = 7.90, p < .05. As can be seen in Figure
43
14, variability in longitudinal speed decreased across segments in both vehicle-automation conditions,
though variability was greater in the intermittent-automation condition during segments 1 and 2,
2
Manual
Mean SD of Longitudinal
Inter.-Auto.
1.5
Speed (mph)
0.5
0
0 1 2 3 4
Segment
Figure 14. Mean SD of longitudinal speed as a function of vehicle-automation condition and transition
segment. Depicted are performance changes from period 5 (segment 0) to period 7 (segments 1 through
4). Error bars are standard errors.
Mean SD of steering wheel rate. Analyses of the mean SD of steering wheel rate for the
transitions from period 2 to 3 and from period 4 to 5 revealed no statistically significant vehicle-
automation × transition segment interactions (all p > .05). For the transition from period 6 to period 7,
statistically significant main effects for roadway condition, F (1, 75) = 886.61, p < .05, and transition
segment, F (2.66, 199.29) = 192.16, p < .05, and statistically significant transition segment × roadway, F
(2.66, 199.29) = 69.96, p < .05, and roadway × vehicle-automation × transition segment, F (2.66, 199.29)
for the transition from period 6 to period 7, separate mixed-model ANOVAs were calculated in which
vehicle-automation and segment were analyzed independently for each roadway condition. Analysis of
the straight roadway condition revealed a statistically significant main effect of transition segment, F
(2.60, 98.84) = 52.16, p < .05, and a statistically significant vehicle-automation × transition segment
44
interaction, F (2.60, 98.84) = 5.36, p < .05. Analysis of the curved roadway condition, however, indicated
a statistically significant main effect of transition segment, F (2.53, 93.44) = 143.40, p < .05. As the main
effect of transition segment in the curved roadway condition was not dependent on vehicle-automation
The transition segment × vehicle-automation interaction in the straight roadway condition was
further examined by means of separate repeated measures ANOVAs in which transition segment was
analyzed independently for each vehicle-automation condition. These analyses indicated statistically
significant main effects of transition segment for both the manual, F (2.01, 38.16) = 42.86, p < .05, and
intermittent-automation conditions, F (1.96, 37.18) = 19.94, p < .05. The mean SD of steering wheel rate
in the straight roadway condition as a function of transition segment and vehicle-automation condition is
illustrated in Figure 15. As can be seen in the figure, variability in steering wheel rate decreased across
segments in both vehicle-automation conditions, though variability was greater in the intermittent-
automation condition during segments 1, suggesting this may be due to a control transition effect.
0.06
Mean SD of Steering Wheel
Manual
Inter.-Auto.
Rate (rad/s)
0.04
0.02
0
0 1 2 3 4
Segment
Figure 15. Mean SD of steering wheel rate as a function of vehicle-automation condition and transition
segment. Depicted are performance changes from period 5 (segment 0) to period 7 (segments 1 through
4). Error bars are standard errors.
Lane deviations. Separate Kruskal-Wallis H-tests were utilized to examine the lane deviations
data for differences in the frequency of lane deviations occurring during control transitions due to the
45
experimental manipulations. As previously, in these tests, the factorial combinations of roadway (straight,
groups for analysis. In the transition from period 2 to period 3, the analysis indicated a statistically
significant difference in the median number of lane deviations between conditions, χ2 (3, N = 80) = 29.69,
p < .05. Post-hoc Mann-Whitney U-tests indicated that, in the curved roadway condition, the median
number of lane deviations was greater in the intermittent-automation condition (median = 1) than in the
manual condition (median = 0), U = 110.50, p < .05; however, no such differences was found between the
manual and intermittent-automation conditions in the straight roadway condition (p > .05).
For the transitions between periods 4 and 5, and periods 6 and 7, separate Kruskal-Wallis H-tests
indicated statistically significant differences in the median number of lane deviations between conditions,
χ2 (3, N = 80) = 27.83 and 25.63, respectively, both p < .05. However, follow up Mann-Whitney U-tests
indicated that the observed differences were due to the roadway factor, rather than vehicle-automation
condition.
The lane deviations data were also analyzed for differences in the number of participants
committing these errors during transition periods, by means of separate Chi-square tests in which vehicle-
automation was analyzed independently for each roadway condition in each transition period. The
analysis of the transition from period 2 to period 3 indicated a statistically significant difference between
conditions in regards to the number of participants who committed lane deviations in the manual and
intermittent-automation conditions for the curved roadway condition, χ2 (1, N = 40) = 8.29, p < .05.
Fewer participants in the curved roadway manual condition (frequency = 4) committed lane deviations
during the transition compared to the intermittent-automation condition (frequency = 13). However, no
such difference between was detected for the straight roadway condition (p > .05).
The Chi-square tests for the transitions from periods 4 to 5, and periods 6 to 7 did not identify any
statistically significant differences between the manual and intermittent-automation conditions during
either of the transition periods (both p > .05). However, trends within the data suggested that in the curved
46
roadway condition, during the transitions from period 4 to 5 and period 6 to 7, fewer participants in the
manual condition committed lane deviations (frequencies = 9 and 8 respectively) compared to the
Major violations. Separate Kruskal-Wallis H-tests were again utilized to examine the major
violations data for differences in the frequency of such violations due to the experimental manipulations.
conditions were detected for any of the transition periods (all p > .05).
The major violations data were also analyzed for differences in the number of participants
committing these errors during transition periods, by means of separate Chi-square tests in which vehicle-
automation was analyzed independently for each roadway condition in each transition period. However,
no statistically significant differences between the manual and intermittent-automation conditions were
detected for any of the transition periods (all p > .05). However, a trend within the data suggested that, for
the transition between periods 4 and 5, in the curved roadway condition, a greater number of participants
DSSQ mood. Mean pre-test and post-test standard scores for all experimental conditions are displayed
for the three DSSQ factors in Table 2. Test-retest correlations between the three DSSQ factors pre- and
post-task were .73, .57, and .50 for worry, task engagement, and distress respectively (all p < .05),
indicating that participants mood states were relatively stable from pre- to post-task.
47
Table 2
Mean Pre-Test and Post-Test Standard Scores for DSSQ Worry, Distress, and Engagement as a
Function of Roadway and Vehicle-Automation Conditions.
Worry Engagement Distress
Exp. Condition Pre Post Pre Post Pre Post
Straight
Manual .29 (.17) -.08 (.24) .36 (.21) -.79 (.19) -.21 (.18) .35 (.20)
Cont.-auto. .18 (.26) -.06 (.28) .50 (.13) -.79 (.20) -.36 (.12) .29 (.18)
Inter.-auto. .27 (.26) .08 (.26) -.04 (.19) -1.41 (.19) -.23 (.17) .81 (.20)
Curved
Manual .19 (.24) -.02 (.20) .26 (.18) -.43 (.20) -.50 (.19) .40 (.17)
Cont.-auto. .03 (.24) -.13 (.23) .51 (.20) -.77 (.24) -.40 (.21) .37 (.16)
Inter.-auto. .33 (.20) .18 (.22) .65 (.21) -.40 (.24) -.46 (.17) .37 (.20)
Mean .22 (.09) .00 (.10) .38 (.08) -.76 (.09) -.36 (.07) .43 (.07)
Note. Values enclosed in parentheses are standard errors.
To determine if participants’ initial mood states were similar before exposure to the experimental
conditions to which they were assigned, separate 2 (roadway) × 3 (vehicle-automation) ANOVAs were
calculated utilizing each of the pre-task DSSQ factors as dependent variables. The results of these
analyses indicated that participants reported similar pre-task levels of worry, task engagement, and
Next, the effects of the experimentally manipulated factors on drivers’ subjective states were
utilizing the three DSSQ factors as dependent variables. In these analyses, roadway and vehicle-
automation were between-groups factors and pre-post task was a within-groups factor. The aim of these
analyses was to test for changes in state from pre- to post-task (indicated by a main effect of pre-post
task) and for variation of such changes with the experimental factors (indicated by pre-post task × factor
interactions).
The analyses of the data for the worry, task engagement, and distress factors revealed that the pre-
post task difference was statistically significant for each factor, Fworry (1, 114) = 9.69, p < .05; Fengagement (1,
114) = 218.26, p < .05; Fdistress (1, 114) = 116.95, p < .05. All of the pre-post × factor interactions in the
48
analyses of each DSSQ dimension lacked significance (p > .05). In total, all three DSSQ factors showed
significant changes in state from pre- to post-task assessment, such that distress increased, but worry and
Driver workload. The effects of the experimentally manipulated factors on drivers’ ratings of
ANOVA, in which roadway and vehicle-automation were between-groups factors and TLX subscale was
a within-groups factor. The results of the analysis indicated statistically significant main effects for
vehicle-automation, F (2, 114) = 3.49, p < .05, and TLX subscale, F (4.18, 476.00) = 43.13, p < .05. In
addition, though not statistically significant, a trend in the data suggested that participants rated the
workload as lower in the straight roadway condition (M = 51.15, SE = 1.92) compared to the curved
A post-hoc Tukey’s Honestly Significant Difference test (HSD; Kirk, 1995) of vehicle-
automation condition indicated that participants rated their workload as higher in the intermittent-
automation condition (M = 58.00, SE = 1.69) compared to both the manual and continuous-automation
conditions (M = 51.12 and 50.97, SE = 2.46 and 2.25, respectively); no such difference was detected
Mean workload ratings for each TLX subscale are illustrated in Figure 16. A post-hoc Tukey’s
HSD test of the TLX subscales indicated that the three highest rated subscales, mental demand, effort, and
frustration, did not differ significantly from each other (all p > .05), but that each of those scales was rated
significantly higher than the temporal demand, physical demand, and performance subscales (all p < .05).
In addition, the temporal demand subscale was rated as significantly higher than the physical demand and
performance subscales (both p > .05); no ratings difference were found between the physical demand and
100
50
25
0
Mental Physical Temporal Perf. Effort Frust.
TLX Subscale
Figure 16. Mean workload ratings for each TLX subscale. Error bars are standard errors.
Correlations between the pre-task and post-task DSSQ secondary factors and the five dimensions
of driver stress vulnerability assessed by the DSI are displayed in Table 3. Examination of the correlations
indicated that dislike of driving was related to lower task engagement and higher distress prior to the
experimental drive, implying that high dislike drivers anticipated a more stressful driving experience.
Post-task, for the whole sample pooled across task conditions, dislike continued to relate to lower
engagement and higher distress. Hazard monitoring was related to increased task engagement pre- and
post-task. Aggression and thrill seeking were related to increased worry pre- and post-task, and thrill
seeking was related to increased pre-task distress. Fatigue proneness was related to increased pre-task
worry, increased pre- and post-task distress, and decreased pre- and post-task task engagement.
50
Table 3
Correlations Between DSSQ Pre- and Post-Task Scores and DSI Factors
Task
Worry Distress
Engagement
DSI factors Pre Post Pre Post Pre Post
Aggression .43** .27** -.02 -.20* .16 .14
Dislike of driving .05 .05 -.27** -.27** .30** .34**
Hazard monitoring .07 .07 .18* .23* .02 -.04
Thrill seeking .40** .27** -.07 -.16 .20* .18
Fatigue proneness .22* .10 -.36** -.27** .33** .38**
Note. * p < .05, ** p < .01
To further explore the observed correlations between DSSQ post-task scores and DSI factors,
several hierarchical multiple regression analyses were conducted. The aim of these regressions was to
assess the predictive power of the DSI factors with other task variables controlled. The purpose of these
analyses was to determine if DSI factors were predictive of post-task DSSQ scores, or if the correlations
For these regressions, effect-coded vectors for the categorical task variables were created, using
the method recommended by Pedhazur (1997). Roadway condition was represented by a single value (1
vs. -1), whereas two vectors for vehicle-automation were created, with the value of each vector coded as
1, 0, or -1. The interaction term was constructed as the product vector of the roadway and vehicle-
automation vectors. These values were then individually used as predictors in each hierarchical multiple
regression.
For the three DSSQ factors as criteria, predictors were entered in four steps. The first step entered
the appropriate DSSQ pre-task score (worry, task engagement, or distress). This variable was entered
first to control for individual differences in initial state that may have carried over post-task. Next, the
roadway and vehicle-automation vectors were entered, followed by the effect-coded vectors for the
roadway × vehicle-automation interaction. In the fourth step of the regression, the DSI factors were
The regression of worry indicated that the only significant predictor of post-task worry was its
pre-task levels, F (1, 118) = 135.20, p < .05, standardized beta = .73. Regression of post-task task
engagement indicated that significant predictors included pre-task task engagement and DSI aggression, F
(11, 108) = 7.49, p < .05, standardized betas = .47 and -.17 respectively. Regression of post-task distress
indicated that significant predictors included pre-task distress, dislike of driving, and fatigue proneness, F
(11, 108) = 5.64, p < .05, standardized betas = .36, .17, and .20 respectively. The outcomes of these
regressions indicated that the DSI factors provide moderate additional predictive power to estimates of
post-task subjective state. Summaries of the results for each regression equation are displayed in Table 4.
To assess the impact of the DSSQ and DSI factors on driver performance, a second series of
hierarchical multiple regressions was calculated. The aim of these regressions was to assess the predictive
power of these factors with task variables controlled. For these analyses, an overall mean index of driver
performance for each of the performance variables recorded by the simulator (mean and SD of lateral
position, longitudinal speed, and steering wheel rate) was determined for each experimental condition
across periods 1, 3, 5, and 7. These values, plus the number of correct pedestrian detections and
concomitant mean reaction times were then individually used as criteria in four-step hierarchical multiple
regressions. As in previous analyses, the continuous-automation condition was excluded from regressions
related to lateral and longitudinal vehicle control. The four steps, entered successively, into these
regression equations were: 1) the effect-coded vectors for roadway and vehicle-automation conditions, 2)
their interaction, 3) the linear terms for the three post-task DSSQ factors, and 4) the linear terms for the
five DSI factors. Summaries of the results for each regression equation are shown in Table 5.
For the mean number of correct pedestrian detections, statistically significant predictors included
roadway condition, vehicle-automation condition, and post-task task engagement, F (8, 111) = 2.93, p <
.05, standardized beta task engagement = .27. Statistically significant predictors of mean correct detection
52
Table 4
Summary Statistics for Regressions of Post-Task DSSQ State Factors
onto Pre-Task DSSQ State Factors, Roadway and Vehicle-Automation
Conditions, Roadway × Vehicle-Automation Interaction, and DSI Factors.
Criterion: Post-task worry
Step Predictors R2 ∆R2 df F for ∆
1 Pre-task worry .53 .54 1, 118 135.20
reaction time were roadway condition and DSI thrill seeking, F (13, 106) = 1.87, p < .05, standardized
beta thrill seeking = -.30. For the mean SD of steering wheel rate, statistically significant predictors included
roadway condition and post-task task engagement, F (6, 73) = 90.01, p < .05, standardized beta task
engagement = -.13.
For the mean lateral position, mean longitudinal speed, and mean SD of longitudinal speed the
regression results did not reveal any statistically significant predictors (all p > .05). For the mean SD of
53
lateral position and the mean steering wheel rate the only statistically significant predictor was roadway
condition, F (2, 77) = 44.42 and 346.14 respectively, both p < .05. The outcomes of these regressions are
broadly consistent with the previously reported ANOVA results, and support the moderate additional
Table 5
Summary Statistics for Regressions of Performance Variables onto
Roadway and Vehicle-Automation Conditions, Roadway × Vehicle-
Automation Interaction, Post-Task DSSQ State Factors, and DSI
Factors.
Criterion: Correct Pedestrian Task Detections
Step Predictors R2 ∆R2 df F for ∆
1 Task variables .07 .07 3, 116 3.10
2 Task variables .08 .01 5, 114 2.08
Interaction of task variables
3 Task variables .17 .09 8, 111 2.93
Interaction of task variables
Post-task DSSQ factors
Table 5 (continued)
Criterion: Mean Longitudinal Speed
Step Predictors R2 ∆R2 df F for ∆
1 No significant predictors
CHAPTER 4
Discussion
The present study was designed to investigate the validity of the resource and effort-regulation
models as foundations for understanding driver performance as affected by automation, drive difficulty,
workload transitions, and subjective state. It was initially predicted that automation would tend to
improve indices of driver performance and hazard monitoring detections, and reduce subjective ratings of
workload. It was also predicted that exposure to difficult driving circumstances (i.e., curved roadways)
would impair driver performance and hazard detection, and increase subjective workload. Differing
predictions regarding the interactive effects of automation, drive difficulty, and workload transitions on
indices of driver performance and hazard monitoring were made based on resource and effort-regulation
models. Resource theory predicted that driver performance and hazard detection would be best under
conditions that minimize drive difficulty and exposure to workload transitions (i.e., in the straight roads
continuous-automation condition) and worst under conditions that impose continuous levels of high task
demand (i.e., in the curved roadway manual condition). By contrast, effort-regulation theory predicted
that driver performance and hazard monitoring would be best under conditions that required continuous
task-directed effort (i.e., in the curved roadway manual condition), and poorest under conditions that
facilitated a misperception of the effort required and which minimize task-directed effort (i.e., in the
In addition, it was predicted that the five DSI factors would be predictive of post-task subjective
state and indices of task performance (Matthews et al., 1998). It was also predicted that task engagement
should relate to superior driver performance, and worry would relate to driver performance impairments.
Overall, the results of the current experiment tended to support predictions from a resource-based
account of driver performance, as drivers displayed decrements in vehicle control on several indices of
performance across the experimental drive (e.g., drivers demonstrated greater variation in lateral position,
longitudinal speed, and steering wheel rates over the duration of the experimental drive), and greater
As predicted, automated vehicle control was found to facilitate driver performance on the
secondary pedestrian monitoring task. However, these benefits were only observed in the continuous-
automation condition, and over the course of the experimental drive, detections declined to levels similar
to those observed in the manual condition. By contrast, periods of automated control did not significantly
Contrary to initial prediction, access to automation also did not reduce workload relative to manual
driving; repeated transitions between manual and automated control were associated with increased driver
workload.
As predicted, drive difficulty did impair driver performance on several indices (e.g., variability in
lateral position, longitudinal speed, and steering wheel inputs) and was related to increased workload.
Results concerning workload transitions were more consistent with a resource-based model of automation
effects, in that performance impairments were generally transient, infrequent, and more pronounced in the
Hypotheses concerning individual difference factors were reasonably supported. The observed
relationships between personality (indexed by DSI factors), subjective state (indexed by the DSSQ), and
driver performance indices were generally consistent with prediction from the transactional model of
driver stress (Matthews, 2001, 2002). Several DSI factors were found to be significant predictors of post-
task driver mood and driver performance. Task engagement was found to relate to superior performance
on the secondary hazard monitoring task and to improved vehicle control (as indexed by steering
variability). Worry did not significantly impact driver performance in this experiment.
Next, the effects of automation, drive difficulty, and workload transitions, and subjective state on
driver performance and perceived workload will be discussed in greater detail. The role of individual
difference factors and the human factors implications of the results will also be discussed.
57
Driver performance. The observed effects of automation in this experiment were generally
consistent with a resource-based model of driver performance. Initially, vehicle automation in the
continuous-automation condition improved driver performance on the pedestrian monitoring task, and
improved detection reaction time in the more difficult curved roadway condition. However, these effects
were transient, as performance benefits declined across the experimental drive. The decrement in
performance on the pedestrian task may have represented an inability of participants to maintain effective
resource allocation or the loss of resources necessary for task performance. These results support
predictions from resource theory in that declines in task performance over time are often cited as evidence
of resource shortfalls on tasks requiring sustained attention (e.g., Parasuraman, Warm, & Dember, 1987).
Performance on the pedestrian task in the intermittent-automation condition did not improve
under automated vehicle control, which may also be interpreted as supporting a resource-based model.
Drivers in the intermittent-automation condition may have expended significant resources during periods
of manual task performance, resulting in resource shortfalls during intervals of automated vehicle control.
Similar performance effects have been observed in vigilance research following an abrupt decline in task
demands (e.g., Gluckman et al., 1993; Ungar et al., 2005). Such shortfalls may have reduced or eliminated
any positive effects that automated vehicle control would have had on driver performance relative to
manual control.
Automated vehicle control did improve driver performance in the areas it directly influenced. The
automation employed in this experiment operated perfectly, in that the system was 100% reliable, and
made no errors in lateral and longitudinal control (i.e., the vehicle did not deviate from the center of the
driver’s lane, and vehicle speed was a constant 35 mph). These results tend to support one of the chief
reasons for automating driver tasks: improved driver safety (e.g., Stanton & Marsden, 1996; Walker et al.,
2001). To the extent that future vehicle automation systems can perform similarly to the automation in the
current experiment, it may be expected to correspondingly improve driver safety, though this may be
Driver workload. Automated control in this experiment did not achieve another key reason cited
for the introduction of automation technologies into vehicles: reduced driver workload (e.g., Stanton &
Marsden, 1996; Walker et al., 2001). Participants in the continuous-automation condition reported similar
levels of workload to that of the manual condition. In addition, participants in the intermittent-automation
condition reported higher workload levels than either of the other two vehicle-automation conditions.
These results support the assertion that automation may not reduce mental workload, but may instead
allocate it to other tasks (Parasuraman, Mouloua, Molloy, & Hilburn, 1996; Reinartz & Gruppe, 1993;
Two potential explanations may account for the workload levels reported by participants across
vehicle-automation conditions. First, the observed similarities in workload between the manual and
continuous-automation conditions may be due to drivers’ level of skill. Young and Stanton (2007) argued
that the effects of automation on driver workload may be moderated by drivers’ skills and experiences,
with diminishing benefits as drivers approach automaticity. Young and Stanton found that a sample of
‘expert’ drivers, selected on the basis of their years of licensure, experienced no reduction in workload
associated with automated vehicle control. Young and Stanton argued that the expert drivers in their
sample had sufficient task experience such that lateral and longitudinal vehicle control contributed little to
their estimates of workload. Though the mean number of years as a licensed driver was fewer for
participants in the current experiment compared to the sample of expert drivers recruited by Young and
Stanton (their sample mean was approximately seven years), participants’ estimated mean number of
miles driven annually in the current experiment exceeded the rate reported by Young and Stanton’s expert
drivers (their sample reported driving approximately 5500 miles per year). The net result may be that
drivers in the current experiment had developed their skills sufficiently that little of their reported
workload was due to vehicular control, a situation analogous to that reported by Young and Stanton
(2007). Instead, participants’ ratings of workload in the current experiment may reflect aspects of driving
that were unrelated to vehicle control, such as monitoring the roadway for pedestrian hazards. As noted
by Young and Stanton (2007; see also, Harms, 1991; Groeger & Clegg, 1997), driving tasks such as
59
hazard monitoring may be too variable for automaticity to develop, and so these elements continue to
require conscious, effortful control. In other words, the observed similarities in workload between
vehicle-automation conditions may reflect the skill of drivers in the current experiment, coupled with
aspects of the driving task that were analogous across conditions, such as hazard monitoring.
the manual and continuous-automation conditions, may indicate that abrupt shifts in task demands
required drivers to dynamically recruit or reallocate resources associated with task performance (Huey &
Wickens, 1993). As driver performance was similar in the manual and intermittent-automation conditions
(e.g., on indices of manual driving performance, error rates, correct pedestrian detections, etc.), the
increased workload experienced by participants in the intermittent-automation condition may also reflect
greater effort required to maintain task performance across control transitions. Drivers may have adopted
task strategies that were suitable under manual vehicle control, but which were insufficient or
inappropriate under automated control, and vice versa. Following a transition in vehicle control, drivers
would then be required to shift strategies, and all associated processes (e.g., attention, task directed effort,
scanning patterns, etc.), to continue task performance. This effect may have been compounded by the
frequency and short duration of shifts in vehicular control experienced by participants in the intermittent-
Drive Difficulty
Driver performance. Driver performance under manual control in the current experiment was
substantially poorer in the curved roadway condition compared to the straight condition. Participants
exhibited impaired control on several indices of performance (e.g., variation in lateral position,
longitudinal speed, and steering wheel inputs, increased lane deviations, etc.), suggesting that exposure to
the curved condition may have affected some general resource, rather than a single critical process. In
addition, participants in the curved roadway condition initially drove at slower speeds, which may
indicate that drivers attempted to compensate for increased task difficulty by driving slowly, a relatively
60
frequently observed strategy in the driving research literature (e.g., Horrey & Simons, 2007). Overall,
these results support the growing body of research literature linking difficult driving environments and
poor driver performance (see Kantowitz & Simsek, 2001, for a partial review of the literature). These
results also tend to support a resource-based account of driver performance as decrements in task
performance with increments in task difficulty have been cited as indicants of capacity limitations (e.g.,
Driver workload. Though not statistically significant, an observed trend within the data suggested
that workload levels were greater in the curved roadway condition compared to the straight condition.
That there was not a larger difference in workload ratings between roadway conditions, given the
First, as suggested by Young and Stanton (2007), the sample of drivers in the current experiment
may have been sufficiently skilled that differences in drive difficulty did not influence their estimates of
workload. In other words, drivers in the current experiment may have been partially insulated against
increased workload in the curved roadway condition by their well developed driving skills. This
explanation is somewhat less than satisfactory, however, in that driver skill did not prevent drivers from
Alternatively, as suggested by Singleton (1989), the observed similarities in workload may be due
to a shift or breakdown in task strategy for participants in the curved roadway condition. Participants may
have adopted (or allowed, in the sense of a strategy breakdown) a task strategy that minimized workload
by accepting increased variability, and consequently error, in vehicle control. This may represent a
questionable, but legitimate task strategy in the context of the current experiment. During the
experimental drive, other traffic was not present on the roadway, nor were law enforcement officers.
Therefore, the potential consequences of poor vehicle control may have been diminished, and
concomitantly the incentive to adopt such a strategy, as a form of workload regulation, was increased. In
61
other words, given the traffic conditions in the experimental drive, the benefits of workload reduction
Driver performance. The effects of workload transitions on driver performance in the current
experiment were more complex than initially hypothesized. Transition effects were more evident in the
curved roadway condition, exhibited in the tendency for drivers in the continuous- and intermittent-
automation conditions to commit lane deviations following a transition, and in the increased variability in
lateral position displayed by drivers in the intermittent-automation condition (though only during the
transition from period 2 to period 3). Across roadway conditions, participants in the continuous-
automation condition also exhibited increased variability in steering. More generally, the effects of
workload transitions observed in the current experiment were related to lateral, rather than longitudinal,
vehicle control, an effect that is consistent with the assertions of Young and Stanton (2002a, 2004, 2007)
that lateral control is more demanding than longitudinal control. In addition, consistent with effects
reported by Scallen et al. (1995), the duration of transition-related control disruptions displayed in the
However, the observed effects of workload transitions on driver performance were somewhat
inconsistent across vehicle-automation conditions, and across transition instances. Drivers in the
drivers in the intermittent-automation condition displayed increased variability in lateral position. These
condition experienced higher workload than did participants in the continuous-automation condition. The
relatively long period of single-task performance under automated vehicle control in the continuous-
automation condition may have allowed participants to successfully recover resources (Gluckman et al.,
1993) or to complete resource reallocation (Huey & Wickens, 1993). Following transition from
62
condition may have allowed participants to make a more effortful attempt at maintaining vehicle control
during the transition (i.e., increased steering inputs to preserve lateral control), compared to a less
effortful solution in the intermittent-automation condition (i.e., increased lateral variation as a result of
automation condition may be due to a simple learning effect, as further experience with the task allowed
drivers to successfully manage subsequent control transitions. Alternatively, it may represent a shift in
task strategy, as described previously. Participants may have adapted their post-transition performance
strategies to include acceptance of a short-term increase in driver errors, compensated for by maintenance
of variability in lateral position and steering inputs. Determination of which explanation better accounts
for the observed effect may represent an area of fruitful future research.
Overall, the observed effects of workload transitions in the current experiment tend to support a
resource-based account of driver performance. Superficially, these effects may seem to support an effort-
regulation model, in that one possible interpretation of the observed increase in lateral variability and
associated driver errors following a control transition is that drivers misperceived the effort required for
successful task performance. However, a key assertion of the effort-regulation model is that the
misperception of effort required for task performance is inversely related to the perceived task demands
of the situation. In driving, this effect has been observed more frequently in conjunction with less
demanding straight roads, as opposed to relatively more demanding curved roads (Matthews, 1996;
Matthews & Desmond, 2002). However, the negative effects of workload transitions in the current
experiment were observed, virtually exclusively, in the curved roadway condition, consistent with
Finally, one outcome of the current experiment that is of potential concern results from the
number of moving violations observed following workload transitions in the continuous- and intermittent-
automation conditions. The overall number of moving violations observed during this experiment was
63
relatively low (i.e., approximately 8% of participants committed such an error), and nearly equally
divided across vehicle-automation conditions. However, the preponderance of major violations observed
in the continuous- and intermittent-automation conditions occurred in conjunction with a transition from
automated to manual vehicle control. In addition, control transitions were associated with increased
incidents of lane deviations, particularly in the curved roadway condition. This suggests that future
workload transitions associated with the proposed vehicle automation systems outlined previously may
present substantial safety implications for drivers. Furthermore, these risks are likely to be compounded
by difficult roadway conditions, such as curved roads, poor weather conditions, snarled traffic, etc.
Clearly, manufacturers of future vehicle automation technologies will need to carefully consider the
Driver workload. Frequent transitions from manual to automated vehicle control, and vice versa,
appeared to negatively impact driver workload. Explanations for this effect are consistent with previously
Individual Differences
The current experiment addressed two issues related to individual differences – the prediction of
subjective states of stress during vehicle operation, and prediction of performance. Analysis of the
individual differences data indicated that both pre- and post-task worry were significantly correlated with
the DSI factors of aggression and thrill seeking, and pre-task with fatigue proneness. Pre- and post-task
task engagement were significantly correlated with dislike of driving, hazard monitoring, and fatigue
proneness, and post-task with aggression. Pre- and post-task distress were significantly correlated with
dislike of driving and fatigue proneness, and pre-task with thrill seeking. The observed correlations
between the DSI factors and post-task worry, task engagement, and distress are generally consistent with
previous experiments (e.g., Matthews et al., 1997; Matthews & Desmond, 2002). Such results further
reinforce the need to consider these personality factors when conducting driver research.
64
The multiple regressions conducted in this experiment to investigate the additional predictive
power of the DSI personality factors indicated that these factors added significantly to prediction of post-
task state, but they were less predictive of driver performance. DSI aggression was negatively related to
post-task task engagement, and dislike of driving and fatigue proneness were related to increased post-
task distress. These results are broadly consistent with previous experiments (e.g., Matthews et al., 1997;
DSI thrill seeking was related to faster reaction times on the pedestrian monitoring task. The
positive effects of the DSI factor of thrill seeking observed in this experiment are consistent with other
research indicating that this personality factor may actually lead to better on-road performance, in some
respects, although thrill-seeking is also linked to risky driving strategies and accident risk (Matthews et
al., 1997). The negative effects of dislike of driving and fatigue proneness on subjective state found in the
current study have also been established previously (Matthews et al., 1997).
The multiple regressions concerning the predictive power of the DSSQ state factors of driver
performance indicated that post-task task engagement was related to improved performance on the
pedestrian monitoring task and to improved vehicle control (as indexed by reduced steering variability).
These facilitative effects underscore the dependent relationship between task engagement and difficult,
attentionally demanding tasks, which has been demonstrated in studies using various tasks requiring
sustained attention, such as vigilance (Matthews & Davies, 1998). This finding provides further evidence
for the role of resource availability as an influence on performance, provided that the task imposes a high
workload.
Contrary to initial predictions, worry was not found to adversely affect driver performance in the
current experiment. However, this may be due to the overall difficulty of the driving task employed in this
experiment. Worry is comprised of self-focused attention, low self esteem, and task-relevant and task-
irrelevant cognitive interference (Matthews et al., 1999). Participants in all conditions may have been
forced by task demands to remain focused on performing the task, limiting the time and resources they
could devote to internal self-examination. This, in turn, may have resulted in an overall decrease in worry
65
for all participants. This argument is supported by the findings of Matthews, Campbell et al. (2002), who
also found that a highly demanding, time-pressured working memory task induced a substantial decrease
in post-task worry, and by Funke et al. (2007) who found a comparable decrease in worry associated with
Broadly, the results of this experiment support previous research indicating the utility of a
resource-based model of driver performance (e.g., Funke et al., 2007; Kantowitz & Simsek, 2001). The
observed facilitative effects of automated vehicle control in this experiment offer some hope that future
in-vehicle automation of vehicle control will indeed aid driver performance, though the degree of aid
The results of this experiment also highlight the need for governmental organizations responsible
for road safety, such as the Department of Transportation, and manufacturers of automated vehicle
technologies to collaborate on guidelines for acceptable transitions in vehicle control. The negative
impact of workload transitions on driver safety observed in this experiment seems to favor infrequent
transitions which occur under tightly controlled circumstances. In this respect, proposed IVHS systems
may be superior to commercial automation endeavors because governments may regulate the construction
of highway on- and off-ramps. Ideally, exit-ramps designed for transitions from automated to manual
vehicle control would be straight and long, with generous road-shoulders bounded by guard-rails to
Alternatively, it may be preferable for vehicle control transitions to occur when the vehicle is in a
parked state. This may be a superior solution to the problem of control transitions, in that drivers could
resume control when they were adequately prepared, rather than having to adjust to vehicle control while
driving. This solution also favors an IVHS system, in that normal street driving may not provide drivers
with an adequate, and perhaps immediate, stopping location when a control transition is required.
Conversely, highway rest stops may provide ideal locations for control transitions to take place. As
66
automated vehicle technologies become more proliferate, state governments may be motivated to increase
It must also be noted that the automation employed in this experiment was involuntary. That is,
participants had no control over either the type of automation they experienced or when the automation
was activated. As Parasuraman and Riley (1997) point out, deliberations over when to activate automation
may impose additional cognitive workload over and above that already required by a task. In addition,
automated vehicle control reflects only one type and level of automation; findings might be substantially
different for other vehicle functions that may be automated (cf., Young & Stanton, 2004) and for other
This study also illustrates the need to consider the influence of individual difference factors on
driver behavior. This need may be particularly relevant in the domain of driver personnel selection, where
employees must be selected based on their driving competence and on their safety records. The
personality traits measured by the DSI may be particularly pertinent to this selection process, as those
traits represent stable, dispositional characteristics that may substantially impact driver performance,
particularly in stressful driving situations (Matthews, 2001). The usefulness of such measures has already
been demonstrated by studies showing that the DSI is predictive of real-world frequencies of accident
involvement and of convictions for driving offences (e.g. Matthews et al., 1997). By incorporating
measures such as the DSI into employee screening processes, safe candidates may be more easily
selected, while less suitable employees are screened out. Furthermore, the results illustrate the need to
evaluate in-vehicle automated devices in drivers of differing personalities and subjective states. For
example, the association between task engagement and better pedestrian detection implies that fatigued
drivers may be vulnerable to a potentially dangerous loss of awareness of hazards during automated
driving.
In sum, the implications of the current study for driving-focused research are clear. Automated
vehicle control is steadily becoming a viable technology. However, the potential human factors
consequences of these alterations to vehicle control have yet to be thoroughly investigated. Such
67
proliferation without the necessary research to establish its safety may result in accidents, injuries, and
negative economic ramifications for both consumers and manufacturers. Due to the initial, exploratory
nature of the current experiment, and the potential ubiquity of future automation technologies, continued
REFERENCES
Anderson, J.R. (1995). Cognitive Psychology and Its Implications (4th ed.). New York: W.H. Freeman
and Co.
Ashley, S. (2001). Driving the info highway. Scientific American, 285, 52-58.
Cox-Fuenzalida, L.E. (2007). Effect of workload history on task performance. Human Factors, 49, 277-
291.
Cox-Fuenzalida, L.E., Beeler, C., & Sohl, L. (2006). Workload history effects: A comparison of sudden
Desmond, P.A., Hancock, P.A., & Monette, J.L. (1998). Fatigue and automation-induced impairments in
de Waard, D., van der Hulst, M., Hoedemaeker, M., & Brookhuis, K.A. (1999). Driver behavior in an
emergency situation in the automated highway system. Transportation Human Factors, 1, 67-82.
Dzindolet, M.T., Peterson, S.A., Pomranky, R.A., Pierce, L.G., & Beck, H.P. (2003). The role of trust in
Dzindolet, M.T., Pierce, L.G., Beck, H.P., Dawe, L.A., & Anderson, B.W. (2001). Predicting misuse and
Endsley, M.R. (1995). Toward a theory of situation awareness in dynamic systems. Human Factors, 37,
32-64.
Fitts, P.M. (1962). Factors in complex skill training. In R. Glaser (Ed.), Training research and education.
Pittsburgh, PA: University of Pittsburgh Press. [Reprinted, 1990. In M. Venturino (Ed.), Selected
reading in human factors (pp. 275-296. Santa Monica, CA: Human Factors Society.]
Funke, G.J., Matthews, G., Warm, J.S., & Emo, A.K. (2007). Vehicle automation: A remedy for driver
Gluckman, J.P., Warm, J.S., Dember, W.N., & Rosa, R.R. (1993). Demand transitions and sustained
Groeger, J.A., & Clegg, B.A. (1997). Automaticity and driving: Time to change gear? In T. Rothengatter
& E. Carbonell Vaya (Eds.), Traffic and Transport Psychology: Theory and Application (pp.
Hancock, P. A., & Parasuraman, R. (1992). Human factors and safety in the design of intelligent vehicle-
Hancock, P. A., & Warm, J. S. (1989). A dynamic model of stress and sustained attention. Human
Harms, L. (1986). Drivers’ attentional responses to environmental variations: A dual-task real traffic
Harms, L. (1991). Variation in drivers’ cognitive load: Effects of driving through village areas and rural
Hart, S.G., & Bartolussi, M.R. (1984). Pilot errors as a source of workload. Human Factors, 26, 545-556.
Hart, S.G., & Staveland, L.E. (1988). Development of a multi-dimensional workload scale: Results of
empirical and theoretical research. In P.A. Hancock and N. Meshkati (Eds.), Human mental
Helton, W.S., Shaw, T.H., Warm, J.S., Matthews, G., Dember, W.N., & Hancock, P.A. (2004). Demand
Mouloua., & P.A. Hancock (Eds.), Human performance, situation awareness and automation: Current
Hockey, G.R.J. (1993). Cognitive-energetical control mechanisms in the management of work demands
and psychological health. In A.D. Baddeley & L. Weiskrantz (Eds.), Attention: Selection,
Hockey, G.R.J. (1997). Compensatory control in the regulation of human performance under stress and
Horrey, W.J., & Simons, D.J. (2007). Examining cognitive interference and adaptive safety behaviours in
70
Horrey, W.J., & Wickens, C.D. (2004). Driving and side task performance: The effects of display clutter,
Houser, A., Pierowicz, J., & Fuglewicz, D. (2005). Concept of operations and voluntary operational
requirements for lane departure warning system (LDWS) on-board commercial motor vehicles
Administration.
Huey, B.M., & Wickens, C.D. (1993). Workload transition: Implications for individual and team
Intelligent Vehicle Highway Safety Act of 2004, H.R. 4931, 108th Cong. (2004).
Intermodal Surface Transportation Efficiency Act of 1991, H.R. 2950, 102nd Cong. (1991) (enacted).
Kantowitz, B. H., & Simsek, O. (2001). Secondary-task measures of driver workload. In P.A. Hancock &
P.A. Desmond (Eds.), Stress, workload, and fatigue (pp. 133-163). Mahwah, NJ: Erlbaum.
Kirk, R.E. (1995). Experimental design: Procedures for the behavioral sciences (3rd ed.). New York:
Kranz, R. (2006). Not all premium features will go mainstream. Retrieved August 19, 2007, from
http://www.autonews.com/apps/pbcs.dll/article?AID=/20060605/SUB/60601051
Lajunen, T., & Summala, H. (1995). Driving experience, personality and skill and safety-motive
Lazarus, R.S., & Folkman, S. (1984). Stress, appraisal and coping. New York: Springer.
Lee, J.D. (2006). Human factors and ergonomics in automation design. In G. Salvendy (Ed.), Handbook
Lee, J. D., & Moray, N. (1992). Trust, control strategies, and allocation of function in human-machine
Ma, R., & Kaber, D.B. (2007). Situation awareness and driving performance in a simulated navigation
Madhavan, P., Wiegmann, D.A., & Lacson, F.C. (2006). Automation failures on tasks easily performed
Matthews, G. (1996). Individual differences in driver stress and performance. Proceedings of the Human
Matthews, G. (2001). A transactional model of driver stress. In P. A. Hancock & P. A. Desmond (Eds.),
Matthews, G. (2002). Towards a transactional ergonomics for driver stress and fatigue. Theoretical Issues
Matthews, G., Campbell, S.E., Desmond, P.A., Huggins, J., Falconer, S., & Joyner, A (1999). Assessment
of task-induced state change: Stress, fatigue and workload components. In M. Scerbo (Ed.),
Automation technology and human performance: Current research and trends (pp. 199-203).
Matthews, G., Campbell, S.E., Falconer, S., Joyner, L.A., Huggins, J., Gilliland, K., Grier, R., & Warm,
Matthews, G., & Davies, D.R. (1998). Arousal and vigilance: The role of task factors. In R.R. Hoffman,
M.F. Sherrick, & J.S. Warm (Eds.), Viewing psychology as a whole: The integrative science of
Matthews, G., Davies, D.R., Westerman, S.J., & Stammers, R.B. (2000). Human perfomance: Cognition,
stress and individual differences. Hove, England: Psychology Press/Taylor & Francis.
Matthews, G., & Desmond, PA (2002). Task-induced fatigue states and simulated driving performance.
Matthews, G., Desmond, P. A., Joyner, L. A., & Carcary, B. (1997). A comprehensive questionnaire
measure of driver stress and affect. In E. Carbonell-Vaya & J. A. Rothengatter (Eds.), Traffic and
transport psychology: Theory and application (pp. 317-326). Amsterdam: Pergamon Press.
Matthews, G., Dorn, L., & Glendon, A. I. (1991). Personality correlates of driver stress. Personality and
72
Matthews, G., Tsuda, A., Xin, G., & Ozeki, Y. (1999) Individual differences in driver stress vulnerability
May, P., Molloy, R., & Parasuraman, R. (1993). Effects of automation reliability and failure rate on
Maxwell, S. E., & Delaney, H. D. (2004). Designing experiments and analyzing data: A model
Moray, N. (1979). Mental workload: Its theory and measurement. New York: Plenum.
Neuman, T. R. (1992). Roadway geometric design. In J. L. Pline (Ed.), Traffic engineering hand-book,
Nilsson, L. (1995). Safety effects of adaptive cruise control in critical traffic situations. Proceedings of
Norman, D. A., & Bobrow, D. G. (1975). On data-limited and resource-limited processes. Cognitive
Psychology, 7, 44-64.
Owens, D. A., Helmers, G., & Sivak, M. (1993). Intelligent vehicle highway systems: A call for user-
Parasuraman, R. (2000). Designing automation for human use: Empirical studies and quantitative models.
Parasuraman, R., Mouloua, M., & Molloy, R. (1996). Effects of adaptive task allocation on monitoring of
Parasuraman, R., Mouloua, M., Molloy, R., & Hilburn, B (1996). Monitoring of automated systems. In R.
Parasuraman & M. Mouloua (Eds.), Automation and human performance: Theory and
Parasuraman, R., & Riley, V. (1997). Humans and automation: Use, misuse, disuse, and abuse. Human
Parasuraman, R., Sheridan, T.B., & Wickens, C.D. (2000). A model for types and levels of human
interaction with automation. IEEE Transactions on Systems Man and Cybernetics -Part A:
Parasuraman, R., Warm, J.S., & Dember, W.N. (1987). Vigilance: Taxonomy and utility. In L.S. Mark,
J.S. Warm, & R.L. Huston (Eds.), Ergonomics and human factors: Recent research. New York:
Springer Verlag.
Pedhazur, E. J. (1997). Multiple regression in behavioral research. New York: Harcourt College
Publishers.
Reinartz, S. J., & Gruppe, T. R. (April, 1993). Information requirements to support operator-automatic
co-operation. Paper presented at the Human Factors in Nuclear Safety Conference, London.
Scallen, S.F., Hancock, P.A., & Duley, J.A. (1995). Pilot performance and preference for short cycles of
Schneider, W. & Shiffrin, R.M. (1977). Controlled and automatic human information processing: I.
Sheridan, T.B. (1997). Supervisory control. In G. Salvendy (Ed.), Handbook of Human Factors (pp.
Shiffrin, R.M. & Schneider, W. (1977). Controlled and automatic human information processing: II.
Perceptual learning, automatic attending, and a general theory. Psychological Review, 84, 127–
190.
Singleton, W. T. (1989). The mind at work: Psychological ergonomics. Cambridge, UK: Cambridge
University Press.
Stanton, N.A., & Marsden, P. (1996). From fly-by-wire to drive-by-wire: Safety implications of
Stanton, N.A., & Young, M.S. (2000). A proposed psychological model of driving. Theoretical Issues in
Stanton, N.A., & Young, M.S. (2005). Driver behaviour with adaptive cruise control. Ergonomics, 48,
74
1294-1313.
Stanton, N.A., Young, M., & McCaulder, B. (1997). Drive-by-wire: The case of driver workload and
reclaiming control with adaptive cruise control. Safety Science, 27, 149-159.
Stanton, N.A., Young, M.S., Walker, G.H., Turner, H., & Randle, S. (2001). Automating the driver’s
Tsang, P.S., & Wickens, C.D. (1988). The structural constraints and strategic control of resource
Ungar, N.R., Matthews, G., Warm, J.S., Dember, W.N., Thomas, J.K., Finomore, V.S., & Shaw, T.H.
(2005). Proceedings of the Human Factors and Ergonomics Society, 49, 1523-1526.
Vanderwerp, D. (2006). 2007 Lexus LS460: Road tests. Retrieved August 19, 2007, from
http://www.caranddriver.com/roadtests/11601/2007-lexus-ls460.html
Walker, G.H., Stanton, N.A., & Young, M.S. (2001) Where is computing driving cars? International
Welford, A. T. (1978). Mental work-load as a function of demand, capacity, strategy and skill.
Wickens, C., & Hollands, J. (2000). Engineering psychology and human performance. Upper Saddle
Wiener, E. L. (1985). Beyond the sterile cockpit. Human Factors, 27, 75-90.
Wiener, E. L. (1989). Human factors of advanced technology (“glass cockpit”) transport aircraft (Rep.
Wiesenfelder, J. (2006). The Lexus LS460: It really does park itself. Retrieved August 19, 2007, from
http://blogs.cars.com/kickingtires/2006/09/ls_460_parking.html
Young, M.S., & Stanton, N.A. (2002a). Attention and automation: New perspectives on mental underload
Young, M.S., & Stanton, N.A. (2002b). Malleable attentional resources theory: A new explanation for the
Young, M.S., & Stanton, N.A. (2004). Taking the load off: Investigations of how adaptive cruise control
Young, M.S., & Stanton, N.A. (2007). What’s skill got to do with it? Vehicle automation and driver
APPENDIX A
Manual Condition
Read these instructions to the participant before the practice phase of the experiment:
“This is an experiment designed to investigate the effects of new automated technologies on
driver performance. It is important that you drive normally, as you would in everyday life. Please observe
the 35 mph speed limit throughout all phases of the experiment. This is not a video game, and you should
not treat it as such.
This practice phase of the experiment is 15 minutes long and is broken into 2 parts. In the first
part of the practice, you will simply be learning to control the simulator. If a crash occurs while you are
driving, the simulator will position your car back on the road and you will be able to resume driving.
During the second part of the practice, you will be given the opportunity to practice a hazard
monitoring task which you will be required to perform later. In this part of the practice phase, pedestrians
will be lined up on both sides of the roadway. Occasionally, one of these pedestrians may begin to walk
into the roadway. Your task is to watch for these potential road hazards and indicate with the paddles on
the wheel [BE SURE TO SHOW THE PARTICIPANTS THE PADDLES NOW] when you observe a
moving pedestrian by depressing the right paddle for a moving pedestrian on the right side of the road, or
by depressing the left paddle for a moving pedestrian on the left side of the road. Make sure that you fully
depress the paddle when you observe a moving pedestrian or the simulator may not record your response.
The appearance of pedestrians in the practice phase will be signaled by a series of orange barrels
appearing on the right and left sides of the roadway. When you observe orange barrels on the side of the
road, get ready to begin the hazard monitoring task.
You may ask the experimenter any questions during the practice run. However, the experimenter
will not be able to answer your questions one the test phase of the drive has started.
Are there any questions?”
Read these instructions to the participant before the main phase of the experiment:
“In this phase of the experiment, you will engage in the hazard monitoring task that you practiced
earlier. Again, your task is to watch for pedestrians walking into the roadway and indicate with the wheel
paddles when you observe a moving pedestrian by depressing the right paddle for a moving pedestrian on
the right side of the road, or by depressing the left paddle for a moving pedestrian on the left side of the
road.. Make sure that you fully depress the paddle when you observe a moving pedestrian or the simulator
may not record your response. Unlike the practice phase of the experiment, there will be no orange barrels
to warn you about the appearance of the pedestrians, so you will need to watch for their appearance. If a
crash occurs while you are driving, the simulator will position your car back on the road and you will be
able to resume driving.
You must maintain vehicle control as much as possible during this phase of the experiment. Good
vehicle control includes maintaining the 35 mph speed limit and detecting the moving pedestrian
throughout this phase of the experiment. It is essential to maintain your speed even if you are finding it
difficult to detect the moving pedestrians. Your performance during this phase of the experiment will be
evaluated based on your adherence to all of these requirements.
Please ask any questions you may have now. The experimenter will be unable to answer your
questions once the drive begins.
Are there any questions?”
the 35 mph speed limit throughout all phases of the experiment. This is not a video game, and you should
not treat it as such.
This practice phase of the experiment is 15 minutes long and is broken into 3 parts. In the first
part of the practice, you will simply be learning to control the simulator. If a crash occurs while you are
driving, the simulator will position your car back on the road and you will be able to resume driving.
During the second part of the drive, you will be given the opportunity to practice driving in a
simulation of an experimental, fully automated vehicle. The automation will automatically control your
vehicle’s speed and heading. Because of this, the gas and brake pedals and the steering wheel of the
simulator will not function. When you observe a green star on both sides of the roadway, this is a signal
that the automation is taking control of the vehicle. The automation is not perfect, however, and it may
cease functioning unexpectedly. If this occurs, you will need to reestablish control of the vehicle.
The third part of the practice drive will begin when you observe a red star on both sides of the
roadway. This is a signal that the vehicle will cease automation functions, and return control of the
vehicle to you. You will then be given the opportunity to practice a hazard monitoring task which you
will be required to perform later. In this part of the practice phase, pedestrians will be lined up on both
sides of the roadway. Occasionally, one of these pedestrians may begin to walk into the roadway. Your
task is to watch for these potential road hazards and indicate with the paddles on the wheel [BE SURE TO
SHOW THE PARTICIPANTS THE PADDLES NOW] when you observe a moving pedestrian by
depressing the right paddle for a moving pedestrian on the right side of the road, or by depressing the left
paddle for a moving pedestrian on the left side of the road. Make sure that you fully depress the paddle
when you observe a moving pedestrian or the simulator may not record your response. The appearance of
pedestrians in the practice phase will be signaled by a series of orange barrels appearing on the right and
left sides of the roadway. When you observe orange barrels on the side of the road, get ready to begin the
hazard monitoring task.
You may ask the experimenter any questions during the practice run. However, the experimenter
will not be able to answer your questions one the test phase of the drive has started.
Are there any questions?”
Read these instructions to the participant before the main phase of the experiment:
“In this phase of the experiment, the automation will again periodically control the speed and
heading of your vehicle. When you observe a green star on both sides of the roadway, this is a signal that
the automation is taking control of your vehicle. When you observe a red star on both sides of the
roadway, this is a signal that the automation is returning control of the vehicle to you. Please remember
that the automation is not perfect, and that you may need to reestablish control of the vehicle if the
automation ceases functioning.
In this phase of the experiment, you will engage in the hazard monitoring task that you practiced
earlier. Again, your task is to watch for pedestrians walking into the roadway and indicate with the wheel
paddles when you observe a moving pedestrian by depressing the right paddle for a moving pedestrian on
the right side of the road, or by depressing the left paddle for a moving pedestrian on the left side of the
road.. Make sure that you fully depress the paddle when you observe a moving pedestrian or the simulator
may not record your response. You will engage in the pedestrian monitoring task throughout this phase of
the experiment, including during times when the automation is controlling your vehicle. Unlike the
practice phase of the experiment, there will be no orange barrels to warn you about the appearance of the
pedestrians, so you will need to watch for their appearance. If a crash occurs while you are driving, the
simulator will position your car back on the road and you will be able to resume driving.
You must maintain vehicle control as much as possible as you transition from automated control
to manual control, from manual control to automated control, and as you complete the hazard monitoring
task. Good vehicle control includes maintaining the 35 mph speed limit and detecting the moving
pedestrian throughout this phase of the experiment. It is essential to maintain your speed even if you are
finding it difficult to detect the moving pedestrians. Your performance during this phase of the
experiment will be evaluated based on your adherence to all of these requirements.
78
Please ask any questions you may have now. The experimenter will be unable to answer your
questions once the drive begins.
Are there any questions?”
79
APPENDIX B
Drive Details
• Roadway characteristics
• Period 1
o Speed limit 35 mph (posted every 3050 ft)
o 2 lane road, each lane 12 ft wide, double yellow line
o Shoulder in both lanes 7 ft wide
o Curves every 1120 ft
o 11 ‘normal’ curves, 300 ft in length
o 4 ‘S’ curves, 600 ft in length
• Period 2
o Speed limit 35 mph (posted every 3050 ft)
o 2 lane road, each lane 12 ft wide, double yellow line
o Shoulder in both lanes 7 ft wide
o Curves every 1120 ft
o 6 ‘normal’ curves, 300 ft in length
o 4 ‘S’ curves, 600 ft in length
• Period 3
o Speed limit 35 mph (not posted)
o 2 lane road, each lane 12 ft wide, double yellow line
o Shoulder in both lanes 7 ft wide
o No curved sections
Drive Features
• Period 1
o Buildings and trees are present at both sides of the roadway. The arrangement is
similar to a small city setting.
o No traffic in the period (oncoming or in drivers’ lane)
• Period 2
o Buildings and trees are present at both sides of the roadway. The arrangement is
similar to a small city setting.
o No traffic in the period (oncoming or in drivers’ lane)
o In the continuous-automation and intermittent-automation conditions:
At 17981 ft, star warning polygons appear 500 ft ahead of driver to signal the
beginning of automated vehicle control.
Automated vehicle control from 18481-30600 ft.
At 30100 ft, star warning polygons appear 500 ft ahead of driver to signal the
end of automated vehicle control.
• Period 3
o No traffic in the period (oncoming or in drivers’ lane)
80
o Orange construction barrels appear (30688-30698 ft) on both sides of the roadway to
alert participants that the third phase is about to begin. Warning barrels consist of
three barrels on each side. Each barrel is separated by 5 ft.
o No buildings appear in Period 3.
o Pedestrian pairs begin to appear on the shoulders of the roadway (30801 ft).
Pedestrians are stationed 7 ft from the roadway. Pedestrian pairs are stationed 200 ft
apart. At 35 mph the pedestrian pairs constituted an event rate of 15 pairs/minute.
Pedestrian Set 1 (30801-33788 ft). This set of pedestrians was designed to
demonstrate the appearance of the moving pedestrians. This set lasted
approximately 1 minute. During Set 1, there were 12 critical signals (moving
pedestrians, explained below), balanced for side of roadway.
o Transition segment, 1520 ft in length (approximately 30 seconds long at 35 mph).
o Orange construction barrels appear (35308-35318 ft) on both sides of the roadway to
alert participants that the Pedestrian Set 2 is about to begin. Warning barrels consist
of three barrels on each side. Each barrel is separated by 5 ft.
Pedestrian Set 2 (35421-47678 ft). This set of pedestrians was designed to
give participants practice with the secondary hazard monitoring task under
conditions that were identical to the experimental drive. This set lasted
approximately 4 minutes. During Set 2, there were 8 critical signals, with an
event rate of 2 critical signals/minute. Critical signals were balanced for side
of roadway.
Drive Characteristics
• During Period 3 of the driving task, participants’ secondary task was to monitor the roadway
for potential road hazards. These hazards took the form of pedestrians walking into the
roadway. Positioned on both sides of the roadway, on the shoulder, were the pedestrian pairs
mentioned previously. Occasionally, one of the pedestrians in a pair would begin to walk
toward the roadway (either the right pedestrian, or the left pedestrian in a pair, but not both
simultaneously). It is important to note that these pedestrians were programmed to stop
walking before they crossed into the roadway. So, the pedestrians were potential road
hazards, which participants did not actually have to avoid. Instead, their task was to observe
the moving pedestrians, and indicate that they had observed them by indicating which side of
the roadway the pedestrian was on, using a set of wheel paddles mounted underneath the
steering wheel. Participants were to depress the right paddle for pedestrians on the right side
of the road, and the left paddle for pedestrians on the left side of the road.
• Pedestrians walked at a speed of .2 ft/sec. Through pilot testing, it was determined that
participants’ detection rate of the critical signals was approximately 65%.
• Participants had 5.5 seconds from their appearance to detect moving pedestrians (5 sec before
reaching the pedestrian, and .5 sec as they passed the participant).
Drive Details
• Roadway characteristics
• Straight roadway condition
o 2 lane road, each lane 12 ft wide, double yellow line
o Shoulder in both lanes 7 ft wide
• Curved roadway condition
o 2 lane road, each lane 12 ft wide, double yellow line
o Shoulder in both lanes 7 ft wide
o Curves every 600 ft
264 ‘S’ curves, 600 ft in length
Drive Features
• All periods
o Buildings and trees are present at both sides of the roadway. The arrangement is
similar to a small city setting.
o No traffic in the experimental drive (oncoming or in drivers’ lane)
o Pedestrian monitoring task begins at 3580 ft (approximately 1 minute into the
experimental drive at 35 mph). As in Period 3 of the Practice Drive, pedestrians are
stationed 7 ft from the roadway. Pedestrian pairs are stationed 200 ft apart. At 35
mph the pedestrian pairs had an event rate of 15 pairs/minute.
o During the experimental drive there were 98 critical signals. These signals had an
overall event rate of 14 critical signals/period, 2 critical signals/minute. Critical
signals were balanced for side of roadway.
o As in Period 3 of the Practice Phase, participants had 5.5 seconds to detect the
moving pedestrians (5 sec before reaching the pedestrian, and .5 sec while passing
the pedestrian).
• Period 1
o In the continuous-automation condition:
At 2830 ft, star warning polygons appear 500 ft ahead of driver to signal the
beginning of automated vehicle control.
Automated vehicle control begins at 3330 ft.
• Period 2
o In the intermittent-automation condition:
At 25140 ft, star warning polygons appear 500 ft ahead of driver to signal the
beginning of automated vehicle control.
Automated vehicle control from 25640-30600 ft.
At 46700 ft, star warning polygons appear 500 ft ahead of driver to signal the
end of automated vehicle control.
• Period 4
o In the intermittent-automation condition:
At 69260 ft, star warning polygons appear 500 ft ahead of driver to signal the
beginning of automated vehicle control.
Automated vehicle control from 69760-91320 ft.
At 90820 ft, star warning polygons appear 500 ft ahead of driver to signal the
end of automated vehicle control.
• Period 6
o In the intermittent-automation condition:
82
At 113380 ft, star warning polygons appear 500 ft ahead of driver to signal
the beginning of automated vehicle control.
Automated vehicle control from 113880-135440 ft.
At 134940 ft, star warning polygons appear 500 ft ahead of driver to signal
the end of automated vehicle control.
• Period 7
o In the continuous-automation condition:
At 157500 ft, star warning polygons appear 500 ft ahead of driver to signal
the end of automated vehicle control.
Automated vehicle control ends at 158000 ft.
Drive Characteristics
• Participants engaged in the pedestrian monitoring task throughout the experimental drive.
Task details are described previously in Drive Features.
83
APPENDIX C
ANOVA Tables
Note. All subsequently presented df and p values in Appendix C are uncorrected. Differences between
values reported here and in Chapter 3: Results are due to corrections controlling alpha error rates using
the Dunn-Sidak alpha correct, and corrections for violations of the sphericity assumption using Box’s
epsilon.
Table C1
Analysis of Variance of Mean Lateral Position
Source df F p
Between Groups
Roadway 1 0.24 .624
Vehicle-automation 1 3.31 .073
Roadway × Vehicle-automation 1 0.05 .821
Error 76 (1.077)
Within Groups
Period 3 1.03 .379
Roadway × Period 3 1.85 .138
Vehicle-automation × Period 3 0.29 .835
Roadway × Vehicle-automation × Period 3 0.76 .517
Error 228 (.122)
Note. Values enclosed in parentheses are mean square errors.
Table C2
Analysis of Variance of Mean SD of Lateral Position
Source df F p
Between Groups
Roadway 1 112.75 .001
Vehicle-automation 1 1.15 .286
Roadway × Vehicle-automation 1 0.91 .344
Error 76 (.481)
Within Groups
Period 3 8.65 .001
Roadway × Period 3 0.97 .408
Vehicle-automation × Period 3 1.65 .178
Roadway × Vehicle-automation × Period 3 1.69 .171
Error 228 (.029)
Note. Values enclosed in parentheses are mean square errors.
Table C3
Analysis of Variance of Mean Longitudinal Speed
Source df F p
Between Groups
Roadway 1 1.43 .236
84
Table C3 (continued)
Source df F p
Vehicle-automation 1 0.08 .779
Roadway × Vehicle-automation 1 1.18 .281
Error 76 (35.473)
Within Groups
Period 3 13.55 .001
Roadway × Period 3 4.17 .007
Vehicle-automation × Period 3 1.81 .145
Roadway × Vehicle-automation × Period 3 1.02 .386
Error 228 (2.864)
Note. Values enclosed in parentheses are mean square errors.
Table C4
Analysis of Variance of Mean Longitudinal Speed in the Straight Roadway Condition
Source df F p
Period 3 2.07 .108
Error 117 (3.355)
Note. Values enclosed in parentheses are mean square errors.
Table C5
Analysis of Variance of Mean Longitudinal Speed in the Curved Roadway Condition
Source df F p
Period 3 17.99 .001
Error 117 (2.434)
Note. Values enclosed in parentheses are mean square errors.
Table C6
Analysis of Variance of Mean SD of Longitudinal Speed
Source df F p
Between Groups
Roadway 1 5.44 .022
Vehicle-automation 1 1.85 .178
Roadway × Vehicle-automation 1 0.09 .768
Error 76 (1.413)
Within Groups
Period 3 4.76 .003
Roadway × Period 3 2.29 .079
Vehicle-automation × Period 3 1.22 .304
Roadway × Vehicle-automation × Period 3 1.24 .296
Error 228 (.195)
Note. Values enclosed in parentheses are mean square errors.
85
Table C7
Analysis of Variance of Mean Steering Wheel Rate
Source df F p
Between Groups
Roadway 1 790.71 .001
Vehicle-automation 1 0.18 .674
Roadway × Vehicle-automation 1 0.85 .360
Error 76 (.002)
Within Groups
Period 3 4.01 .008
Roadway × Period 3 2.90 .036
Vehicle-automation × Period 3 0.57 .635
Roadway × Vehicle-automation × Period 3 0.86 .463
Error 228 (.000)
Note. Values enclosed in parentheses are mean square errors.
Table C8
Analysis of Variance of Mean SD of Steering Wheel Rate
Source df F p
Between Groups
Roadway 1 621.08 .001
Vehicle-automation 1 0.006 .941
Roadway × Vehicle-automation 1 0.824 .367
Error 76 (.004)
Within Groups
Period 3 15.73 .001
Roadway × Period 3 15.93 .001
Vehicle-automation × Period 3 1.39 .247
Roadway × Vehicle-automation × Period 3 0.65 .582
Error 228 (.000)
Note. Values enclosed in parentheses are mean square errors.
Table C9
Analysis of Variance of Mean Longitudinal Speed in the Curved Roadway Condition
Source df F p
Period 3 9.62 .001
Error 117 (.000)
Note. Values enclosed in parentheses are mean square errors.
Table C10
Analysis of Variance of Mean Longitudinal Speed in the Curved Roadway Condition
Source df F p
Period 3 16.93 .001
Error 117 (.000)
Note. Values enclosed in parentheses are mean square errors.
86
Table C11
Analysis of Variance of Percentage of Correct Pedestrian Detections
Source df F p
Between Groups
Roadway 1 2.86 .094
Vehicle-automation 2 3.19 .045
Roadway × Vehicle-automation 2 0.58 .559
Error 114 (49.03)
Within Groups
Period 6 3.87 .001
Roadway × Period 6 4.82 .001
Vehicle-automation × Period 12 5.29 .001
Roadway × Vehicle-automation × Period 12 1.57 .096
Error 684 (4.60)
Note. Values enclosed in parentheses are mean square errors.
Table C12
Analysis of Variance of Percentage of Correct Pedestrian Detections in the Straight Roadway
Condition
Source df F p
Period 6 5.91 .001
Error 354 (4.17)
Note. Values enclosed in parentheses are mean square errors.
Table C13
Analysis of Variance of Percentage of Correct Pedestrian Detections in the Curved Roadway
Condition
Source df F p
Period 6 2.64 .016
Error 354 (5.79)
Note. Values enclosed in parentheses are mean square errors.
Table C14
Analysis of Variance of Percentage of Correct Pedestrian Detections in the Manual Vehicle-
Automation Condition
Source df F p
Period 6 0.628 .708
Error 234 (4.66)
Note. Values enclosed in parentheses are mean square errors.
87
Table C15
Analysis of Variance of Percentage of Correct Pedestrian Detections in the Continuous-
Automation Vehicle-Automation Condition
Source df F p
Period 6 11.38 .001
Error 234 (3.80)
Note. Values enclosed in parentheses are mean square errors.
Table C16
Analysis of Variance of Percentage of Correct Pedestrian Detections in the Intermittent-
Automation Vehicle-Automation Condition
Source df F p
Period 6 3.42 .003
Error 234 (5.92)
Note. Values enclosed in parentheses are mean square errors.
Table C17
Analysis of Variance of Mean Reaction Times for Correct Pedestrian Detections
Source df F p
Between Groups
Roadway 1 3.92 .05
Vehicle-automation 2 2.45 .091
Roadway × Vehicle-automation 2 0.03 .969
Error 108 (.202)
Within Groups
Period 6 5.43 .001
Roadway × Period 6 1.29 .260
Vehicle-automation × Period 12 2.25 .009
Roadway × Vehicle-automation × Period 12 2.12 .014
Error 648 (.036)
Note. Values enclosed in parentheses are mean square errors.
Table C18
Analysis of Variance of Mean Reaction Times for Correct Pedestrian Detections in the Straight
Roadway Condition
Source df F p
Between Groups
Vehicle-automation 2 1.38 .260
Error 54 (.176)
Within Groups
Period 6 3.08 .006
Vehicle-automation × Period 12 2.25 .154
Error 324 (.042)
Note. Values enclosed in parentheses are mean square errors.
88
Table C19
Analysis of Variance of Mean Reaction Times for Correct Pedestrian Detections in the Curved
Roadway Condition
Source df F p
Between Groups
Vehicle-automation 2 1.13 .332
Error 54 (.227)
Within Groups
Period 6 3.75 .001
Vehicle-automation × Period 12 3.3 .001
Error 324 (.031)
Note. Values enclosed in parentheses are mean square errors.
Table C20
Analysis of Variance of Mean Reaction Times for Correct Pedestrian Detections in the Curved
Roadway Manual Vehicle-Automation Condition
Source df F p
Period 6 2.94 .010
Error 114 (.034)
Note. Values enclosed in parentheses are mean square errors.
Table C21
Analysis of Variance of Mean Reaction Times for Correct Pedestrian Detections in the Curved
Roadway Continuous-Automation Vehicle-Automation Condition
Source df F p
Period 6 12.19 .001
Error 114 (.012)
Note. Values enclosed in parentheses are mean square errors.
Table C22
Analysis of Variance of Mean Reaction Times for Correct Pedestrian Detections in the Curved
Roadway Intermittent-Automation Vehicle-Automation Condition
Source df F p
Period 6 1.40 .224
Error 96 (.049)
Note. Values enclosed in parentheses are mean square errors.
Table C23
Analysis of Variance of Mean SD of Lateral Position for the Transition from Period 7 to Post-
task Baseline
Source df F p
Between Groups
Roadway 1 13.93 .001
Vehicle-automation 1 1.81 .183
89
Table C24
Analysis of Variance of Mean SD of Lateral Position for the Transition from Period 7 to Post-
task Baseline in the Straight Roadway Condition
Source df F p
Between Groups
Vehicle-automation 1 0.06 .807
Error 35 (.325)
Within Groups
Segment 4 1180.40 .001
Vehicle-automation × Segment 4 1.95 .105
Error 140 (.164)
Note. Values enclosed in parentheses are mean square errors.
Table C25
Analysis of Variance of Mean SD of Lateral Position for the Transition from Period 7 to Post-
task Baseline in the Curved Roadway Condition
Source df F p
Between Groups
Vehicle-automation 1 4.01 .053
Error 35 (.424)
Within Groups
Segment 4 334.65 .001
Vehicle-automation × Segment 4 2.90 .024
Error 280 (.486)
Note. Values enclosed in parentheses are mean square errors.
Table C26
Analysis of Variance of Mean SD of Longitudinal Speed for the Transition from Period 7 to
Post-task Baseline
Source df F p
Between Groups
Roadway 1 5.50 .022
Vehicle-automation 1 0.48 .493
90
Table C27
Analysis of Variance of Mean SD of Steering Wheel Rate for the Transition from Period 7 to
Post-task Baseline
Source df F p
Between Groups
Roadway 1 26.19 .001
Vehicle-automation 1 0.77 .382
Roadway × Vehicle-automation 1 0.57 .452
Error 70 (.002)
Within Groups
Segment 4 39.45 .001
Roadway × Segment 4 27.58 .001
Vehicle-automation × Segment 4 5.32 .001
Roadway × Vehicle-automation × Segment 4 1.48 .208
Error 280 (.002)
Note. Values enclosed in parentheses are mean square errors.
Table C28
Analysis of Variance of Mean SD of Steering Wheel Rate for the Transition from Period 7 to
Post-task Baseline in the Manual Vehicle-Automation Condition
Source df F p
Segment 4 23.99 .001
Error 144 (.001)
Note. Values enclosed in parentheses are mean square errors.
Table C29
Analysis of Variance of Mean SD of Steering Wheel Rate for the Transition from Period 7 to
Post-task Baseline in the Continuous-Automation Vehicle-Automation Condition
Source df F p
Segment 4 15.46 .001
Error 144 (.004)
Note. Values enclosed in parentheses are mean square errors.
91
Table C30
Analysis of Variance of Mean SD of Lateral Position for the Transition from Period 2 to Period 3
Source df F p
Between Groups
Roadway 1 175.14 .001
Vehicle-automation 1 10.89 .001
Roadway × Vehicle-automation 1 8.88 .004
Error 76 (.591)
Within Groups
Segment 4 9.92 .001
Roadway × Segment 4 10.57 .001
Vehicle-automation × Segment 4 5.27 .001
Roadway × Vehicle-automation × Segment 4 4.95 .001
Error 280 (.272)
Note. Values enclosed in parentheses are mean square errors.
Table C31
Analysis of Variance of Mean SD of Lateral Position for the Transition from Period 2 to Period 3
in the Straight Roadway Condition
Source df F p
Between Groups
Vehicle-automation 1 0.11 .741
Error 38 (.275)
Within Groups
Segment 4 2.18 .073
Vehicle-automation × Segment 4 0.13 .973
Error 152 (.081)
Note. Values enclosed in parentheses are mean square errors.
Table C32
Analysis of Variance of Mean SD of Lateral Position for the Transition from Period 2 to Period 3
in the Curved Roadway Condition
Source df F p
Between Groups
Vehicle-automation 1 11.66 .001
Error 38 (.907)
Within Groups
Segment 4 11.64 .001
Vehicle-automation × Segment 4 5.97 .001
Error 152 (.464)
Note. Values enclosed in parentheses are mean square errors.
92
Table C33
Analysis of Variance of Mean SD of Lateral Position for the Transition from Period 2 to Period 3
in the Curved Roadway Manual Vehicle-Automation Condition
Source df F p
Segment 4 7.36 .001
Error 76 (.830)
Note. Values enclosed in parentheses are mean square errors.
Table C34
Analysis of Variance of Mean SD of Lateral Position for the Transition from Period 2 to Period 3
in the Curved Roadway Intermittent-Automation Vehicle-Automation Condition
Source df F p
Segment 4 8.95 .001
Error 76 (.845)
Note. Values enclosed in parentheses are mean square errors.
Table C35
Analysis of Variance of Mean SD of Lateral Position for the Transition from Period 4 to Period 5
Source df F p
Between Groups
Roadway 1 3.41 .069
Vehicle-automation 1 3.00 .087
Roadway × Vehicle-automation 1 3.32 .072
Error 76 (3040.900)
Within Groups
Segment 4 1.36 .249
Roadway × Segment 4 1.28 .276
Vehicle-automation × Segment 4 1.26 .284
Roadway × Vehicle-automation × Segment 4 1.34 .256
Error 304 (1499.450)
Note. Values enclosed in parentheses are mean square errors.
Table C36
Analysis of Variance of Mean SD of Lateral Position for the Transition from Period 6 to Period 7
Source df F p
Between Groups
Roadway 1 23.59 .001
Vehicle-automation 1 0.84 .362
Roadway × Vehicle-automation 1 12.40 .001
Error 76 (1.638)
Within Groups
Segment 4 7.75 .001
Roadway × Segment 4 1.53 .194
Vehicle-automation × Segment 4 2.22 .067
93
Table C37
Analysis of Variance of Mean SD of Longitudinal Speed for the Transition from Period 2 to
Period 3
Source df F p
Between Groups
Roadway 1 16.89 .001
Vehicle-automation 1 4.79 .032
Roadway × Vehicle-automation 1 0.07 .791
Error 76 (.480)
Within Groups
Segment 4 20.88 .001
Roadway × Segment 4 0.78 .539
Vehicle-automation × Segment 4 0.84 .502
Roadway × Vehicle-automation × Segment 4 0.61 .658
Error 304 (.232)
Note. Values enclosed in parentheses are mean square errors.
Table C38
Analysis of Variance of Mean SD of Longitudinal Speed for the Transition from Period 4 to
Period 5
Source df F p
Between Groups
Roadway 1 0.37 .546
Vehicle-automation 1 0.39 .534
Roadway × Vehicle-automation 1 3.89 .052
Error 76 (6031.181)
Within Groups
Segment 4 2.22 .066
Roadway × Segment 4 0.14 .968
Vehicle-automation × Segment 4 0.15 .963
Roadway × Vehicle-automation × Segment 4 2.22 .067
Error 304 (2622.252)
Note. Values enclosed in parentheses are mean square errors.
94
Table C39
Analysis of Variance of Mean SD of Longitudinal Speed for the Transition from Period 6 to
Period 7
Source df F p
Between Groups
Roadway 1 0.05 .828
Vehicle-automation 1 2.73 .103
Roadway × Vehicle-automation 1 0.00 .965
Error 76 (1.845)
Within Groups
Segment 4 15.43 .001
Roadway × Segment 4 0.68 .606
Vehicle-automation × Segment 4 4.40 .002
Roadway × Vehicle-automation × Segment 4 1.26 .285
Error 304 (.421)
Note. Values enclosed in parentheses are mean square errors.
Table C40
Analysis of Variance of Mean SD of Longitudinal Speed for the Transition from Period 6 to
Period 7 in the Manual Vehicle-Automation Condition
Source df F p
Segment 4 12.30 .001
Error 156 (.387)
Note. Values enclosed in parentheses are mean square errors.
Table C41
Analysis of Variance of Mean SD of Longitudinal Speed for the Transition from Period 6 to
Period 7 in the Intermittent-Automation Vehicle-Automation Condition
Source df F p
Segment 4 7.90 .001
Error 156 (.455)
Note. Values enclosed in parentheses are mean square errors.
Table C42
Analysis of Variance of Mean SD of Steering Wheel Rate for the Transition from Period 2 to
Period 3
Source df F p
Between Groups
Roadway 1 844.54 .001
Vehicle-automation 1 0.33 .566
Roadway × Vehicle-automation 1 0.63 .429
Error 76 (.001)
Within Groups
Segment 4 236.68 .001
95
Table C41
Analysis of Variance of Mean SD of Steering Wheel Rate for the Transition from Period 4 to
Period 5
Source df F p
Between Groups
Roadway 1 0.37 .545
Vehicle-automation 1 0.35 .553
Roadway × Vehicle-automation 1 3.88 .053
Error 76 (5645.657)
Within Groups
Segment 4 2.21 .068
Roadway × Segment 4 0.14 .968
Vehicle-automation × Segment 4 0.14 .967
Roadway × Vehicle-automation × Segment 4 2.22 .067
Error 304 (2496.938)
Note. Values enclosed in parentheses are mean square errors.
Table C42
Analysis of Variance of Mean SD of Steering Wheel Rate for the Transition from Period 6 to
Period 7
Source df F p
Between Groups
Roadway 1 886.61 .001
Vehicle-automation 1 0.06 .812
Roadway × Vehicle-automation 1 0.40 .528
Error 75 (.001)
Within Groups
Segment 4 192.16 .001
Roadway × Segment 4 69.96 .001
Vehicle-automation × Segment 4 0.77 .546
Roadway × Vehicle-automation × Segment 4 3.02 .018
Error 300 (.000)
Note. Values enclosed in parentheses are mean square errors.
96
Table C43
Analysis of Variance of Mean SD of Steering Wheel Rate for the Transition from Period 6 to
Period 7 in the Straight Roadway Condition
Source df F p
Between Groups
Vehicle-automation 1 0.18 .674
Error 38 (.001)
Within Groups
Segment 4 52.16 .001
Vehicle-automation × Segment 4 5.36 .001
Error 152 (.000)
Note. Values enclosed in parentheses are mean square errors.
Table C44
Analysis of Variance of Mean SD of Steering Wheel Rate for the Transition from Period 6 to
Period 7 in the Curved Roadway Condition
Source df F p
Between Groups
Vehicle-automation 1 0.24 .628
Error 38 (.002)
Within Groups
Segment 4 143.40 .001
Vehicle-automation × Segment 4 1.19 .317
Error 152 (.001)
Note. Values enclosed in parentheses are mean square errors.
Table C45
Analysis of Variance of Mean SD of Steering Wheel Rate for the Transition from Period 6 to
Period 7 in the Straight Roadway Manual Vehicle-Automation Condition
Source df F p
Segment 4 42.86 .001
Error 76 (.000)
Note. Values enclosed in parentheses are mean square errors.
Table C46
Analysis of Variance of Mean SD of Steering Wheel Rate for the Transition from Period 6 to
Period 7 in the Straight Roadway Intermittent-Automation Vehicle-Automation Condition
Source df F p
Segment 4 19.94 .001
Error 76 (.000)
Note. Values enclosed in parentheses are mean square errors.
97
Table C47
Analysis of Variance of Mean Pre-task DSSQ Worry Scores
Source df F p
Roadway 1 0.12 .730
Vehicle-automation 2 0.36 .697
Roadway × Vehicle-automation 2 0.11 .895
Error 114 (1.056)
Note. Values enclosed in parentheses are mean square errors.
Table C48
Analysis of Variance of Mean Pre-task DSSQ Task Engagement Scores
Source df F p
Roadway 1 1.82 .180
Vehicle-automation 2 0.75 .475
Roadway × Vehicle-automation 2 2.66 .074
Error 114 (.680)
Note. Values enclosed in parentheses are mean square errors.
Table C49
Analysis of Variance of Mean Pre-task DSSQ Distress Scores
Source df F p
Roadway 1 1.69 .196
Vehicle-automation 2 0.02 .979
Roadway × Vehicle-automation 2 0.28 .754
Error 114 (.680)
Note. Values enclosed in parentheses are mean square errors.
Table C50
Analysis of Variance of Mean DSSQ Worry Scores
Source df F p
Between Groups
Roadway 1 0.01 .922
Vehicle-automation 2 0.46 .634
Roadway × Vehicle-automation 2 0.10 .908
Error 114 (1.910)
Within Groups
Pre-post Task 1 9.69 .001
Roadway × Pre-post Task 1 0.46 .500
Vehicle-automation × Pre-post Task 2 0.25 .780
Roadway × Vehicle-automation × Pre-post Task 2 0.07 .929
Error 114 (.295)
Note. Values enclosed in parentheses are mean square errors.
98
Table C51
Analysis of Variance of Mean DSSQ Task Engagement Scores
Source df F p
Between Groups
Roadway 1 5.45 .021
Vehicle-automation 2 0.537 .586
Roadway × Vehicle-automation 2 3.28 .041
Error 114 (1.216)
Within Groups
Pre-post Task 1 218.26 .001
Roadway × Pre-post Task 1 2.84 .095
Vehicle-automation × Pre-post Task 2 2.13 .124
Roadway × Vehicle-automation × Pre-post Task 2 0.73 .483
Error 114 (.352)
Note. Values enclosed in parentheses are mean square errors.
Table C52
Analysis of Variance of Mean DSSQ Distress Scores
Source df F p
Between Groups
Roadway 1 1.34 .250
Vehicle-automation 2 0.49 .613
Roadway × Vehicle-automation 2 0.68 .510
Error 114 (.957)
Within Groups
Pre-post Task 1 116.95 .001
Roadway × Pre-post Task 1 0.30 .587
Vehicle-automation × Pre-post Task 2 0.95 .391
Roadway × Vehicle-automation × Pre-post Task 2 1.15 .319
Error 114 (.318)
Note. Values enclosed in parentheses are mean square errors.
Table C53
Analysis of Variance of Mean TLX Workload Scores
Source df F p
Between Groups
Roadway 1 3.16 .078
Vehicle-automation 2 3.49 .034
Roadway × Vehicle-automation 2 0.34 .715
Error 114 (1110.555)
Within Groups
Subscale 5 43.13 .001
Roadway × Subscale 5 1.33 .249
Vehicle-automation × Subscale 10 0.90 .534
Roadway × Vehicle-automation × Subscale 10 0.79 .639
99
APPENDIX D
Regression Tables
Table D1
Summary of Hierarchical Regression Analysis for Variables Predicting Post-Task Worry
Variable df F for ∆ R2 ∆R2 B SE B β
Step 1 1, 118 135.20 .534 .534
Pre-task worry .762 .065 .731*
Step 2 4, 115 33.47 .538 .004
Pre-task worry .762 .066 .731*
Roadway vector -.046 .066 -.044
Vehicle-automation vector 1 -.056 .094 -.044
Vehicle-automation vector 2 -.003 .094 -.003
Step 3 6, 113 21.97 .538 .000
Pre-task worry .761 .067 .731*
Roadway vector -.046 .067 -.044
Vehicle-automation vector 1 -.056 .095 -.044
Vehicle-automation vector 2 -.003 .095 -.003
Roadway × automation vector 1 -.027 .095 -.021
Roadway × automation vector 2 .027 .095 .021
Step 4 11, 108 11.79 .546 .007
Pre-task worry .801 .080 .769*
Roadway vector -.037 .070 -.035
Vehicle-automation vector 1 -.048 .097 -.037
Vehicle-automation vector 2 -.013 .097 -.010
Roadway × automation vector 1 -.038 .098 -.030
Roadway × automation vector 2 .038 .098 .030
DSI aggression -.003 .005 -.046
DSI dislike of driving .004 .006 .045
DSI hazard monitoring -.001 .006 -.010
DSI thrill seeking .000 .004 -.008
DSI fatigue proneness -.005 .005 -.079
*p < .05
Table D2
Summary of Hierarchical Regression Analysis for Variables Predicting Post-Task Task Engagement
Variable df F for ∆ R2 ∆R2 B SE B β
Step 1 1, 118 53.71 .313 .313
Pre-task task engagement .630 .086 .559*
Step 2 4, 115 15.92 .356 .044
Pre-task task engagement .613 .086 .543*
Roadway vector -.156 .073 -.161*
Vehicle-automation vector 1 .183 .103 .154
Vehicle-automation vector 2 -.109 .104 -.091
Step 3 6, 113 10.99 .369 .012
Pre-task task engagement .596 .088 .528*
101
Table D2 (continued)
Variable df F for ∆ R2 ∆R2 B SE B β
Roadway vector -.159 .073 -.163*
Vehicle-automation vector 1 .182 .103 .153
Vehicle-automation vector 2 -.106 .104 -.089
Roadway × automation vector 1 -.050 .104 -.042
Roadway × automation vector 2 .150 .103 .126
Step 4 11, 108 7.49 .433 .064
Pre-task task engagement .548 .094 .486*
Roadway vector -.124 .073 -.128
Vehicle-automation vector 1 .191 .100 .160
Vehicle-automation vector 2 -.111 .102 -.093
Roadway × automation vector 1 -.085 .102 -.072
Roadway × automation vector 2 .190 .102 .159
DSI aggression -.010 .005 -.167*
DSI dislike of driving -.010 .007 -.133
DSI hazard monitoring .007 .006 .087
DSI thrill seeking -.004 .004 -.086
DSI fatigue proneness .002 .005 .038
*p < .05
Table D3
Summary of Hierarchical Regression Analysis for Variables Predicting Post-Task
Distress
Variable df F for ∆ R2 ∆R2 B SE B β
Step 1 1, 118 39.59 .251 .251
Pre-task distress .517 .082 .501*
Step 2 4, 115 10.44 .266 .015
Pre-task distress .512 .083 .496*
Roadway vector .001 .065 .001
Vehicle-automation vector 1 -.055 .091 -.055
Vehicle-automation vector 2 -.085 .091 -.086
Step 3 6, 113 7.45 .283 .017
Pre-task distress .508 .083 .492*
Roadway vector .001 .065 .001
Vehicle-automation vector 1 -.055 .091 -.055
Vehicle-automation vector 2 -.086 .091 -.086
Roadway × automation vector 1 -.098 .091 -.099
Roadway × automation vector 2 -.049 .091 -.049
Step 4 11, 108 5.64 .365 .081
Pre-task distress .367 .090 .355*
Roadway vector -.037 .064 -.045
Vehicle-automation vector 1 -.053 .088 -.053
Vehicle-automation vector 2 -.081 .089 -.082
Roadway × automation vector 1 -.055 .089 -.056
Roadway × automation vector 2 -.083 .091 -.084
DSI aggression -.001 .005 -.027
DSI dislike of driving .011 .006 .173*
102
Table D3 (continued)
Variable df F for ∆ R2 ∆R2 B SE B β
DSI hazard monitoring .000 .005 .006
DSI thrill seeking .006 .003 .148
DSI fatigue proneness .010 .005 .196*
*p < .05
Table D4
Summary of Hierarchical Regression Analysis for Variables Predicting the Mean Number of Correct Pedestrian
Task Detections
Variable df F for ∆ R2 ∆R2 B SE B β
Step 1 3, 116 3.10 .074 .074
Roadway vector 2.858 1.685 .152
Vehicle-automation vector 1 -4.983 2.383 -.216*
Vehicle-automation vector 2 5.442 2.383 .236*
Step 2 5, 114 2.08 .084 .009
Roadway vector 2.858 1.691 .152
Vehicle-automation vector 1 -4.983 2.392 -.216*
Vehicle-automation vector 2 5.442 2.392 .236*
Roadway × automation vector 1 2.267 2.392 .098
Roadway × automation vector 2 -2.208 2.392 -.096
Step 3 8, 111 2.93 .174 .091
Roadway vector 4.000 1.680 .212*
Vehicle-automation vector 1 -5.883 2.318 -.255*
Vehicle-automation vector 2 5.355 2.316 .232*
Roadway × automation vector 1 1.994 2.307 .086
Roadway × automation vector 2 -3.173 2.339 -.137
Post-task worry -2.460 1.582 -.136
Post-task task engagement 5.242 1.877 .270*
Post-task distress .384 2.157 .016
Step 4 13, 106 2.64 .245 .070
Roadway vector 3.478 1.667 .184*
Vehicle-automation vector 1 -6.408 2.288 -.277*
Vehicle-automation vector 2 5.286 2.289 .229*
Roadway × automation vector 1 2.691 2.284 .116
Roadway × automation vector 2 -4.054 2.359 -.176
Post-task worry -3.308 1.634 -.183*
Post-task task engagement 6.936 1.929 .357*
Post-task distress -.650 2.332 -.028
DSI aggression .184 .118 .155
DSI dislike of driving .198 .153 .129
DSI hazard monitoring -.241 .137 -.159
DSI thrill seeking .093 .089 .107
DSI fatigue proneness -.050 .124 -.041
*p < .05
103
Table D5
Summary of Hierarchical Regression Analysis for Variables Predicting Mean Pedestrian Task Detection Reaction
Time
Variable df F for ∆ R2 ∆R2 B SE B β
Step 1 3, 116 3.00 .072 .072
Roadway vector -.035 .015 -.205*
Vehicle-automation vector 1 -.009 .022 -.044
Vehicle-automation vector 2 .040 .022 .191
Step 2 5, 114 1.78 .072 .000
Roadway vector -.035 .015 -.205*
Vehicle-automation vector 1 -.009 .022 -.044
Vehicle-automation vector 2 .040 .022 .191
Roadway × automation vector 1 -.001 .022 -.004
Roadway × automation vector 2 .005 .022 .024
Step 3 8, 111 1.10 .073 .001
Roadway vector -.035 .016 -.204*
Vehicle-automation vector 1 -.009 .022 -.044
Vehicle-automation vector 2 .041 .022 .194
Roadway × automation vector 1 -.001 .022 -.002
Roadway × automation vector 2 .005 .023 .025
Post-task worry -.001 .015 -.006
Post-task task engagement .002 .018 .011
Post-task distress .007 .021 .032
Step 4 13, 106 1.87 .186 .113
Roadway vector -.031 .016 -.182*
Vehicle-automation vector 1 -.001 .022 -.005
Vehicle-automation vector 2 .037 .022 .177
Roadway × automation vector 1 -.012 .022 -.056
Roadway × automation vector 2 .018 .022 .086
Post-task worry .014 .015 .082
Post-task task engagement -.010 .018 -.055
Post-task distress .015 .022 .071
DSI aggression -.001 .001 -.115
DSI dislike of driving -.001 .001 -.077
DSI hazard monitoring .001 .001 .080
DSI thrill seeking -.002 .001 -.296*
DSI fatigue proneness .001 .001 .083
*p < .05
Table D6
Summary of Hierarchical Regression Analysis for Variables Predicting Mean Lateral
Position
2
Variable df F for ∆ R ∆R2 B SE B β
Step 1 2, 77 2.37 .058 .058
Roadway vector -.003 .051 -.007
Vehicle-automation vector -.112 .051 -.241*
Step 2 3, 76 1.60 .059 .001
Roadway vector -.003 .052 -.007
104
Table D6 (continued)
Variable df F for ∆ R2 ∆R2 B SE B β
Vehicle-automation vector -.112 .052 -.241*
Roadway × automation vector -.016 .052 -.035
Step 3 6, 73 1.16 .087 .028
Roadway vector .010 .056 .021
Vehicle-automation vector -.110 .053 -.236*
Roadway × automation vector -.022 .053 -.048
Post-task worry .077 .053 .165
Post-task task engagement .026 .061 .055
Post-task distress -.005 .066 -.010
Step 4 11, 68 0.76 .109 .022
Roadway vector -.001 .058 -.003
Vehicle-automation vector -.111 .055 -.24*
Roadway × automation vector -.019 .055 -.041
Post-task worry .080 .057 .172
Post-task task engagement .036 .064 .075
Post-task distress -.033 .080 -.059
DSI aggression -.003 .004 -.090
DSI dislike of driving .004 .005 .101
DSI hazard monitoring -.003 .004 -.081
DSI thrill seeking .002 .003 .081
DSI fatigue proneness .001 .005 .037
*p < .05
Table D7
Summary of Hierarchical Regression Analysis for Variables Predicting Mean SD of Lateral Position
Variable df F for ∆ R2 ∆R2 B SE B β
Step 1 2, 77 44.42 .536 .536
Roadway vector -.285 .031 -.725*
Vehicle-automation vector -.040 .031 -.102
Step 2 3, 76 29.65 .539 .004
Roadway vector -.285 .031 -.725*
Vehicle-automation vector -.040 .031 -.102
Roadway × automation vector .024 .031 .060
Step 3 6, 73 16.48 .575 .036
Roadway vector -.313 .032 -.796*
Vehicle-automation vector -.029 .031 -.074
Roadway × automation vector .036 .031 .092
Post-task worry -.003 .031 -.007
Post-task task engagement -.085 .035 -.213*
Post-task distress -.009 .038 -.019
Step 4 11, 68 9.30 .601 .026
Roadway vector -.311 .033 -.791*
Vehicle-automation vector -.033 .031 -.084
Roadway × automation vector .039 .031 .099
Post-task worry -.011 .032 -.028
Post-task task engagement -.070 .037 -.174
105
Table D7 (continued)
Variable df F for ∆ R2 ∆R2 B SE B β
Post-task distress -.006 .045 -.014
DSI aggression .004 .002 .134
DSI dislike of driving .003 .003 .099
DSI hazard monitoring -.002 .003 -.053
DSI thrill seeking .000 .002 .017
DSI fatigue proneness -.002 .003 -.076
*p < .05
Table D8
Summary of Hierarchical Regression Analysis for Variables Predicting Mean Longitudinal
Speed
Variable df F for ∆ R2 ∆R2 B SE B β
Step 1 2, 77 0.47 .012 .012
Roadway vector .273 .286 .108
Vehicle-automation vector -.040 .286 -.016
Step 2 3, 76 0.67 .026 .014
Roadway vector .273 .286 .108
Vehicle-automation vector -.040 .286 -.016
Roadway × automation vector -.298 .286 -.118
Step 3 6, 73 0.87 .067 .041
Roadway vector .278 .307 .110
Vehicle-automation vector .013 .291 .005
Roadway × automation vector -.295 .292 -.117
Post-task worry .506 .291 .199
Post-task task engagement -.049 .333 -.019
Post-task distress .010 .362 .003
Step 4 11, 68 1.21 .163 .097
Roadway vector .275 .305 .109
Vehicle-automation vector -.040 .288 -.016
Roadway × automation vector -.209 .289 -.083
Post-task worry .295 .302 .116
Post-task task engagement .028 .340 .011
Post-task distress -.056 .422 -.019
DSI aggression .014 .022 .077
DSI dislike of driving .039 .027 .195
DSI hazard monitoring -.003 .024 -.017
DSI thrill seeking .034 .015 .288*
DSI fatigue proneness -.012 .024 -.078
*p < .05
Table D9
Summary of Hierarchical Regression Analysis for Variables Predicting Mean SD of Longitudinal
Speed
2
Variable df F for ∆ R ∆R2 B SE B β
Step 1 2, 77 2.99 .072 .072
106
Table D9 (continued)
Variable df F for ∆ R2 ∆R2 B SE B β
Roadway vector -.154 .064 -.263*
Vehicle-automation vector -.031 .064 -.052
Step 2 3, 76 2.12 .077 .005
Roadway vector -.154 .065 -.263*
Vehicle-automation vector -.031 .065 -.052
Roadway × automation vector -.042 .065 -.072
Step 3 6, 73 1.66 .120 .043
Roadway vector -.186 .069 -.318*
Vehicle-automation vector -.020 .066 -.035
Roadway × automation vector -.034 .066 -.058
Post-task worry .023 .066 .038
Post-task task engagement -.125 .075 -.209
Post-task distress -.090 .082 -.128
Step 4 11, 68 1.36 .180 .060
Roadway vector -.168 .070 -.286*
Vehicle-automation vector -.025 .066 -.043
Roadway × automation vector -.037 .067 -.063
Post-task worry .001 .069 .002
Post-task task engagement -.113 .078 -.189
Post-task distress -.093 .097 -.133
DSI aggression .009 .005 .218
DSI dislike of driving .005 .006 .115
DSI hazard monitoring .005 .005 .103
DSI thrill seeking -.001 .003 -.031
DSI fatigue proneness -.003 .006 -.069
*p < .05
Table D10
Summary of Hierarchical Regression Analysis for Variables Predicting Mean Steering Wheel Rate
Variable df F for ∆ R2 ∆R2 B SE B β
Step 1 2, 77 346.14 .900 .900
Roadway vector -.048 .002 -.948*
Vehicle-automation vector -.001 .002 -.019
Step 2 3, 76 231.10 .901 .001
Roadway vector -.048 .002 -.948*
Vehicle-automation vector -.001 .002 -.019
Roadway × automation vector .002 .002 .036
Step 3 6, 73 119.47 .908 .006
Roadway vector -.050 .002 -.980*
Vehicle-automation vector .000 .002 -.006
Roadway × automation vector .003 .002 .052
Post-task worry -.001 .002 -.020
Post-task task engagement -.004 .002 -.083*
Post-task distress .001 .002 .014
Step 4 11, 68 67.39 .916 .008
Roadway vector -.050 .002 -.978*
107
Table D11
Summary of Hierarchical Regression Analysis for Variables Predicting Mean SD of Steering Wheel
Rate
Variable df F for ∆ R2 ∆R2 B SE B β
Step 1 2, 77 246.91 .865 .865
Roadway vector -.062 .003 -.930*
Vehicle-automation vector .000 .003 -.007
Step 2 3, 76 164.18 .866 .001
Roadway vector -.062 .003 -.930*
Vehicle-automation vector .000 .003 -.007
Roadway × automation vector .002 .003 .035
Step 3 6, 73 90.01 .881 .015
Roadway vector -.065 .003 -.977*
Vehicle-automation vector .001 .003 .008
Roadway × automation vector .004 .003 .057
Post-task worry -.003 .003 -.051
Post-task task engagement -.009 .003 -.126*
Post-task distress .001 .003 .008
Step 4 11, 68 50.73 .891 .010
Roadway vector -.065 .003 -.972*
Vehicle-automation vector .000 .003 .004
Roadway × automation vector .004 .003 .054
Post-task worry -.004 .003 -.064
Post-task task engagement -.008 .003 -.119*
Post-task distress -.003 .004 -.035
DSI aggression .000 .000 .029
DSI dislike of driving .000 .000 .087
DSI hazard monitoring .000 .000 .054
DSI thrill seeking .000 .000 .025
DSI fatigue proneness .000 .000 .028
*p < .05
108
APPENDIX E
Psychometric Instruments
109
DSI
Please check one box only unless otherwise indicated (do not write in boxes at right margin).
Section A
1. Please state your age in years: ____________
5. Please state the year when you obtained your full driving license: ______
7. Estimate roughly how many miles you personally have driven in the past year:
Less than 5000 miles 5000-10,000 miles 10,000-15,000 miles
15,000-20,000 miles Over 20,000 miles
9. Please state which of these types of road you use frequently (check one or more boxes as appropriate):
Freeways Other main roads Urban roads Country roads
10. During the last three years, how many minor road accidents have you been involved in?
(A minor accident is one in which no-one required medical treatment, AND costs of damage to vehicles and property
were less than $800).
11. During the last three years, how many major road accidents have you been involved in?
(A major accident is one in which EITHER someone required medical treatment, OR costs of damage to vehicles and property
were greater than $800, or both).
12. During the last three years, have you ever been convicted for:
Please answer the following questions on the basis of your usual or typical feelings about driving. Each question asks you to answer
according to how strongly you agree with one or other of two alternative answers. Please read each of the two alternatives carefully
before answering. To answer, mark the horizontal line at the point which expresses your answer most accurately. Be sure to answer all the
questions, even if some of them don't seem to apply to you very well: guess as best you can if need be.
DCQ
These questions are concerned with how you usually deal with driving when it is difficult, stressful or upsetting. Think of those occasions
during the last year when driving was particularly stressful. Perhaps you nearly had an accident, or you were stuck in a traffic jam, or you
had to drive for a long time in poor visibility and heavy traffic. Use your experiences of driving during the last year to indicate how much
you usually engage in the following activities when driving is difficult, stressful or upsetting, by CIRCLING one of the numbers from 0 to
5 to the right of each question.
Not at all Very much
1. Relieved my feelings by taking risks or driving fast 0 1 2 3 4 5
2. Cheered myself up by thinking about things unrelated to the drive 0 1 2 3 4 5
3. Stayed detached or distanced from the situation 0 1 2 3 4 5
4. Tried to make other drivers more aware of me by driving close behind them 0 1 2 3 4 5
5. Wished that I was a more confident and forceful driver 0 1 2 3 4 5
6. Ignored my feelings about the drive 0 1 2 3 4 5
7. Made sure I avoided reckless or impulsive actions 0 1 2 3 4 5
8. Showed other drivers what I thought of them 0 1 2 3 4 5
9. Drove assertively or aggressively 0 1 2 3 4 5
10. Tried to gain something worthwhile from the drive 0 1 2 3 4 5
11. Showed other drivers I was in control of the situation 0 1 2 3 4 5
12. Made an extra effort to drive safely 0 1 2 3 4 5
13. Felt that I was becoming a more experienced driver 0 1 2 3 4 5
14. Made an effort to stay calm and relaxed 0 1 2 3 4 5
15. Swore at other drivers (aloud or silently) 0 1 2 3 4 5
16. Thought about good times I've had 0 1 2 3 4 5
17. Wished that I found driving more enjoyable 0 1 2 3 4 5
18. Made sure I kept a safe distance from the car in front 0 1 2 3 4 5
19. Went on as if nothing had happened 0 1 2 3 4 5
20. Refused to believe that anything unpleasant had happened 0 1 2 3 4 5
21. Told myself there wasn't really any problem 0 1 2 3 4 5
22. Let other drivers know they were at fault 0 1 2 3 4 5
23. Criticized myself for not driving better 0 1 2 3 4 5
24. Thought about the consequences of having an accident 0 1 2 3 4 5
25. Flashed the car lights or used the horn in anger 0 1 2 3 4 5
26. Felt I was learning how to cope with stress 0 1 2 3 4 5
27. Deliberately slowed down when I met a difficult traffic situation or 0 1 2 3 4 5
bad weather
28. Made a special effort to look out for hazards 0 1 2 3 4 5
29. Blamed myself for getting too emotional or upset 0 1 2 3 4 5
30. Concentrated hard on what I had to do next 0 1 2 3 4 5
31. Worried about what I was going to do next 0 1 2 3 4 5
32. Looked on the drive as a useful experience 0 1 2 3 4 5
33. Worried about my shortcomings as a driver 0 1 2 3 4 5
34. Thought about the benefits I would get from making the journey 0 1 2 3 4 5
35. Learnt from my mistakes 0 1 2 3 4 5
114
PRE-TASK DUNDEE STRESS
STATE QUESTIONNAIRE
General Instructions. This questionnaire is concerned with your feelings and thoughts at the moment. We would like
to build up a detailed picture of your current state of mind, so there are quite a few questions, divided into four
sections. Please answer every question, even if you find it difficult. Answer, as honestly as you can, what is true of
you. Please do not choose a reply just because it seems like the ‘right thing to say’. Your answers will be kept entirely
confidential. Also, be sure to answer according to how you feel AT THE MOMENT. Don’t just put down how you
usually feel. You should try and work quite quickly: there is no need to think very hard about the answers. The first
answer you think of is usually the best.
Before you start, please provide some general information about yourself.
First, there is a list of words which describe people’s moods or feelings. Please indicate how well each word describes
how you feel AT THE MOMENT. For each word, circle the answer from 1 to 4 which best describes your mood.
Please answer some questions about your attitude to the task you are about to do. Rate your agreement with the
following statements by circling one of the following answers:
In this section, we are concerned with your thoughts about yourself: how your mind is working, how confident you
feel, and how well you expect to perform on the task. Below are some statements which may describe your style of
thought RIGHT NOW. Read each one carefully and indicate how true each statement is of your thoughts AT THE
MOMENT. To answer, circle one of the following answers:
This set of questions concerns the kinds of thoughts that go through people’s heads at particular times, for
example while they are doing some task or activity. Below is a list of thoughts, some of which you might have
had recently. Please indicate roughly how often you had each thought DURING THE LAST TEN MINUTES
or so, by circling a number from the list below.
General Instructions
This questionnaire is concerned with your feelings and thoughts while you were performing the task. We would
like to build up a detailed picture of your current state of mind, so there are quite a few questions, divided into
six sections. Please answer every question, even if you find it difficult. Answer, as honestly as you can, what is
true of you. Please do not choose a reply just because it seems like the 'right thing to say'. Your answers will be
kept entirely confidential. Also, be sure to answer according to how you felt WHILE PERFORMING THE
TASK. Don't just put down how you usually feel. You should try and work quite quickly: there is no need to
think very hard about the answers. The first answer you think of is usually the best.
1. MOOD STATE
First, there is a list of words which describe people's moods or feelings. Please indicate how well each word
describes how you felt WHILE PERFORMING THE TASK. For each word, circle the answer from 1 to 4
which best describes your mood.
1. Happy 1 2 3 4
2. Dissatisfied 1 2 3 4
3. Energetic 1 2 3 4
4. Relaxed 1 2 3 4
5. Alert 1 2 3 4
6. Nervous 1 2 3 4
7. Passive 1 2 3 4
8. Cheerful 1 2 3 4
9. Tense 1 2 3 4
10. Jittery 1 2 3 4
11. Sluggish 1 2 3 4
12. Sorry 1 2 3 4
13. Composed 1 2 3 4
14. Depressed 1 2 3 4
15. Restful 1 2 3 4
16. Vigorous 1 2 3 4
17. Anxious 1 2 3 4
18. Satisfied 1 2 3 4
19. Unenterprising 1 2 3 4
20. Sad 1 2 3 4
21. Calm 1 2 3 4
22. Active 1 2 3 4
23. Contented 1 2 3 4
24. Tired 1 2 3 4
25. Impatient 1 2 3 4
26. Annoyed 1 2 3 4
27. Angry 1 2 3 4
28. Irritated 1 2 3 4
29. Grouchy 1 2 3 4
119
2. MOTIVATION AND WORKLOAD
Please answer the following questions about your attitude to the task you have just done. Rate your agreement
with the following statements by circling one of the following answers:
16. Please rate the MENTAL DEMAND of the task: How much mental and perceptual activity was required?
Low 0 1 2 3 4 5 6 7 8 9 10 High
17. Please rate the PHYSICAL DEMAND of the task: How much physical activity was required?
Low 0 1 2 3 4 5 6 7 8 9 10 High
18. Please rate the TEMPORAL DEMAND of the task: How much time pressure did you feel due to the pace at
which the task elements occurred?
Low 0 1 2 3 4 5 6 7 8 9 10 High
19. Please rate your PERFORMANCE: How successful do you think you were in accomplishing the goals of the
task?
Low 0 1 2 3 4 5 6 7 8 9 10 High
20. Please rate your EFFORT: How hard did you have to work (mentally and physically) to accomplish your
level of performance?
Low 0 1 2 3 4 5 6 7 8 9 10 High
21. Please rate your FRUSTRATION: How discouraged, irritated, stressed and annoyed did you feel during the
task?
Low 0 1 2 3 4 5 6 7 8 9 10 High
120
3. THINKING STYLE
In this section, we are concerned with your thoughts about yourself: how your mind is working, how confident
you feel, and how well you believed you performed on the task. Below are some statements which may
describe your style of thought during task performance. Read each one carefully and indicate how true each
statement was of your thoughts WHILE PERFORMING THE TASK. To answer circle one of the following
answers: Extremely = 4 Very much = 3 Somewhat = 2 A little bit = 1 Not at all = 0
This set of questions concerns the kinds of thoughts that go through people's heads at particular times, for
example while they are doing some task or activity. Below is a list of thoughts, some of which you might have
had recently. Please indicate roughly how often you had each thought during THE LAST TEN MINUTES
(while performing the task), by circling a number from the list below.
Next, please answer some questions about the task. Please indicate what you thought of the task while you were
performing it. Please try to rate the task itself rather than your personal reactions to it. For each adjective or
sentence circle the appropriate number, on the six point scales provided (where 0 = not at all to 5 = very much
so).
Threatening 0 1 2 3 4 5 Enjoyable 0 1 2 3 4 5
Fearful 0 1 2 3 4 5 Exhilarating 0 1 2 3 4 5
Worrying 0 1 2 3 4 5 Informative 0 1 2 3 4 5
Frightening 0 1 2 3 4 5 Challenging 0 1 2 3 4 5
Terrifying 0 1 2 3 4 5 Stimulating 0 1 2 3 4 5
Hostile 0 1 2 3 4 5 Exciting 0 1 2 3 4 5
Finally, think about how you dealt with any difficulties or problems which arose while you were performing the
task. Below are listed some options for dealing with problems such as poor performance or negative reactions
to doing the task. Please indicate how much you used each option, specifically as a deliberately chosen way of
dealing with problems. To answer circle one of the following answers:
I ...
NASA-TLX
We are interested in the "workload" you experienced during the experiment. Workload is
difficult to define but can be seen as made up of different factors (e.g. physical or mental
components). A set of six rating scales has been developed to evaluate the workload experienced
by an individual during different tasks. Please read the descriptions of the scales carefully. If you
have a question about any of the scales, please ask the experimenter about it. It is extremely
important that they be clear to you. After reading each description, please mark each scale with a
horizontal line at the point which expresses your experiences during the task most accurately.
MENTAL DEMAND Low/High How much mental and perceptual activity was
required (e.g., thinking, deciding, calculating,
remembering, looking, searching, etc.)? Was
the task easy or demanding, simple or
complex, exacting or forgiving?
TEMPORAL DEMAND Low/High How much time pressure did you feel due to
the rate or pace at which the tasks or task
elements occurred? Was the pace slow and
leisurely or rapid and frantic?
EFFORT Low/High How hard did you have to work (mentally and
physically) to accomplish this level of
performance?
Condition: _______________
Instructions: For each of the scales presented below, please mark each scale with a horizontal line
at the point that matches your experience with the task you just completed. Please note that the
"Performance" scale goes from "good" on the left to "poor" on the right.
0 25 50 75 100 0 25 50 75 100
Physical Demand Effort
Low High Low High
0 25 50 75 100 0 25 50 75 100
Temporal Demand Frustration
Low High Low High
0 25 50 75 100 0 25 50 75 100