Content uploaded by James L Szalma
Author content
All content in this area was uploaded by James L Szalma
Content may be subject to copyright.
Individual Differences in Response to Automation:
The Five Factor Model of Personality
James L. Szalma and Grant S. Taylor
University of Central Florida
This study examined the relationship of operator personality (Five Factor Model) and characteristics of
the task and of adaptive automation (reliability and adaptiveness—whether the automation was well-
matched to changes in task demand) to operator performance, workload, stress, and coping. This
represents the first investigation of how the Five Factors relate to human response to automation.
One-hundred-sixty-one college students experienced either 75% or 95% reliable automation provided
with task loads of either two or four displays to be monitored. The task required threat detection in a
simulated uninhabited ground vehicle (UGV) task. Task demand exerted the strongest influence on
outcome variables. Automation characteristics did not directly impact workload or stress, but effects did
emerge in the context of trait-task interactions that varied as a function of the dimension of workload and
stress. The pattern of relationships of traits to dependent variables was generally moderated by at least
one task factor. Neuroticism was related to poorer performance in some conditions, and all five traits
were associated with at least one measure of workload and stress. Neuroticism generally predicted
increased workload and stress and the other traits predicted decreased levels of these states. However, in
the case of the relation of Extraversion and Agreeableness to Worry, Frustration, and avoidant coping,
the direction of effects varied across task conditions. The results support incorporation of individual
differences into automation design by identifying the relevant person characteristics and using the
information to determine what functions to automate and the form and level of automation.
Keywords: automation, individual differences, performance, stress, workload
Supplemental materials: http://dx.doi.org/10.1037/a0024170.supp
The effect of automation on human response varies as a function
of its type, level, adaptiveness, and reliability (Parasuraman, Sheri-
dan, & Wickens, 2000). In principle automation should improve
performance and ease workload (Hancock & Chignell, 1988), but
empirical evidence has been inconsistent (Young & Stanton,
2004). Facilitative effects are most likely to occur when automa-
tion is well matched to the level of task load (Parasuraman &
Hancock, 2001). For instance, Parasuraman, Mouloua, and Hilburn
(1999) reported that workload-matched adaptive automation im-
proved failure detection in a monitoring task, but that the perfor-
mance and workload associated with automation poorly matched
to task demand was of similar magnitude to that of a control group
that did not receive automation support.
Automation may also impact operator stress, particularly if there
is a monitoring requirement. Research has established that moni-
toring tasks are stressful, and that the stress response is multidi-
mensional (Hancock & Warm, 1989; Warm, Parasuraman, &
Matthews, 2008). Several studies have employed the Dundee
Stress State Questionnaire (DSSQ) to assess the dimensions of
stress across a variety of cognitive tasks. The DSSQ is a factor-
analytically derived self-report measure that assesses task-related
changes in cognitive state. Three factors have been identified as a
‘Big 3⬘of stress (Matthews et al., 1999), each of which is associ-
ated with a different core relational theme (Lazarus, 1999). Dis-
tress, which reflects both cognitive and affective processes, is
linked to the theme of perceived overload of processing capacity.
Task Engagement reflects cognitive and energetic processes and is
related to the theme of commitment of effort. Worry reflects only
cognitive processes and is associated with a theme of self-
evaluation (Matthews et al., 2002). Studies using the DSSQ have
established that vigilance increases Distress and reduces Task-
Engagement (Warm et al., 2008). Few studies have examined the
impact of automation on these factors, but there is evidence that
automation reduces Distress but that Task Engagement is also
reduced (Funke, Matthews, Warm, & Emo, 2007).
The DSSQ is often administered in combination with the Coping
Inventory for Task Situations (CITS; Matthews & Campbell, 1998),
which measures three strategies for coping with stress: task-focused,
in which the individual focuses on task-related processes; emotion-
focused, in which the person engages in efforts to regulate their
James L. Szalma and Grant S. Taylor, Department of Psychology,
University of Central Florida.
This work was supported in part by a contract from the U.S. Army
Research Laboratory (ARL), J. L. Szalma, Principal Investigator
(#W911NF0620041). The views expressed here are those of the authors
and do not necessarily reflect the official policy or position of the Depart-
ment of the Army, the Department of Defense, or any official agency of the
United States government. We wish to thank Jennifer Ross, Jennifer Scott,
and David Tademy for assistance in creating the stimuli and in data
collection.
Correspondence concerning this article should be addressed to James L.
Szalma, Performance Research Laboratory, Department of Psychology, PO
Box 161390, University of Central Florida, Orlando, FL 32816. E-mail:
James.Szalma@ucf.edu
Journal of Experimental Psychology: Applied © 2011 American Psychological Association
2011, Vol. 17, No. 2, 71–96 1076-898X/11/$12.00 DOI: 10.1037/a0024170
71
emotional response; and avoidant, in which attention is diverted away
from the task to avoid confronting the source of stress.
Individual Differences in Response to Automation
The importance of individual differences for theoretical models
of automation and for the design and operation of automated
systems has been noted (Oron-Gilad, Szalma, Thropp, & Hancock,
2005; Parasuraman & Manzey, 2010). For instance, Singh, Mol-
loy, and Parasuraman (1993a) developed a scale to measure an
individual’s complacency potential, and they reported that it was
inversely related to automation failure detection (Singh, Molloy, &
Parasuraman, 1993b). This relationship is more likely to occur
under conditions conducive to complacency (i.e., automation of
consistently high reliability; Prinzel, Freeman, & Prinzel, 2005).
Interpersonal Trust
Trust is another trait that has been recognized as an important
determinant of system performance (e.g., Lee & Moray, 1992).
Rotter (1980) defined interpersonal trust as an expectancy that a
person can rely on the statement or action of another person or
group. However, he emphasized an important distinction between
general (i.e., dispositional) and specific (i.e., context dependent)
expectancies. General trust is thus a stable personality trait, and
specific trust is a state. Rotter (1980) argued that general expec-
tancies should exert a greater influence on trust behavior in novel
or unfamiliar situations or when the situational cues are appraised
by the person as consistent with their general expectancies. In
contrast, expectancies in more familiar circumstances will be more
strongly influenced by prior specific experience in similar situa-
tions than expectancies related to a general propensity to trust.
Lee and See (2004) argued that the expectancies related to
interpersonal trust may extend to interaction with nonhuman
agents. Response to automation might therefore be determined by
a combination of the situation-specific cues (including the auto-
mation characteristics), the appraised similarity of the situation to
prior experience, and the relationship between specific personality
traits (e.g., the Five Factors) and generalized expectancies. Indeed,
Merritt and Ilgen (2008) provided empirical evidence that both
dispositional and history-based trust affect user perceptions of
automation.
Based on these considerations, personality traits related to dis-
positional trust would be expected to affect response to automation
when individuals appraise the situation as similar to their previous
experience with the automated task. Rotter (1980) was careful to
point out that trust consists of expectations of reliability in the
absence of evidence to the contrary. Thus, general tendencies to
trust may influence acceptance of an agent when the trustworthi-
ness of the automation is ambiguous (e.g., during the early phase
of task performance, or cases in which the task is difficult), such
that traits associated with the propensity to trust will be more likely
to correlate positively with trust under such conditions. As the
reliability becomes less ambiguous (e.g., as operators become
familiar with the automation), trust behavior should depend more
on experience with the situation-specific cues than on trait char-
acteristics.
Automation and the Major Dimensions of Personality
Few studies have investigated how major personality factors
relate to operator response to automation. The Five Factor Model
(FFM; Costa & McCrae, 1992) is a taxonomy that proposes five
universal traits that constitute human personality. These are Ex-
traversion, Neuroticism, Conscientiousness, Agreeableness, and
Openness to Experience. Although the FFM is not universally
accepted (e.g., Block, 2001), it is one of the most widely adopted
approaches to personality description (Matthews, Deary, & White-
man, 2003). With respect to human performance, Extraversion and
Neuroticism have been studied extensively (Eysenck & Eysenck,
1985), but programmatic research on the relation of the other three
factors to task performance is mostly lacking (Matthews et al.,
2003). Indeed, to date no studies have simultaneously examined all
five factors in the context of a task with automation support.
The Five Factors of Personality
Neuroticism. Neuroticism is defined as an individual’s typ-
ical level of emotional stability or emotionality, the tendency to
experience negative affective states such as anxiety, sadness, an-
ger, and guilt (Costa & McCrae, 1992). It is related to greater
vulnerability to stress (higher Distress and Worry; Matthews et al.,
1999), and the use of emotion-focused and avoidant coping strat-
egies (Matthews & Campbell, 1998). Individuals high in Neurot-
icism tend to prefer and adapt better to positively affective envi-
ronments that are unambiguous and unthreatening. In terms of
performance, the relation of the trait to cognitive processes is task
dependent (Matthews et al., 2003; Szalma, 2008). However, when
effects are observed, Neuroticism predicts impairment of working
memory, attentional resources, and sustained attention, biases in
selective attention for negative information, a tendency to appraise
environmental demands as threats, and greater sensitivity to threat
information (Matthews et al., 2003).
Individuals high on this trait respond with stronger negative
affect to tasks that include threat stimuli and uncertainty regarding
the occurrence of events, particularly when the task is relatively
difficult (Matthews et al., 2003). Further, higher Neuroticism is
associated with a lower capacity to adapt to environmental change.
For instance, Cox-Fuenzalida, Swickert, and Hittner (2004) re-
ported that higher Neuroticism was associated with performance
impairment after transition from low to high workload or from
high to low workload.
Extraversion. Extraversion is defined primarily in terms of
preferences for social interaction, but it also includes characteris-
tics of assertiveness, activity level, preference for excitement and
stimulation, and generally positive affect (Costa & McCrae, 1992).
Individuals high in Extraversion benefit from environments that
provide opportunities for interaction and stimulation. The trait has
been associated with a “cognitive patterning” of greater working
memory and resource capacities, superior divided attention, but
more lenient response criteria and poorer sustained attention (Mat-
thews et al., 2003). Extraversion is also associated with lower
levels of posttask Distress (Matthews et al., 1999), and to more
task-focused and less emotion-focused coping (Matthews &
Campbell, 1998; Penley & Tomaka, 2002).
From the cognitive patterning, one might expect that Extraver-
sion should correlate positively with performance and negatively
72 SZALMA AND TAYLOR
with workload and stress under conditions of a time-constrained
task requiring divided attention to multiple displays or tasks.
Differences in response to automation as a function of Extraver-
sion are likely to be influenced by the reliability and the level of
automation. More extraverted individuals may have greater re-
source capacity for compensatory effort when the level of auto-
mation or its reliability is relatively low, but at high reliability or
high levels of automation Extraversion may be related to greater
complacency and poorer performance in the detection of automa-
tion failures.
Conscientiousness. Conscientious individuals actively self-
regulate their behavior to achieve goals and to actively plan,
organize, and complete tasks (Costa & McCrae, 1992). They tend
to thrive in environments in which they can act autonomously and
demonstrate self-efficacy by successfully performing moderately
challenging tasks. They should therefore achieve better perfor-
mance and report lower levels of stress and workload when the
environment supports the attainment of their goals. Any factor that
thwarts goal attainment may increase perceived workload and
stress, and performance may also be impaired if the resources
available for allocation are insufficient to effectively cope with the
demand. However, in general Conscientiousness is positively cor-
related with performance. Indeed, it is one of the strongest per-
sonality predictors of job performance (Barrick, Mount, & Judge,
2001), although there has been limited research on the relation of
this trait to task performance (Matthews et al., 2003).
In addition to performance, this trait predicts greater Task En-
gagement and less Distress and Worry (Matthews et al., 1999), and
the use of more task-focused and less emotion-focused or avoidant
coping strategies (Matthews & Campbell, 1998; Penley & To-
maka, 2002). The pattern of these relationships suggests that
conscientious individuals may devote more cognitive resources to
a task and they may be less likely to appraise the additional effort
as aversive relative to those lower on the trait.
Automation and task characteristics that are related to either the
level of challenge (task demand), or how well the aid supports or
interferes with effective performance (automation reliability) are
likely to moderate the effects of Conscientiousness. Conscientious-
ness should be positively associated with performance when au-
tomation is reliable, and conscientious operators should be less
susceptible to complacency, misuse, or disuse. How these individ-
uals respond to automation level will depend on how it impacts
goal attainment. If automation of any type aids performance by
allowing load shedding but permits the operator to exercise his or
her capacities (e.g., decision automation that permits operators to
make the final decision based on their own inspection of the
information), then higher levels of automation will benefit perfor-
mance and potentially reduce workload and stress.
Agreeableness. Agreeableness is defined as an interpersonal
trait associated with characteristics of sympathy, altruism, helpful-
ness, tender mindedness, and the propensity to trust others (Costa
& McCrae, 1992). Agreeable individuals adapt well to interper-
sonal settings requiring social interaction and cooperation, and the
trait generally correlates with lower Distress (Matthews et al.,
1999), less avoidant coping (Matthews & Campbell, 1998), and
more positive affect (Penley & Tomaka, 2002). Matthews et al.
(2003) suggested that the cognitive processes underlying Agree-
ableness may consist of the content of schemas representing be-
liefs, motivations, and actions styles (e.g., social skills) for inter-
personal relationships.
Due to the importance of cognitive expectancies for trust (Rot-
ter, 1980) and of trust in automation (Lee & See, 2004), the
relation of Agreeableness to user response is likely to manifest in
the belief content of schemas regarding automation reliability.
Those higher in Agreeableness should exhibit more accurate cal-
ibration of trust to the trustworthiness of the automated aid (Lee &
See, 2004), and they may therefore be less vulnerable to overre-
liance/misuse or disuse. Individuals low on this trait may have
expectancies of low reliability, be less likely to appropriately
calibrate their level of trust, and thus be more susceptible to disuse.
Openness. Openness to Experience, sometimes labeled Intel-
lect, consists of active imagination, aesthetic sensitivity, attention
to feelings, intellectual curiosity, and independent judgment and
enjoyment of variety and novelty (Costa & McCrae, 1992). Open
individuals effectively adapt to novel situations or environments
that provide opportunities to engage their intellectual capacities.
They enjoy intellectually challenging activities, and to the extent
that a task satisfies this tendency, this trait should correlate with
better performance and lower perceived workload and stress. In-
deed, Openness correlates positively with an individual’s typical
level of intellectual engagement in activities and with cognitive
ability measures (Ackerman & Heggestad, 1997). Further, Open-
ness correlates negatively with Distress (Matthews et al., 1999),
and open individuals are more likely to make challenge rather than
threat appraisals and engage in more task-focused coping than
avoidant coping (Penley & Tomaka, 2002). Individuals higher on
this trait may therefore have more cognitive resources to devote to
a task (or they may be more willing to engage their capacities).
Task characteristics related to resource demand (i.e., task de-
mand, automation reliability) should moderate the relation of
Openness to performance. Open individuals should benefit from
automation that performs repetitive or mundane tasks, but high-
level decision automation that is highly reliable may not provide
sufficient cognitive stimulation for them and may thus increase
their workload and stress. However, they may be less vulnerable to
misuse if the “raw data” are available, as they would be more
likely to engage effort to compare the automation response to their
own evaluation of the display. Those low in Openness may benefit
from high-level automation that relieves task load, but they may
also be more vulnerable to misuse.
Personality and Human Response to Automation
No studies have examined the relationships of Neuroticism,
Conscientiousness, Agreeableness, or Openness to human re-
sponse to automation. However, two studies investigated the ef-
fects of Extraversion. Thus, in addition to complacency potential,
Singh et al. (1993b) also examined the correlation between per-
formance and Extraversion as a function of automation reliability.
However, the relationship was not statistically significant. More
recently Merritt and Ilgen (2008) reported that Extraversion was
positively correlated with willingness to trust automation, which
they argued was due to a generally higher propensity of extraverts
to trust in the context of interpersonal interactions. However, they
did not report the correlation of Extraversion with performance,
workload, or stress response.
73
PERSONALITY AND ADAPTIVE AUTOMATION
Relation of the five factors to interpersonal trust. Previous
research has established a link between Rotter’s interpersonal trust
construct and other dimensions of personality. Interpersonal trust
correlates positively with facets of Extraversion, Agreeableness,
and Openness (Couch, Adams, & Jones, 1996; Evans & Revelle,
2008; Mooradian, Renzl, & Matzler, 2006), and negatively with
facets of Neuroticism (Breen, Endler, Prociuk, & Okada, 1978;
Couch et al., 1996; Mooradian et al., 2006). Evans and Revelle
(2008) argued that Extraversion and Agreeableness, as compo-
nents of trust, are related to willingness to accept vulnerability, and
that Agreeableness motivates trust in situations of high uncertainty
or risk.
Moderation of Individual Differences by Task and
Automation Characteristics
Parasuraman et al. (2000) presented a model delineating the
important characteristics of automation, and more recently Thropp
(2006) incorporated individual differences in operator characteris-
tics into this model (see also Oron-Gilad et al., 2005). For a given
personality trait, some task properties may be more important than
others, and it is likely that the automation and task characteristics
most relevant to operators (i.e., have the greatest impact on oper-
ator response) will be those that influence the pattern of cognitive
and affective processes that characterize a particular trait (cf.,
Matthews, 2008). For instance, due to the greater sensitivity of
individuals high in Neuroticism to threat stimuli and their greater
stress vulnerability, threat-related task characteristics may moder-
ate the relation of this trait to operator response. In particular, tasks
requiring the detection of threats (e.g., terrorists) may exert a
stronger effect on individuals higher in Neuroticism than would be
observed in an automated task involving process control or even an
aviation-related task.
The degree of uncertainty associated with a task may also
moderate personality effects. For instance, high Neuroticism indi-
viduals are more likely to appraise ambiguous situations as threats,
while Extraversion and Conscientiousness are associated with
challenge appraisals (Matthews, 2008). One way in which uncer-
tainty manifests in automated tasks is in whether the “raw data” or
information regarding stimulus or system characteristics are ac-
cessible to the human. If this information is available, the operator
can judge the reliability of the automation. However, when this
information is not provided, the operator may be more dependent
on the automation, particularly if it is crucial for task performance
(Rice, 2009). It may be that personality effects vary as a function
of the degree to which the automation is crucial for performance of
a task, such that when the automation is crucial, most individuals
will use it regardless of personality differences. Greater variability
across individuals may occur when use of the automation is not
crucial or required for task performance, as in the present study.
The Present Study
This study was designed to examine the joint effects of task and
person characteristics on the performance, workload, stress, and
coping associated with an automated threat-detection task. The
task characteristics examined were automation reliability (95%
reliable versus 75% reliable), task demand (two versus four dis-
plays to be monitored), and the workload-adaptiveness of the
automation (i.e., workload-matched automation that was engaged
when the task demand was high and disengaged when it was low,
versus workload-mismatched automation that was engaged at
lower task loads and disengaged when task load was higher;
Parasuraman et al., 1999; Parasuraman & Hancock, 2001). The
dependent variables were performance (proportion of correct de-
cisions and response time), agreement with automation, perceived
workload, stress, and coping strategy. After three 10-min blocks of
experience with one of the automation conditions, all participants
completed a 10-min block of trials with 95% reliable automation
under conditions of higher task demand, and participants could
choose to use or turn off the automation. Provision of choice
during this final block permitted a behavioral measure of automa-
tion use.
Task and Automation Characteristics
The task used in this study does not capture all elements repre-
sentative of automation. First, our “high” reliability consisted of
95% reliable automation to provide sufficient data for analysis, but
this is lower than most designers or users would tolerate. The 75%
reliable automation represents the level of reliability one standard
deviation below the mean of unaided human performance. This
value was selected so that performance in this condition would not
be as good as or better than unaided monitoring. Note that 75%
reliability is within the confidence interval of the automation
effectiveness threshold identified by Wickens and Dixon (2007).
They also reported that dependence on automation is stronger at
higher levels of task demand, even in situations in which the “raw
data” are available to the operator. It is therefore possible that low
reliability automation will not lead to disuse when task demand is
high.
Second, this task required threat detection using a simulated
uninhabited ground vehicle (UGV), and as such is likely quite
different from automation in civilian industrial or transportation
domains. Third, in operating a UGV, human operators typically
have control over the movement of the vehicle and the rotation of
the camera in order to scan the environment, and they also have
some control over the time spent inspecting a scene. In the present
study, these parameters were constrained in order to achieve the
necessary level of experimental control. Fourth, there was no
explicit algorithm used to generate automation responses, and the
participants were not informed of how the automation worked.
Finally, UGV operators are not (currently) required to scan more
than one source, and automation would not be invoked consistently
by matching or mismatching its engagement according to the
number of displays to be monitored. However, in this study, task
demand was operationalized by the number of sources to be
monitored, and the workload-matching manipulation permitted
analysis of individual differences in response to automation under
different levels of demand.
Hypotheses
Few studies have examined the joint influence of person and
automation/task characteristics on human response. Specific pre-
dictions for trait effects are therefore difficult to specify precisely.
However, based on the above considerations and extant research,
the hypotheses described below were tested. Note that although the
74 SZALMA AND TAYLOR
traits are examined separately in the present work, tests of trait–
trait interactions are possible (e.g., see Szalma, 2008). The traits
were examined separately in this study to simplify analysis and
interpretation.
Performance
Based on previous research, the predictions for performance are
associated mostly with Neuroticism and Extraversion. The atten-
tional bias for threat information may provide a performance
advantage to those higher in Neuroticism, given that the task in this
study required threat detection, but the imposition of poor auto-
mation or higher task load could offset this benefit. Hence, per-
formance by those higher in Neuroticism may be facilitated by
reliable automation but be impaired under conditions of low reli-
ability, and these effects may be stronger at higher levels of task
demand. Because individuals high on this trait are emotionally
labile and low in dispositional trust, they may overreact to poor
automation and be more susceptible to disuse and less agreement
with automation. In addition, changes in the level of automation
and the level of task demand may impair performance for individ-
uals high in Neuroticism, as they have difficulty in adapting to
change (Cox-Fuenzalida et al., 2004).
The cognitive patterning of Extraversion suggests that task load
and workload-adaptiveness should moderate the relationship of
this trait to performance, such that extraverts should perform better
under the more stimulating conditions of higher task demand.
Because individuals higher in Extraversion do not perform well
under conditions of underload (Matthews et al., 2003), this trait
may be negatively related to performance when the automation is
highly reliable, particularly the workload-mismatched automation
condition in which the automation is provided for the less demand-
ing task. Paradoxically, more extraverted individuals may perform
better with low-reliability automation because it offers more stim-
ulation. As Extraversion is positively correlated with trust, it
should also be associated with a more accurate calibration of trust
to the trustworthiness of the automation. Hence, this trait should
correlate positively with agreement with automation recommen-
dation when it is reliable, but Extraversion should correlate neg-
atively with agreement when the automation is unreliable.
Operator skill. Note that there is a possibility that as partici-
pants’ threat detection improves the automation may become less
useful to them. Meyer and Biton (2002) reported that the diagnostic
value of a warning decreases as a function of operator skill in
performing the task, because better operators have fewer performance
failures that lead to warnings. Although in the present study automa-
tion failures were independent of operator response, the findings of
Meyer and Biton (2002) suggest that individuals who are better
performers may be less vulnerable to automation failures at either
reliability level and less sensitive to changes in the level of automa-
tion. For instance, if individuals higher in Conscientiousness and
Openness devote more effort to task performance, they may improve
with time on task and the effects of differences in reliability and level
of automation may be attenuated for these individuals.
Subjective Response to Automation: Perceived
Workload, Stress, and Coping
Extraversion, Conscientiousness, Agreeableness, and Openness
should be negatively related to perceived workload and to pre- and
posttask stress, and Neuroticism should be positively related to
workload and stress. Neuroticism was expected to be negatively
related to task-focused coping and positively related to emotion-
focused and avoidant coping, Each of the other traits would be
expected to be positively related to task-focused coping and in-
versely related to emotion-focused and avoidant coping (Penley &
Tomaka, 2002).
Note that the relationships of each trait to the dependent mea-
sures were expected to be moderated by task and automation
characteristics. The positive relationship of Neuroticism with
workload and stress should be strongest in cases where task
demand is highest. Indeed, highly reliable, workload-matched au-
tomation should show the greatest benefit for individuals higher in
Neuroticism in terms of reductions in workload and stress. Given
that extraverts tend to prefer stimulating environments, Extraver-
sion should be negatively related to workload and stress when no-
or low-reliability automation is provided, but positively related to
these measures when demand is low and the automation is reliable.
This pattern would likely be reversed for Conscientiousness, as
conscientious individuals are more likely to appraise poor auto-
mation or removal of reliable automation as a potential threat to
achieving their performance goals. They may therefore report
higher levels of workload and stress as a result of their compen-
satory efforts. Agreeableness would be expected to be negatively
related to workload, stress, and emotion-focused or avoidant cop-
ing under conditions of reliable automation. Finally, Openness
should be negatively related to workload and stress, and this effect
should be strongest under the more demanding conditions.
Method
Participants
A total of 161 university students (age: M⫽19.8, SD ⫽2.59)
participated, of which 52 were male and 109 female (see Table 1).
Participants chose to receive either course credit or cash payment
(at a rate of $8/hour) for their participation.
Experimental Design
The present study employed a 2 (reliability) by 2 (workload-
adaptiveness) by 4 (block/task demand) mixed design with re-
peated measures on the last factor. The number of displays to be
monitored varied as a function of block, changing in a pattern of
2–4–2–4 displays for blocks 1–4, respectively. This pattern of
task demand was experienced by all participants (see Table 1). In
block 4, all participants experienced the 95% reliable automation
which they could choose to engage or disengage during that block.
This latter condition was designed to provide a behavioral measure
of the use of automation in a common condition for all experi-
mental groups. Each participant was assigned at random to one of
the four experimental conditions.
Procedure
The experiment consisted of two separate sessions. In the first
session, participants completed the 300 question version of the five
factor inventory provided by the International Personality Item
Pool (IPIP) a no-cost alternative to the NEO PI-R. Previous re-
75
PERSONALITY AND ADAPTIVE AUTOMATION
search has established that the IPIP-NEO scales are valid alterna-
tive measures of the five factors and their facets (Goldberg et al.,
2006). The questionnaire was administered via computer and re-
quired approximately one hour to complete. Data were collected in
groups of up to 15 participants.
Task. In the second session, participants completed the au-
tomated threat detection task. Participants were presented with a
scenario that terrorists had infiltrated an office building in the
United States, and they were told that they would participate in a
reconnaissance mission to identify terrorists, civilians, friendly
forces, and improvised explosive devices (IEDs) within the build-
ing. A team of unmanned ground vehicles (UGVs) would be sent
into the building to transmit surveillance video to a remote oper-
ator (the participant).
Participants were instructed that the movement of the UGVs
would be fully automated; their task was to monitor the video
feeds being transmitted from each vehicle. These videos were
actually a series of prerecorded clips that were presented in a
random order held constant across all participants. Participants
viewed a total of 420 short video clips; each 6.4 cm high by 8.9 cm
wide (approximately 9.1° by 12.5° of visual angle), of the simu-
lated view of a UGV maneuvering through different rooms (see
Figure 1; color figure available online as Supplemental Material).
Each video clip would pan across a single room for 10 seconds as
the UGV remained stationary.
The participant’s task was to visually scan the video of each
room. After presentation of each clip, the participant had 7 seconds
to respond regarding whether they saw a terrorist, friendly soldier,
civilian, IED, or none of the above (i.e., the room was “empty” or
“clean”); no single video contained more than one of these items.
Participants were familiarized with the stimuli and the task during
a training period immediately prior to initiating the task. The
videos were presented in four 10-min blocks of 35 trials each. Task
demand was manipulated by requiring participants to monitor
video displays from 2 UGVs on each trial in blocks 1 and 3 (lower
task demand) and 4 UGVs on each trial in blocks 2 and 4 (higher
task demand).
Automated decision aid. During the 7-s response period, an
automated diagnostic/decision aid was either engaged or disen-
gaged, depending on the condition and block (see Table 1). A
horizontal array of response buttons indicating the possible event
categories was presented below each video clip. When engaged,
the automation provided a recommendation by highlighting one of
the event categories, but the participant was instructed to make the
final decision (level 4 automation; Parasuraman et al., 2000). In
blocks without automation support, the same response options
Table 1
Experimental Design for the Present Study
Adaptiveness Reliability Block 1 Block 2 Block 3 Block 4
Workload-Matched
Automation
75 % Reliable
(n⫽42; 27 females)
2 Displays
Automation off
4 Displays
Automation on
2 Displays
Automation off
4 Displays (95% reliable)
Automation on
95 % Reliable
(n⫽38; 25 females)
2 Displays
Automation off
4 Displays
Automation on
2 Displays
Automation off
4 Displays (95% reliable)
Automation on
Workload-Mismatched
Automation
75 % Reliable
(n⫽41; 29 females)
2 Displays
Automation on
4 Displays
Automation off
2 Displays
Automation on
4 Displays (95% reliable)
Automation on
95 % Reliable
(n⫽40; 29 females)
2 Displays
Automation on
4 Displays
Automation off
2 Displays
Automation on
4 Displays (95% reliable)
Automation on
Figure 1. Example task interface used in the present study (for a 4-display block during which the automation
was engaged). The image on the left presents examples of the display during the 10-s video presentations. The
image on the right shows the display during the response interval. Recommendations of the automated aid are
highlighted. (Color version available online as Supplemental Material.)
76 SZALMA AND TAYLOR
were presented but the computer did not highlight a recommen-
dation (level 1 automation: completely manual). Note that the
participant was not completely dependent on the automation to
perform the task, as the “raw data” (i.e., the video streams) were
provided.
Reliability was defined as the accuracy of the automation rec-
ommendations. Each participant received either “high” (95%) or
“low” (75%) reliability automation. These values were determined
based on pilot data to be one standard deviation above and below
the mean correct detection rate achieved by human observers
(without automation support) for each stimulus type. For instance,
participants tended to perform better when detecting terrorists or
friendly soldiers than when detecting IEDs, so automation reliabil-
ity varied across signal types in the same pattern. Participants had
no direct knowledge of the reliability of the system, aside from
being instructed that “the automation may make mistakes as it is
based on a computer vision algorithm.” Participants did not receive
any information or feedback regarding the accuracy of the auto-
mation or of their own response.
Each participant was also assigned to one of two workload-
adaptiveness groups (workload-matched or workload-
mismatched automation). The workload-matched automation
provided no automation support when task demands were low
(blocks 1 and 3), and automation was engaged when task
demand was high (block 2). The workload-mismatched auto-
mation consisted of the opposite pattern: automation was pro-
vided when task demands were low (blocks 1 and 3) but not
when task demands were high (block 2).
In the fourth block there were four video clips to monitor, and
all participants had direct control over whether the automated aid
was engaged, and they could turn it on or off at any time during the
block. All participants received 95% reliable automation for this
block when they chose to engage it. To encourage participants to
achieve their best performance possible during block 4, they re-
ceived a cash bonus based on their performance in that block of
$.05 for each correct response and a deduction of $.01 for each
incorrect response, resulting in a maximum bonus of $7. Most
participants received between $4 and $6.
Workload, stress, and coping measures. Measures of per-
ceived workload (NASA-TLX; Hart & Staveland, 1988), stress
(DSSQ; Matthews et al., 2002), and coping (CITS; Matthews &
Campbell, 1998) were administered to participants after each
block, with an additional administration of the DSSQ before the
first block to assess pretask cognitive state. Participants responded
to each instrument on the computers used for the experimental
task.
Results
Independent Variable Analyses
Analyses of the manipulated variables were conducted to deter-
mine their effects independent of relationships involving person-
ality. For all analyses involving repeated measures the degrees of
freedom were adjusted for violations of the sphericity assumption
(Maxwell & Delaney, 2004). The Bonferroni correction was ap-
plied to all post hoc comparisons. Measures of association were
computed for all ANOVA Ftests (
2
). Values of
2
⫽.01, .06,
and .14 are considered small, medium, and large associations,
respectively (Maxwell & Delaney, 2004).
Performance Accuracy
The means and standard deviations for performance accuracy
are shown in Table 2. As can be seen in the table, decision
accuracy varied as a function of stimulus category. Correctly
identifying IEDs and empty rooms were the most difficult,
while identifying terrorists or friendly military was easier.
Performance effects examined in this study were for overall
decision accuracy (collapsed across stimulus categories). The
data for overall performance accuracy are displayed in Figure
2a. ANOVA revealed significant effects for reliability, F(1,
157) ⫽6.48, p⫽.01,
2
⫽.03, block, F(1, 382) ⫽50.59, p⬍
.001,
2
⫽.13, and a significant reliability by adaptiveness by
block interaction, F(2, 382) ⫽18.80, p⬍.001,
2
⫽.07. Tests
of the adaptability by block interaction within each level of
reliability revealed a significant interaction at 95% reliability,
F(2, 157) ⫽32.92, p⬍.001,
2
⫽.18, but not at low reliability
(p⫽.14). Tests of the effects of adaptiveness within each block
at 95% reliability revealed significant differences in block 1,
F(1, 76) ⫽14.37, p⬍.001),
2
⫽.15, with the workload-
mismatched group (M⫽.85; SD ⫽.10) achieving a higher
scores than the workload-matched group (M⫽.78; SD ⫽.06,
d⫽.86). A significant group difference was also observed in
block 2, F(1, 76) ⫽45.20, p⬍.001,
2
⫽.37, but in this case
the workload-matched group (M⫽.88; SD ⫽.06) performed
better than the workload-mismatched group (M⫽.78; SD ⫽
.06, d⫽1.52). Note that both groups experienced the same
level of demand (four displays) in the second block. There were
no group differences during blocks 3 or 4 (p⬎.07 in each
case). The pattern of results indicates that during the first two
blocks the schedule of automation (adaptiveness) did not impact
performance per se, but that 95% reliable automation improved
decision accuracy when it was engaged.
Table 2
Mean Accuracy and Response Times for Each Stimulus Type
and Each Block (Standard Deviations in Parentheses)
(N⫽161)
Stimulus
Block
1234
Accuracy (proportions)
Empty Room .92 (.10) .86 (.12) .84 (.16) .92 (.09)
Friendly Military .97 (.08) .95 (.05) .98 (.09) .96 (.04)
Civilian .83 (.11) .90 (.07) .92 (.11) .94 (.05)
IED .55 (.18) .56 (.18) .63 (.21) .68 (.20)
Terrorist .97 (.09) .95 (.06) .98 (.09) .97 (.04)
Response Time (seconds)
Empty Room 1.08 (.22) .82 (.11) .96 (.22) .80 (.11)
Friendly Military 1.02 (.28) .93 (.13) .97 (.25) .97 (.17)
Civilian 1.06 (.26) 1.00 (.16) 1.04 (.23) .88 (.12)
IED 1.05 (.25) .94 (.15) .91 (.23) .88 (.12)
Terrorist 1.07 (.30) 1.05 (.22) .85 (.20) .90 (.16)
77
PERSONALITY AND ADAPTIVE AUTOMATION
Response Time
The means and standard deviations for response time are shown
in Table 2 for the different stimulus categories. The data for overall
response time to correct decisions is shown in Figure 2b. ANOVA
revealed significant effects for block, F(2, 323) ⫽86.50, p⬍.001
2
⫽.21, reliability, F(1, 156) ⫽5.85, p⫽.02,
2
⫽.03, and a
block by reliability interaction, F(2, 323) ⫽3.77, p⫽.02,
2
⫽
.01. All other sources of variance failed to reach statistical signif-
icance (p⬎.05 in each case). Tests of the effects of reliability
within each block indicated a significant effect for blocks 1, F(1,
159) ⫽5.85, p⫽.02,
2
⫽.03, and 2, F(1, 159) ⫽5.04, p⫽.03,
2
⫽.02. In block 1, response time was significantly faster at 95%
reliability (M⫽1.00, SD ⫽.18) than at 75% reliability (M⫽1.08,
SD ⫽.23, d⫽.38). The pattern was similar in block 2, with faster
response time at 95% reliability (M⫽.90, SD ⫽.11) than at 75%
reliability (M⫽.94, SD ⫽.11, d⫽.35). There were no significant
differences between reliability conditions in blocks 3 or 4 (p⬎
.07).
Agreement with Automation
Agreement with automation was determined by computing the
proportion of trials on which the individual selected the category
recommended by the automation. Two types of agreement were
computed—one in which the recommendation by the computer
was correct and the other for cases in which the participant agreed
with an incorrect recommendation. Because of the workload-
adaptiveness manipulation, agreement with automation could not
be analyzed for the full factorial design. Analyses were therefore
Figure 2. a. Proportion of correct decisions as a function of block and workload-matching condition at
each level of automation reliability. Note. Error bars are standard errors. b. Response time as a function of
block and workload-matching condition at each level of automation reliability. Note. Error bars are standard
errors.
ⴱ
p⬍.05.
78 SZALMA AND TAYLOR
computed separately for the workload-mismatched (blocks 1, 3,
and 4) and workload-matched (blocks 2 and 4) conditions.
Workload-mismatched groups. ANOVA revealed a signif-
icant block effect, F(1, 91) ⫽9.56, p⫽.001,
2
⫽.13. Post hoc
comparisons indicated that agreement with correct automation was
significantly higher in block 3 (M⫽.91, SD ⫽.05) than in blocks
1(M⫽.88, SD ⫽.06, d⫽.25) or 4 (M⫽.86, SD ⫽.10, d⫽.42).
The levels of agreement for the latter two blocks were not signif-
icantly different from one another (p⫽.19). The main effect for
reliability and the interaction between block and reliability were
not statistically significant (p⬎.07 in each case). For agreement
with incorrect automation, there were significant effects for block,
F(1, 112) ⫽7.89, p⫽.001,
2
⫽.12 and the block by reliability
interaction, F(1, 112) ⫽5.11, p⫽.008,
2
⫽.08. Tests for the
effect of reliability within each block resulted in a significant
effect for block 1, F(1, 79) ⫽11.96, p⫽.001,
2
⫽.13.
Agreement with the incorrect decision aid in block 1 was higher
for participants who received 75% reliable automation (M⫽.28,
SD ⫽.11) than for those who received 95% reliable automation
(M⫽.18, SD ⫽.15, d⫽.77).
Workload-matched groups. For the workload-matched con-
ditions, analysis of agreement with correct automation revealed
significant effects for block, F(1, 64) ⫽11.68, p⫽.001,
2
⫽.15,
reliability, F(1, 64) ⫽5.68, p⫽.02,
2
⫽.07, and the interaction
between these factors, F(1, 64) ⫽4.04, p⫽.049
2
⫽.06. Tests
for the effect of reliability within each block indicated a significant
effect only for block 2, F(1, 78) ⫽6.68, p⫽.01,
2
⫽.08.
Agreement with correct automation was higher in the 95% condi-
tion (M⫽.87, SD ⫽.08) than in the 75% condition (M⫽.82,
SD ⫽.05, d⫽.58). Analysis of agreement with incorrect auto-
mation revealed significant effects for reliability, F(1, 61) ⫽6.34,
p⫽.01,
2
⫽.09, and for block by reliability, F(1, 61) ⫽3.88,
p⫽.05,
2
⫽.05. Tests for the effect of reliability within each
block indicated a significant difference for block 2, F(1, 78) ⫽
28.77, p⬍.001,
2
⫽.26. Agreement with incorrect automation
was higher at 95% reliability (M⫽.56, SD ⫽.22) than at 75%
reliability (M⫽.35, SD ⫽.10, d⫽1.20).
Subjective Response to Automation: Perceived
Workload, Stress, and Coping
Perceived workload. For global workload and the subscales,
the only statistically significant effects were for block (see Table
3). In general perceived workload was higher when four displays
were to be monitored (blocks 2 and 4) relative to the 2-display
blocks (1 and 3), independent of automation reliability or adap-
tiveness.
Stress. For all three scales the only significant effect was for
block (see Table 4 and Figure 3). Post hoc tests indicated that
post-Task Engagement declined in block 3 and increased in block
4. Posttask Distress increased from the pretask state until block 2,
and then declined from block 2 to blocks 3 and 4. Posttask Worry
declined from the pretask state to block 3, after which it remained
stable. In sum, stress varied as a function of time on task, level of
demand, and stress scale, but it was not impacted by the properties
of the automation.
Stress coping. For task-focused coping significant effects
were observed for block (see Table 4), and for the reliability by
adaptiveness interaction, F(1, 157) ⫽5.20, p⫽.024,
2
⫽.03.
For the block effect post hoc comparisons indicated that task-
focused coping scores in block 3 were significantly lower than
those in the other three blocks (p⬍.001 in each case). The
interaction was analyzed by examining the adaptiveness effect
within each level of reliability. A statistically significant difference
between adaptiveness groups was observed at high reliability, F(1,
76) ⫽3.92, p⫽.051,
2
⫽.05 (see Figure 4). Task-focused
coping was significantly higher among those who received reliable
workload-mismatched automation (M⫽17.44, SD ⫽4.75) than
those in the workload-matched reliable automation condition (M⫽
15.31, SD ⫽4.74, d⫽.45). The adaptiveness effect at low
reliability was not statistically significant (p⫽.21).
For both emotion-focused and avoidant coping the only signif-
icant effect observed was for block (Table 4 and Figure 5); Post
hoc tests indicated that emotion-focused coping was used more in
block 2 than in either block 3 or 4, and that avoidant coping was
used more in block 3 than in the other three blocks. In sum,
task-focused coping varied as a function of block, reliability, and
adaptiveness. Emotion-focused and avoidant coping varied as a
function of task demand/block. Neither of these latter strategies
varied as a function of automation properties.
Individual Differences Analyses
Correlational analyses were conducted for personality traits and
the dependent measures across adaptiveness and reliability condi-
tions to evaluate the relationships independent of the automation
Table 3
Summary Statistics for ANOVAs of Perceived Workload Scales (Standard Deviations in Parentheses) (N⫽161)
Scale Fdf1, df2
2
Block
Significant comparisons1234
GWL 60.05
ⴱⴱⴱ
2,415 .31 52.90 (19.89) 64.74 (21.33) 48.76 (21.62) 59.52 (22.67) All
MD 46.42
ⴱⴱⴱ
2,438 .24 261.81 (141.53) 321.60 (138.98) 213.58 (136.07) 270.53 (140.12) All except B1 vs. B4
TD 26.06
ⴱⴱⴱ
2,453 .14 159.04 (114.82) 209.47 (128.18) 129.32 (115.42) 184.48 (133.46) All except B1 vs. B4
PW 4.71
ⴱⴱ
2,434 .01 374.00 (89.08) 388.42 (91.29) 356.79 (117.99) 363.76 (107.68) B2 vs. B3 & B4
E 27.66
ⴱⴱⴱ
2,460 .15 141.36 (103.69) 163.32 (108.64) 198.17 (117.58) 209.56 (116.66) B2 vs. B1 & B3; B4 vs. B1 & B3
F 2.68
ⴱ
2,447 .01 79.60 (9.65) 103.23 (108.75) 89.37 (118.02) 84.84 (106.28) B2 vs. B1
Note. GWL ⫽Global Workload; TD ⫽Temporal Demand; PW ⫽Performance Workload; E ⫽Effort; F ⫽Frustration; B1-B4 ⫽blocks 1 to 4,
respectively.
ⴱ
p⬍.05.
ⴱⴱ
p⬍.01.
ⴱⴱⴱ
p⬍.001.
79
PERSONALITY AND ADAPTIVE AUTOMATION
characteristics. The relationships among the independent variables,
personality traits, and dependent measures were evaluated by path
analysis using AMOS (Arbuckle, 2009). To simplify analyses and
interpretation, separate path models were tested for each of the five
trait domains within each dependent variable.
In path analysis, interactive effects of independent variables can be
tested via the multiple groups approach (Kline, 2005). The first step is
to fit the model with no constraints (i.e., the unconstrained model). If
an adequate fit is obtained, then the model is tested with the constraint
that model parameters are equivalent across groups. The
2
difference
test is used to determine whether the constrained models represented
a better fit than the unconstrained model (Brown, 2006; Kline, 2005).
If the unconstrained model fits the data best, then there are significant
differences in model parameters (e.g., the path coefficient, which
indicates the relationship between two variables) across experimental
conditions. This is interpreted to mean that the relationship between a
trait and a dependent measure varies as a function of experimental
condition (i.e., an interactive effect).
Table 4
Summary Statistics for ANOVAs of Stress and Coping Scales (standard deviations in parentheses) (N⫽161)
Scale Fdf1, df2
2
Pre-task
Block
Significant comparisons1234
TE 39.85
ⴱⴱⴱ
3,524 .17 .61 (.81) .61 (1.02) .55 (1.04) ⫺.02 (1.00) .33 (1.03) B3 vs. pre, B1, B2, & B4; B4 vs. pre,
B1,&B2
D 41.01
ⴱⴱⴱ
3,488 .19 ⫺.60 (1.01) .006 (1.07) .27 (1.17) ⫺.02 (1.10) ⫺.03 (1.08) B1 vs. pre & B2; B2 vs. B3
W 163.60
ⴱⴱⴱ
2,340 .69 .23 (1.07) ⫺.67 (.88) ⫺.78 (.86) ⫺.89 (.89) ⫺.84 (.90) B1 vs. pre & B2; B2 vs. B3
TC 20.04
ⴱⴱⴱ
2,398 .11 — 16.88 (5.24) 16.97 (5.41) 14.78 (5.56) 16.14 (5.84) B3 vs. B1, B2, & B4
EC 6.94
ⴱⴱⴱ
2,432 .04 — 5.00 (4.51) 5.59 (4.70) 4.62 (4.63) 4.72 (4.53) B2 vs. B3; B2 vs. B4
AC 10.23
ⴱⴱⴱ
2,432 .06 — 4.27 (4.15) 4.65 (4.24) 5.82 (5.33) 5.02 (5.14) B3 vs. B1, B2, & B4
Note. TE ⫽Task Engagement; D ⫽Distress; W ⫽Worry; TC ⫽task-focused coping; EC ⫽emotion-focused coping; AC ⫽avoidant coping; B1-B4 ⫽
blocks 1 to 4, respectively.
ⴱⴱⴱ
p⬍.001.
Figure 3. Pre- and posttask scores as a function of block for the three stress scales. Note. Error bars are
standard errors.
80 SZALMA AND TAYLOR
If the structural weights model, in which the path coefficients
are constrained to be equal across groups, provides a better fit to
the data, this indicates that there are no significant group differ-
ences in predicting the relationship of the trait to the dependent
measure. Cases in which the structural covariance model fit best
indicate that in addition to the path coefficients, the covariances
between variables are constrained to be equal. In some cases, the
structural residual model may provide the best fit, indicating that
all model parameters were equivalent across groups. Although the
best fitting model is reported here, for the purposes of the present
study, only the equivalence of the path coefficients was relevant,
because differences in variances, covariances, and residuals may
be due to sampling error or other artifacts that do not reflect and
are not substantively related to the relationships to be tested.
Model fit was tested for the four experimental conditions (75%
reliability/workload-mismatched; 75% reliability/workload-
matched; 95% reliability/workload-mismatched; 95% reliability/
workload-matched). Cases in which an adequate fit could not be
obtained for the unconstrained model were analyzed by fitting
separate models for low and high reliability conditions. If adequate
model fit could not be obtained for the separate reliability groups,
the analysis was conducted by collapsing across all automation
conditions.
Model fit was assessed using multiple indices, as a single index
only assesses a particular aspect of fit (Brown, 2006; Kline, 2005).
In addition to model
2
, the Root Mean Square Error of approx-
imation (RMSEA), the comparative fit index (CFI), and the
Tucker-Lewis Index (TLI) were used. The Akaike information
criterion (AIC) is also reported to permit comparison of relative fit
across models tested. Based on recommendations described in
Brown (2006), the criteria for a good fit was an RMSEA ⱕ.05,
and a CFI index and TLI index of .95 or greater. Criteria for a
moderate fit were an RMSEA between .05 and .10, and a CFI and
TLI index ranging from .90 to .95. These statistics are summarized
for each path analysis in Table 5.
Results are organized by dependent measure and trait. Both
unstandardized and standardized path coefficients are reported, but
standardized coefficients should only be interpreted when compar-
ing coefficients derived from the same sample (Kline, 2005).
When comparisons across groups are made, unstandardized values
should be used. This is because although unstandardized path
coefficients may be equivalent across groups, the standardized
Figure 4. Task-focused coping score as a function of automation reliability and workload-adaptiveness. Note.
Error bars are standard errors.
81
PERSONALITY AND ADAPTIVE AUTOMATION
values may differ due to differences in the variances across groups.
Note that across all conditions there were no statistically signifi-
cant effects involving traits on response time or on the proportion
of trials during block 4 on which participants chose to engage the
automation. These results are therefore not reported.
Performance Accuracy and Agreement With
Automation
Correlational analysis. Among the five traits, only Neurot-
icism and Conscientiousness significantly correlated with perfor-
mance, and only Neuroticism was significantly correlated with
agreement measures. Higher Neuroticism was related to lower
accuracy during blocks 2–4 and less agreement with correct au-
tomation in blocks 3 and 4 (see Table 6). No significant correla-
tions were observed for agreement with incorrect automation.
Conscientiousness was positively correlated with accuracy in
block 3.
Path analyses. Only Neuroticism was related to performance
accuracy and agreement with automation. Neuroticism predicted
less agreement with correct automation in the workload-matched
condition (i.e., the Neuroticism by workload adaptiveness interac-
tion—N ⫻WLA in Figure 6). For performance, comparison
across the four groups resulted in a model with poor fit, but a good
fit was obtained for the structural weights model comparing
workload-adaptiveness conditions at 95% reliability (see Figure 6).
Neuroticism predicted lower performance during block 3 (2 dis-
plays). For the 75% reliability group the structural residuals model
yielded a good fit. Neuroticism predicted lower performance in
block 2 (4 displays; Figure 6), with a marginally significant effect
in block 4.
Note that the failure to obtain a good fit across the four groups,
and the good fit obtained for the reliability groups when analyzed
separately, indicates that the relationship of Neuroticism to per-
formance depends on automation reliability (i.e., a trait by reli-
ability interaction—N ⫻R in Figure 6). The different patterns of
significant paths for the two reliability conditions indicates that
Neuroticism effects also depended on task demand: at 95% reli-
ability performance decrements were observed when the demands
were lower (the two displays in block 3); while at 75% reliability
it was during the higher demand blocks that Neuroticism was
related to poorer performance. Note this was true regardless of
whether the automation was engaged (i.e., across workload-
matched/mismatched conditions).
Subjective Response to Automation: Perceived
Workload, Stress, and Coping
Perceived Workload
Correlational analyses. There were no statistically signifi-
cant correlations of Global Workload, Mental Demand, or Effort
across the five traits (p⬎.05 in each case). Temporal Demand was
positively correlated with Neuroticism in block 4, and Perfor-
mance Workload correlated positively with Neuroticism (blocks
1–3) and negatively with Extraversion (blocks 2 and 3; see Table
6). Frustration was negatively correlated with Extraversion (block
2) and positively correlated with Neuroticism (blocks 1 and 2) and
Agreeableness (block 3), although the correlation with Agreeable-
ness should be interpreted with caution because the number of
significant correlations with that trait could have occurred by
chance. There were no statistically significant workload correla-
tions involving Conscientiousness or Openness.
Path analyses. There were no statistically significant effects
for Global Workload or Mental Demand involving any of the five
traits. Neuroticism predicted greater Temporal Demand in block 4
(Figure 7a). Across reliability and adaptiveness groups, Neuroti-
cism and Openness predicted less Effort after blocks 1 and 4,
respectively. Neuroticism predicted greater Frustration after block
Figure 5. Emotion-focused and Avoidant Coping as a function of block. Note. Error bars are standard errors.
82 SZALMA AND TAYLOR
1 (2 displays) and after the two 4-display blocks (2 and 4).
Agreeableness predicted higher Frustration in block 3. Extraver-
sion predicted less Frustration in block 2 but higher Frustration in
block 3 in the 75% reliability/workload-mismatched condition
(indicating an interaction effect—E ⫻R⫻WLA in Figure 7a).
Across all reliability and adaptiveness conditions, Neuroticism
and Agreeableness predicted greater Performance Workload (Fig-
ure 7b). The relationship of Conscientiousness to Performance
Workload varied as a function of automation reliability. In the 75%
reliability/workload-matched Condition Conscientiousness pre-
dicted greater Performance Workload in block 3. In the 95%
reliability/workload-matched Condition Conscientiousness pre-
dicted less Performance Workload in block 1 and greater workload
in block 3.
Stress Scales
Correlational analyses. Neuroticism was positively corre-
lated and Extraversion negatively correlated with pretask Distress
and Worry and posttask state for each of the four blocks. Consci-
entiousness was negatively correlated to pretask Distress but not
with the posttask state. Neuroticism was negatively correlated with
Task Engagement for pretask state and for all four blocks (see
Table 6). Task Engagement correlated positively with Conscien-
tiousness for pretask state and post-Task Engagement after block
3, although there is a 6% probability that these significant corre-
lations occurred by chance. The trait–stress relationships were
generally consistent with previous research (Matthews et al.,
1999). That is, higher Neuroticism was associated with higher
levels and Extraversion and Conscientiousness with lower levels
of stress.
Path analyses. Across reliability and adaptiveness condi-
tions, higher Neuroticism was associated with lower and Consci-
entiousness with higher pre-Task Engagement. Neuroticism was
associated with lower post-Task Engagement in block 3, with a
marginally significant negative effect in block 4 (see Figure 8a).
The other four traits were not significantly related to post-Task
Table 5
Summary of Fit Statistic for Path Analyses for the Big Five Traits (N⫽161)
Exp. cond DV Model fit
2
CFI TLI RMSEA AIC
Neuroticism
95 p(C) SW
2
(7) ⫽8.06, p⫽.33 .99 .98 .05 (90% CI: ⬍.00–.15) 74.06
75 p(C) SR
2
(23) ⫽24.38, p⫽.38 .99 .99 .03 (90% CI: ⬍.00–.10) 58.38
M Agg-C SW
2
(3) ⫽2.44, p⫽.49 1.00 1.06 ⬍.00 (90% CI: ⬍.00–.18) 32.44
All TD SR
2
(60) ⫽62.18, p⫽.40 .99 .99 .02 (90% CI: ⬍.00–.05) 102.18
All PW SR
2
(61) ⫽5.53, p⫽1.00 1.00 1.06 ⬍.00 (90% CI: ⬍.00–⬍.00) 43.53
All Effort SR
2
(61) ⫽61.80, p⫽.45 1.00 1.00 .01 (90% CI: ⬍.00–.05) 99.80
All F SR
2
(60) ⫽0.00, p⫽1.00 1.00 1.05 ⬍.00 (90% CI: ⬍.00–⬍.00) 40.00
All TE SR
2
(84) ⫽93.55, p⫽.22 .99 .99 .03 (90% CI: ⬍.00–.05) 141.55
All D SC
2
(60) ⫽62.23, p⫽.40 1.00 1.00 .02 (90% CI: ⬍.00–.05) 158.23
75/Mis 75/M 95/Mis 95/M W U
2
(12) ⫽23.16, p⫽.03 1.00 .94 .08 (90% CI: .03–.12) 215.16
All TC SC
2
(43) ⫽42.55, p⫽.49 1.00 1.00 ⬍.00 (90% CI: ⬍.00–.05) 116.55
All EC SC
2
(43) ⫽51.34, p⫽.18 .99 .99 .04 (90% CI: ⬍.00–.07) 125.32
All AC SC
2
(43) ⫽45.94, p⫽.35 .99 .99 .02 (90% CI: ⬍.00–.06) 119.94
Extraversion
All D SC
2
(64) ⫽67.31, p⫽.36 1.00 1.00 .02 (90% CI: ⬍.00–.05) 155.31
75/mis 75/M 95/M W U
2
(8) ⫽12.83, p⫽.12 1.00 .96 .06 (90% CI: ⬍.00–.12) 212.83
All EC SC
2
(35) ⫽31.00, p⫽.66 1.00 1.01 ⬍.00 (90% CI: ⬍.00–.05) 121.00
Conscientiousness
All TE SC
2
(64) ⫽71.94, p⫽.23 .99 .99 .03 (90% CI: ⬍.00–.06) 159.94
All D SC
2
(68) ⫽78.778, p⫽.175 .98 .98 .03 (90% CI: ⬍.00–.06) 158.78
Agreeableness
All PW SC
2
(39) ⫽48.18, p⫽.15 .94 .94 .04 (90% CI: ⬍.00–.07) 130.18
All F SR
2
(60) ⫽74.37, p⫽.10 .93 .95 .04 (90% CI: ⬍.00–.07) 114.37
75/M W U
2
(8) ⫽12.01, p⫽.15 1.00 .97 .06 (90% CI: ⬍.00–.12) 212.01
95 M EC U
2
(4) ⫽6.24, p⫽.18 .97 .96 .06 (90% CI: ⬍.00–.15) 158.24
Openness
All Effort SR
2
(61) ⫽65.97, p⫽.31 .98 .98 .02 (90% CI: ⬍.00–.06) 103.97
All D SC
2
(60) ⫽62.64, p⫽.38 1.00 1.00 .02 (90% CI: ⬍.00–.05) 158.64
Note. 95 ⫽95% reliability; 75 ⫽75% reliability; Agg-C ⫽Agreement with automation, recommendation correct; p(C) ⫽proportion correct; TD ⫽
Temporal Demand; PW ⫽Performance Workload; F ⫽Frustration; TE ⫽Task Engagement; D ⫽Distress; W ⫽Worry; TC ⫽Task-focused coping; EC ⫽
Emotion-focused Coping; AC ⫽Avoidant Coping; SW ⫽structural weights path model; SR ⫽structural residual path model; SC ⫽structural covariance
path model; U ⫽Unconstrained path model.
ⴱ
p⬍.05.
ⴱⴱ
p⬍.01.
ⴱⴱⴱ
p⬍.001.
83
PERSONALITY AND ADAPTIVE AUTOMATION
Engagement. For Distress, all traits except Agreeableness signifi-
cantly predicted pretask state in the expected direction: Neuroti-
cism predicted higher pretask Distress, and Extraversion, Consci-
entiousness, and Openness each predicted lower pretask Distress.
However, none of these traits predicted posttask state, indicating
that the relationships of traits to posttask Distress (i.e., the first
order correlations in Table 6) were mediated by pretask state.
Worry was the only one of the three stress dimensions in which
the relationships of traits to posttask state were moderated by
reliability and adaptiveness conditions. In the 95% reliability
workload-matched condition, Neuroticism predicted higher post-
task Worry in the first block (2 displays), with a marginally
significant effect for the last block (4 displays, Figure 8b). Neu-
roticism also predicted increased Worry in block 1 for the 75%
reliability groups and in block 2 for the 75% reliable workload-
matched group (4 displays, automation engaged).
Agreeableness predicted less Worry after block 3 and increased
Worry after block 4 in the 75% reliability workload-matched
condition (see Figure 8c). Thus, after the experience of relatively
poor automation under the demanding conditions of block 2, those
higher in Agreeableness exhibited a reduction in the cognitive
symptoms of stress in the subsequent block when the automation
was disengaged and task demands were decreased.
The relationship of Extraversion to posttask Worry varied as a
function of task demand, reliability, and adaptiveness. When au-
tomation was matched to task demand, Extraversion predicted less
Worry in block 2 (4 displays) of the 75% reliability condition and
increased Worry in blocks 2, 3, and 4 of the 95% reliability
condition (Figure 8c). In the 75% reliable workload-mismatched
condition, Extraversion predicted less Worry in block 1. Essen-
tially, more extraverted individuals reported less Worry when the
automation was poor or the task demands were high, and they
reported more Worry when demand was low or when highly
reliable automation was provided.
Coping Scales
Correlational analyses. Task-focused coping did not signif-
icantly correlate with the traits. Neuroticism was positively corre-
lated with both emotion-focused and avoidant coping. Extraver-
sion was associated with less avoidant coping in blocks 2 and 3,
and less emotion-focused coping in blocks 2–4 (see Table 6).
Agreeableness was associated with less avoidant coping in block 1,
but there is an 11% probability that this pattern could have oc-
curred by chance. No significant correlations involving Conscien-
tiousness and Openness were observed.
Path analysis. Higher Neuroticism was associated with less
task-focused coping in blocks 1 and 4 across all experimental
conditions (see Figure 9a). The other four traits were not signifi-
cantly related to this form of coping. Across all experimental
conditions emotion-focused coping was positively related to Neu-
roticism in blocks 2 and 3 and negatively related to Extraversion in
block 2. Agreeableness predicted less emotion-focused coping in
block 2, but only in the 95% reliable workload-matched automa-
tion condition (i.e., 4 displays, automation engaged). Across all
experimental conditions avoidant coping was positively related
to Neuroticism in the two display conditions (blocks 1 and 3)
and negatively related to Extraversion in block 2 (Figure 9b).
Table 6
Correlations of the Five Factors to Dependent Variables (N⫽161)
Pretask state Block 1 (2 displays) Block 2 (4 displays) Block 3 (2 displays) Block 4 (4 displays)
Neuroticism
Accuracy — ⫺.09 ⫺.15
ⴱ
⫺.23
ⴱⴱ
⫺.16
ⴱ
Agreement-RC —
Mismatched — ⫺.18 — ⫺.26
ⴱ
⫺.06
Matched — — ⫺.08 — ⫺.30
ⴱ
Temporal Demand — .03 .05 .13 .21
ⴱⴱ
Perceived Performance — ⫺.17
ⴱ
⫺.15
ⴱ
⫺.17
ⴱ
⫺.13
Frustration — .22
ⴱⴱ
.24
ⴱⴱ
.11 .18
ⴱ
Task Engagement ⫺.30
ⴱⴱⴱ
⫺.23
ⴱⴱ
⫺.19
ⴱ
⫺.32
ⴱⴱⴱ
⫺.28
ⴱⴱⴱ
Distress .54
ⴱⴱⴱ
.34
ⴱⴱⴱ
.32
ⴱⴱⴱ
.32
ⴱⴱⴱ
.30
ⴱⴱⴱ
Worry .32
ⴱⴱⴱ
.41
ⴱⴱⴱ
.50
ⴱⴱⴱ
.42
ⴱⴱⴱ
.43
ⴱⴱⴱ
Emotion-focused coping — .33
ⴱⴱⴱ
.36
ⴱⴱⴱ
.43
ⴱⴱⴱ
.38
ⴱⴱⴱ
Avoidant coping — .35
ⴱⴱⴱ
.32
ⴱⴱⴱ
.33
ⴱⴱⴱ
.28
ⴱⴱⴱ
Extraversion
Perceived Performance — .10 .17
ⴱ
.16
ⴱ
.03
Frustration — .04 ⫺.17
ⴱ
.04 ⫺.03
Distress ⫺.39
ⴱⴱⴱ
⫺.23
ⴱⴱ
⫺.23
ⴱⴱ
⫺.16
ⴱ
⫺.18
ⴱ
Worry ⫺.19
ⴱ
⫺.22
ⴱⴱ
⫺.22
ⴱⴱ
⫺.17
ⴱ
⫺.26
ⴱⴱⴱ
Emotion-focused coping — ⫺.09 ⫺.18
ⴱ
⫺.19
ⴱ
⫺.17
ⴱ
Avoidant Coping — ⫺.12 ⫺.22
ⴱⴱ
⫺.17
ⴱ
⫺.14
Conscientiousness
Accuracy — .04 .03 .16
ⴱ
.04
Task Engagement .19
ⴱ
.08 .13 .16
ⴱ
.10
Distress ⫺.24
ⴱⴱ
⫺.14 ⫺.09 ⫺.09 ⫺.09
Agreeableness
Frustration — .12 .07 .18
ⴱ
.14
Avoidant coping — ⫺.16
ⴱ
⫺.09 ⫺.12 .07
ⴱ
p⬍.05.
ⴱⴱ
p⬍.01.
ⴱⴱⴱ
p⬍.001.
84 SZALMA AND TAYLOR
Agreeableness predicted less avoidant coping in block 1 for the
95% workload-mismatched group, but more avoidant coping in
block 4 for the 75% workload-matched group. In both cases the
automation was engaged. Conscientiousness and Openness
were not significantly related to either emotion-focused or
avoidant coping.
Discussion
The purpose for the present study was to investigate the joint
effects of human and automation/task characteristics on perfor-
mance, workload, stress, coping, and the degree to which operators
agreed with the automation. Among the independent variables,
task demand was most strongly related to participant response. For
traits, most of the significant relationships were associated with
Neuroticism and Extraversion (e.g., see Table 6), although all five
traits were associated with differences on at least one measure of
perceived workload and stress.
However, the patterns and magnitudes of the associations varied
as a function of trait, dependent measure, and experimental con-
dition. Indeed, across dependent variables, if a significant corre-
lation was observed in the correlational analysis, it was generally
also significant in the path analysis. The converse was not true. For
several scales (e.g., Effort, task-focused coping, the relationship
between Worry and Agreeableness and between Performance
Workload and Conscientiousness) there were no significant corre-
lations but there were significant paths.
This is possible because first order correlations reflect the rela-
tionship between two variables, without the removal of shared
variance with other variables. Path coefficients, in contrast, reflect
association between two variables after controlling for the variance
shared with other variables in the model. Path coefficients are thus
analogous to the coefficients in multiple regression. Such cases of
masking of task moderation effects in the simple correlational
analyses lends support to Szalma’s (2008, 2009) argument that
task and person characteristics should not be examined in isolation
but that instead their joint effects should be systematically ex-
plored.
Performance
Task/Automation Characteristics
Automation of higher reliability increased performance accu-
racy and reduced response time, but the workload-adaptiveness of
the automation did not have a significant impact on these variables.
Performance was improved when high reliability automation was
provided, regardless of the level of demand imposed by the task. In
addition, the rate of agreement with correct automation was higher
for those in the 95% reliable automation relative to participants in
the 75% reliable condition. The pattern was reversed for agreement
with incorrect automation.
These results were generally consistent with previous research
regarding automation reliability (Parasuraman et al., 2000), that
complacency is higher when both the automation and task de-
mands are high. That is, agreement with incorrect automation was
substantially higher at 95% as compared to 75% reliability, but
only in the 4-display condition of block 2. In contrast, the reverse
trend was observed in the initial 2-display block (block 1). This
pattern may reflect greater reliance at high task load when auto-
mation is reliable. It may be that with four displays there was
insufficient time to thoroughly scan the display, inducing those in
the high reliability condition to rely more on the aid. At the lower
demand there may have been more time to scan the “raw data” to
cross-check the automation.
Trait Effects
Accuracy. It was predicted that Neuroticism should be neg-
atively related to performance and that the other traits would show
positive relationships. Only Neuroticism and Conscientiousness
were significantly related to performance. The pattern of results
indicates that providing highly reliable automation attenuates the
negative association between Neuroticism and performance accu-
racy under demanding conditions (i.e., the relation of Neuroticism
to performance in block 2 was restricted to the 75% reliable
automation conditions). The performance deficits associated with
high Neuroticism may be due to fewer resources available for task
performance, and reliable automation may benefit high Neuroti-
cism individuals by freeing up resources for allocation to the task.
With respect to Conscientiousness, this trait is correlated with job
performance, but it was only weakly related to accuracy in this
study. It may be that this general trend does not always extend to
performance of a specific task.
The absence of performance effects for Extraversion may be due
to the cognitive patterning of the trait (Matthews et al., 2003). The
task required divided attention and a time constraint for participant
response; conditions favorable to those higher in Extraversion.
However, this trait is also associated with poorer vigilance. The
advantages in divided attention for more extraverted participants
and the stimulating effects of observing multiple video-based
Figure 6. Significant paths for performance accuracy and agreement with
correct automation. Note.
ⴱ
p⬍.05;
ⴱⴱ
p⬍.01;
ⴱⴱⴱ
p⬍.001;
#
.05 ⬍p⬍
.10. N⫽Neuroticism; R ⫽Reliability; N⫻R⫽Neuroticism by reliability
interaction; WLA ⫽workload-adaptive variable; N⫻WLA ⫽Neuroti-
cism by workload-adaptiveness interaction; M ⫽workload-matched con-
dition; 75% ⫽75% reliable automation condition; 95% ⫽95% reliable
automation condition.
85
PERSONALITY AND ADAPTIVE AUTOMATION
Figure 7 (opposite).
86 SZALMA AND TAYLOR
threat stimuli may have therefore been offset by the negative
effects of the monitoring requirement.
The nature of the task itself may also have exerted an influence
on the pattern of relationships between the five traits and perfor-
mance. Threat stimuli and ambiguity regarding threat are particu-
larly relevant and salient to individuals higher in Neuroticism. It
may be that the performance effects were dominated by threat
characteristics of the task rather than of the automation, facilitating
the relation of Neuroticism to performance and attenuating the
relationships of the other four factors. An important issue for
future research is whether the pattern of performance effects
observed herein extends to automation tasks in which the events to
be controlled are not inherently threatening or as likely to elicit an
emotional response.
Agreement with automation. Agreement with automation
was generally unrelated to personality. The exception was the
negative relationship between Neuroticism and agreement with
accurate automation during block 4. This relationship may be
due to the propensity of high Neuroticism individuals to be
cautious in their decision making (Matthews, 2008). However,
the fact that the effects were limited to the workload-matched
group in block 4 suggests that the experience of automation in
block 2, in which the demands were similar to those in block 4,
set the expectancies for the latter block more strongly for those
high versus low in Neuroticism. However, the relation of Neu-
roticism to agreement in block 4 did not depend on the reli-
ability of the automation in block 2, suggesting that the influ-
ence of the trait may reflect lower trust based on general
expectancies rather than previous experience of interaction with
the automated agent.
The other four traits were not significantly related to agree-
ment with automation. For Extraversion, this result is consistent
with those of Singh et al. (1993b), who reported that this trait
was unrelated to automation-induced complacency. However, it
was expected that Agreeableness would be associated with
agreement, as trust is a major component of this trait. Those
higher in Agreeableness may not have differed from those
lower on the trait in sensitivity to the trustworthiness of the
automation. A second possibility is that trait effects were at-
tenuated because participants could cross-check the automation
with the “raw data” itself (i.e., the video clips). It was expected
that providing such information would induce more individual
differences in agreement, but perhaps in some task domains the
information suppresses individual differences. Future investi-
gations should examine whether and how access to the “raw”
information moderates the relationships of traits to agreement
with decision automation.
Subjective Response to Automation: Perceived
Workload, Stress, and Coping
Task/Automation Characteristics
Only task demand (2 versus 4 displays) exerted an influence on
perceived workload, with higher demand generally associated with
higher workload scores. Note that for automation reliability there
were performance effects but no significant differences in per-
ceived workload, a pattern of workload insensitivity (Parasuraman
& Hancock, 2001). It may be that participants used the reliable
automation to verify the results of their own scanning of the
display, improving performance but not relieving their effort or
mental work.
Reliability and workload-adaptiveness also did not significantly
affect posttask stress: only the level of task demand exerted an
effect on the cognitive state measures. However, the pattern of
changes may not reflect demand per se, but may result from
experience with the task. Distress increased to a maximum in block
2 (4 displays) and fell off in the last two blocks (2 and 4 displays,
respectively). Worry declined over blocks, and Task Engagement
remained stable until blocks 3 (2 displays) and 4 (4 displays), when
it declined and increased, respectively.
This pattern of results are consistent with those of Szalma et al.
(2004), who reported that symptoms of Distress increased over
period on watch but declined toward the end of a visual sustained
attention task, and that Task Engagement was relatively stable
early in the watch but fluctuated during the latter half of the vigil.
The changes with respect to block in this study are consistent with
the conclusion of Szalma et al. (2004) that stress symptoms change
over time on task and that the pattern of change depends on which
dimension of stress is measured.
With respect to coping, automation of higher reliability induced
less task-focused coping when it was workload-mismatched, pos-
sibly because in that condition the automated aid was provided
when task demands were lower and participants could divert
attention away from the task without sacrificing performance. An
interesting question for future consideration is whether such a
reduction in task-focused coping is related to complacency or
misuse of automation. Emotion-focused and avoidant coping var-
ied as a function of task demand/block, but not automation con-
dition. However, this may be due to the interactive effects with
personality.
Trait Effects
Perceived workload. In general, where significant effects
were observed, Neuroticism and Agreeableness predicted greater
Figure 7 (opposite). a. Significant paths for Temporal Demand, Effort, and Frustration. Note.
ⴱ
p⬍.05;
ⴱⴱ
p⬍.01;
ⴱⴱⴱ
p⬍.001; N⫽Neuroticism; E ⫽
Extraversion; A ⫽Agreeableness; O ⫽Openness; R ⫽Reliability; WLA ⫽workload-adaptive variable; E ⫻R⫻WLA ⫽Extraversion by reliability by
workload-adaptiveness interaction; Mis ⫽workload-mismatched condition; 75% ⫽75% reliable automation condition; 95% ⫽95% reliable automation
condition; TD ⫽Temporal Demand; F⫽Frustration; All ⫽collapsed across groups. b. Significant paths for Performance Workload. Note.
ⴱ
p⬍.05;
ⴱⴱ
p⬍.01;
ⴱⴱⴱ
p⬍.001; N⫽Neuroticism; C ⫽Conscientiousness; A ⫽Agreeableness; R ⫽Reliability; WLA ⫽workload-adaptive variable; C ⫻R⫻
WLA ⫽Conscientiousness by reliability by workload-adaptiveness interaction; M ⫽workload-matched condition; 75% ⫽75% reliable automation
condition; 95% ⫽95% reliable automation condition; PW ⫽Performance Workload; All ⫽collapsed across groups.
87
PERSONALITY AND ADAPTIVE AUTOMATION
workload and Openness predicted lower workload. These results
conformed to expectation for Neuroticism and Openness, but were
in the opposite direction of those expected for Agreeableness. Note
that in both cases in which the latter trait predicted higher work-
load it did so in one of the two later blocks (see Figures 7a and 7b).
It may be that experience with the task induces increased workload
for individuals higher in Agreeableness. However, the reasons for
this are not clear and remain to be determined.
Figure 8 (opposite).
88 SZALMA AND TAYLOR
In both reliability groups, Conscientiousness predicted an in-
crease in Performance Workload in block 3 when no automation
was provided and following the higher demand of block 2 when
automation was engaged. This may be a workload transition effect
(cf., Cox-Fuenzalida et al., 2004), as workload increased as a
function of Conscientiousness when demand changed from a
higher to a lower level. The more demanding condition may have
suppressed differences as a function of the trait. However, for
those higher in Conscientiousness, transition to the less demanding
task may have facilitated more self-evaluative cognitions regard-
ing their performance during the previous block, thereby increas-
ing their Performance Workload.
Note that for Conscientiousness, the only workload effect was
for Performance Workload under conditions of lower demand and
no automation support. The specificity of the relation of the trait to
this dimension of workload may be due to the greater concern of
conscientious individuals with their performance and their ten-
dency to evaluate themselves against their performance goals.
However, provision of automation or higher task demands atten-
uated the relationship of Conscientiousness to this dimension of
workload, perhaps because providing an aid (even an imperfect
one) increased the confidence of conscientious individuals.
Extraversion predicted less Frustration when reliability was low
and task demand high (block 2 of the 75% reliable workload-
mismatched condition) but greater Frustration when unreliable
automation was removed and task demand decreased (block 3 of
the 75% reliable workload-mismatched condition). This pattern is
what one might expect if extraverted individuals were responding
to the change in stimulation offered by the task. Because they
prefer stimulating environments, the combination of lower demand
and the vigilance requirement may have been aversive to these
individuals. That is, the higher Frustration in block 3 likely reflects
the reduction in task demand from 4 to 2 displays. In contrast to
Neuroticism, individuals higher in Extraversion seemed to expe-
rience a workload benefit from higher task demand.
Stress. In general, the observed stress effects were in the
expected direction. Neuroticism was associated with less Task-
Engagement and greater levels of Distress and Worry, Extraver-
sion correlated negatively with Distress and Worry, and Consci-
entiousness was related to greater Task-Engagement and less
Distress. However, the path analyses revealed that the significant
relationships between traits and the Task Engagement and Distress
components of stress were primarily in pretask state. Task char-
acteristics were therefore relatively weak moderators of the
personality–stress relationship along the energetic and affective
dimensions. It was for the cognitive component (Worry) that task
factors moderated the trait–stress relationships, indicating that
nontask thoughts and self-reflective processes are more sensitive
than the emotional or energetic facets of stress to interactive
effects of trait and task factors.
Two particularly interesting findings in regard to task–
personality interactive stress effects are the dependencies of the
relation of Agreeableness and Extraversion to posttask Worry on
reliability, adaptiveness, and task demand. Agreeableness was
associated with less Worry after block 3 and more Worry after
block 4 in the 75% reliable workload-matched condition. Note that
in the latter block the automation was 95% reliable. Apparently,
removal of unreliable automation and the reduction in demand in
the subsequent block relieved the cognitive aspects of stress for
agreeable individuals, but the provision of reliable automation in
the fourth block increased their stress. It is unlikely that this was
a response to increased task demand, as Agreeableness was not
related to Worry in the second block, in which there were four
displays. It is also unlikely that this reflects a change in trust
behavior, as performance and agreement with automation did not
vary as a function of Agreeableness. As this pattern was not
observed in the group that received 95% reliable automation in
previous blocks, it may be that the 75% reliable automation in
block 2 set the expectancies of those higher in Agreeableness,
thereby increasing their Worry in block 4 when task demands were
the same as those experienced in block 2.
However, there are two issues to be resolved by future research
before accepting such an explanation. First, it is not clear why the
effects of prior experience with 75% reliable automation on re-
sponse to 95% reliable automation were confined to the cognitive
component of stress, although it may be due to the central role of
cognitive expectancy in trust (Lee & See, 2004; Rotter, 1980).
Second, if prior experience influenced their cognitive stress state,
it is unclear why this experience would also not affect their
performance or agreement with the automated aid in those condi-
tions. It may be that the four-display condition was sufficiently
demanding it outweighed the trait effects on performance and
agreement, but that changes in posttask Worry were dominated by
the trait, suggesting that traits may predict stress at levels of
demand lower than those associated with performance decrements
(Szalma, 2008). That is, individuals high in Agreeableness main-
tained their performance and agreement behavior at levels similar
to those low on the trait, but this was accomplished at the cost of
greater Worry.
Extraversion was related to less Worry at 75% reliability
when the automation was engaged. However, in the 95% reli-
ability workload-matched condition, Extraversion predicted
greater Worry when task load was increased from two to four
displays (block 2) and when task load decreased from four to
Figure 8 (opposite). a. Significant paths for Distress and Task Engagement. Note.
ⴱ
p⬍.05;
ⴱⴱ
p⬍.01;
ⴱⴱⴱ
p⬍.001;
#
.05 ⬍p⬍.10. N⫽Neuroticism;
E⫽Extraversion; C ⫽Conscientiousness; O ⫽Openness; TE ⫽Task Engagement; D ⫽Distress; All ⫽collapsed across groups. b. Significant paths
for Worry and Neuroticism. Note.
ⴱ
p⬍.05;
ⴱⴱ
p⬍.01;
ⴱⴱⴱ
p⬍.001;
#
.05 ⬍p⬍.10. N⫽Neuroticism; R ⫽Reliability; WLA ⫽workload-adaptive
variable; N⫻R⫻WLA ⫽Neuroticism by reliability by workload-adaptiveness interaction; M ⫽workload-matched condition; Mis ⫽workload-
mismatched condition; 75% ⫽75% reliable automation condition; 95% ⫽95% reliable automation condition; W ⫽Worry; All ⫽collapsed across groups.
c. Significant paths for Worry and Extraversion and Agreeableness. Note.
ⴱ
p⬍.05;
ⴱⴱ
p⬍.01;
ⴱⴱⴱ
p⬍.001;
#
.05 ⬍p⬍.10; E ⫽Extraversion; A ⫽
Agreeableness; R ⫽Reliability; WLA ⫽workload-adaptive variable; E ⫻R⫻WLA ⫽Extraversion by reliability by workload-adaptiveness interaction;
A⫻R⫻WLA ⫽Agreeableness by reliability by workload-adaptiveness interaction; M ⫽workload-matched condition; Mis ⫽workload-mismatched
condition; 75% ⫽75% reliable automation condition; 95% ⫽95% reliable automation condition; W ⫽Worry; All ⫽collapsed across groups.
89
PERSONALITY AND ADAPTIVE AUTOMATION
Figure 9. Significant paths for Task-focused and Emotion-focused coping. Note.
ⴱ
p⬍.05;
ⴱⴱ
p⬍.01;
ⴱⴱⴱ
p⬍.001; N⫽Neuroticism; E ⫽Extraversion; A ⫽Agreeableness; R ⫽Reliability; WLA ⫽
workload-adaptive variable; M ⫽workload-matched condition; A ⫻R⫻M⫽Agreeableness by reliability
by workload-matched condition interaction; 95% ⫽95% reliable automation condition; TC ⫽Task-focused
coping; EC ⫽Emotion-focused coping; All ⫽collapsed across groups. b.Significant paths for Avoidant
coping. Note.
ⴱ
p⬍.05;
ⴱⴱ
p⬍.01;
ⴱⴱⴱ
p⬍.001;
#
.05 ⬍p⬍.10; N⫽Neuroticism; E ⫽Extraversion;
A⫽Agreeableness; R ⫽Reliability; WLA ⫽workload-adaptive variable; M ⫽workload-matched
condition; Mis ⫽workload-mismatched condition; M ⫽workload-matched condition; E ⫻R⫻M⫽
Extraversion by reliability by workload-matched condition interaction; A ⫻R⫻M⫽Agreeableness by
reliability by workload-matched condition interaction; 5% ⫽75% reliable automation condition; 95% ⫽
95% reliable automation condition; AC ⫽Avoidant coping; All ⫽collapsed across groups.
90 SZALMA AND TAYLOR
two displays (block 3). Thus, at higher task demand (block 2)
Extraversion predicted greater Worry at high reliability and less
Worry at low reliability. This result is consistent with the
hypothesis that Extraversion would be positively related to
stress when the task provides less stimulation and more of a
monitoring requirement, although this pattern was reversed in
the final block in which participants experienced the higher task
load with high reliability automation. This latter effect may be
due to the control participants could exert on automation en-
gagement and the monetary reward for good performance.
Extraversion is positively correlated with sensitivity to rewards
(Matthews et al., 2003).
Coping. Only Neuroticism was related to task-focused
coping, in the expected direction. Under the more demanding
conditions of block 2, both emotion-focused and avoidant cop-
ing decreased as a function of Extraversion, and emotion-
focused coping increased as a function of Neuroticism. How-
ever, the relation of this trait to the three coping strategies also
manifested in the lower demand conditions, and the effects were
not stronger for higher or lower demands. Apparently, high
Neuroticism is associated with use of these strategies regardless
of the reliability and workload-adaptiveness of the automated
aid. The decline in emotion-focused and avoidant coping as a
function of Extraversion in block 2 (higher demand) may result
from a strategy of diverting cognitive resources away from
emotion-focused and avoidant processing to deal with the in-
creased task demands. Conscientiousness and Openness were
not related to coping strategy, suggesting that the relationship
of these variables to workload and stress depend on factors
other than how they typically cope with demanding tasks.
The pattern of results for the relationship of Agreeableness to
coping strategy indicates that individuals high on this trait are
vulnerable to the less effective coping strategies when demand is
high or the automation is unreliable, situations in which their
Worry also increased. Less reliable automation may therefore be
associated with a “hidden cost” (Hockey, 1997) of performance in
terms of stress (Worry) and avoidant coping that persists in a
subsequent condition in which the automation is reliable. As this
form of coping can be associated with poor performance (Mat-
thews & Campbell, 1998), Agreeableness may be a risk factor for
automation disuse when the individual has prior experience with
unreliable automation.
Task and Person Characteristics: Relative Effects and
Interactions
Task variables exerted stronger effects on performance than
person characteristics (see Figure 10a), and among the former
block accounted for more variability (13–21%) than workload-
adaptiveness or reliability. Traits were associated with small to
medium effects on perceived workload (2–10%), with larger ef-
fects associated with the 2-display blocks. Task variables exerted
a strong effect on workload, particularly block/task demand (17–
69%). As can be seen in Figure 10b, however, task variables
tended to dominate the scales related to appraisals of task demand
(i.e., mental and temporal demand) and traits dominated the scales
reflecting appraisals related to the self (i.e., performance workload
and frustration). The latter effects were related to Neuroticism,
Extraversion, and Conscientiousness.
For posttask stress, task characteristics dominated over traits in
terms of the proportion of variance accounted for in the dependent
measures (see Figure 10c), with block/task demand exerting the
only substantial influence. Note, however, that the contribution of
block versus traits was more closely matched on the Distress scale,
and these were larger for the first two bocks (5–9%) relative to the
last two blocks (1–6%). Traits primarily affected posttask stress
indirectly via pretask state. For instance Neuroticism and Extra-
version accounted for 9–32% of the variance in pretask state.
Task variables were weakly associated with coping (1–3%
for task-focused coping), with the exception of block (11% for
task-focused coping). In contrast, traits exerted a medium to
large effect, with Neuroticism accounting for 4–10% of the
variance in emotion-focused and 2–8% of the variance in
avoidant coping. The stronger effects occurred under the less
demanding (2-display) blocks. Extraversion was associated
with a small (2%) effect on emotion-focused coping but a
medium effect (6–8%) on avoidant coping. Agreeableness was
most strongly related to coping, accounting for 7–11% of the
variance in avoidant coping (2-display condition) and 8% of the
variance in emotion-focused coping (4-display condition). Au-
tomation characteristics did not generally moderate the trait-
coping relations, with the exception of Agreeableness. It may be
that the interaction of Agreeableness with automation reliability
impacts the adoption of avoidant coping strategies in response
to the trustworthiness of the automated aid.
In sum, task demand seemed to exert the strongest influence
on outcome variables. Automation characteristics primarily af-
fected performance, but their relationship to stress and work-
load did emerge in the context of trait-independent variable
interactions. The pattern of relationships of traits to dependent
variables generally occurred for those outcome measures that
are closely related to the trait (e.g., Conscientiousness and
Performance Workload; Neuroticism and stress/workload) and
under experimental conditions to which a trait is most sensitive
(e.g., level of stimulation for Extraversion, task overload for
Neuroticism, prior exposure to poor automation for Agreeable-
ness, no automation to support performance for Conscientious-
ness). Differences as a function of trait generally manifested
more strongly as a function of task demand rather than auto-
mation characteristics. Whether the pattern of interactive ef-
fects over blocks is due to demand transitions or time on task is
a matter for future research.
Incorporating Individual Differences into Models of
Automation: Implications for Research and
Application
The results of the present study indicate that the impact of
automation and task demand on participant response depends to a
substantial degree on human personality traits. A practical impli-
cation is that the identification of specific profiles to predict
human–automation interaction is unlikely to be useful as a generic
selection measure. No single profile emerged from this study that
was independent of task variables, and the complexity of the
interactive effects suggest that for selection tools to be effective,
the optimal trait profiles would need to be developed separately for
each level, type, and perhaps even domain of application of auto-
mation.
91
PERSONALITY AND ADAPTIVE AUTOMATION
The potential variability in personality effects as a function of
automation properties may lead some to conclude that it would be
best to exclude individual differences considerations altogether,
and to design the automation on general principles applied gener-
ically to all operators. However, the results of this and other work
(Merritt & Ilgen, 2008; Prinzel et al., 2005; Singh et al., 1993b;
Thropp, 2006) clearly indicate that to do so would create the
potential for inefficiencies in training and operation and possibly
Figure 10. Mean Proportion of variance accounted for in performance measures (A), workload measures (B),
and stress/coping measures (C) by task and personality trait variables.
92 SZALMA AND TAYLOR
increase the risk of performance failure. The results of the current
study indicate that one should not assume that variations in task
parameters will have a common, normative effect across individ-
uals (cf., Parasuraman & Manzey, 2010). Hence, as Szalma (2008,
2009) has noted, interface and training design should not be “one
size fits all,” but should instead incorporate the person character-
istics.
Szalma (2009) proposed initial guidelines for incorporating
individual differences into the design process (see Figure 11),
in which analyses of both the task and the person are used to
shape the design of the interface. However, the empirical da-
tabase required to support an adequate person-task analysis is
mostly lacking. Programmatic research is needed to identify the
complex interactions of task characteristics and the cognitive
and affective traits of individuals across different domains in
which automation is or may be used. Information from studies
such as the one reported herein could then be used to structure
how, whether, and when automation is implemented, which
would lead to the development of truly adaptive automation.
For instance, the current results indicate that the provision of
highly reliable automation under conditions of high demand
may attenuate performance decrements and the cognitive di-
mensions of stress for those higher in Neuroticism. On the other
hand, providing reliable automation support may be aversive to
more extraverted individuals, who may need adaptive automa-
tion that allows for active engagement in a task when demands
are lower.
The above considerations indicate the task confronting research-
ers, but the practitioner faces the challenge of how to account for
individual differences in automation design based on extant re-
search. A potential answer may be found by incorporating indi-
vidual differences into models of automation. Thropp (2006) in-
corporated a consideration of individual differences into the
Parasuraman et al. (2000) model for automation design. A syn-
Figure 11. General guidelines for incorporating individual differences
into the design of human-technology interfaces. (After Szalma, 2009).
Figure 10 (continued).
93
PERSONALITY AND ADAPTIVE AUTOMATION
thesis of this model and the steps recommended by Szalma
(2009) provide a sequence of steps for the practitioner, as
shown in Figure 12.
The first step is to evaluate both the task and user characteristics,
and to identify the skills required to successfully operate the
system. The next step is to identify or empirically determine
the relevant affective and cognitive traits that correlate with
these skills. A specific selection instrument for the operation of the
specific automated task could then be developed or, perhaps more
fruitfully, the analysis of person and task could be used to decide
what should be automated and what form it should take. Such
automation would adapt to the needs/preferences of the specific
user (Thropp, 2006), thereby incorporating individuation in design
that will likely emerge as an important factor in the development
of future human–technology interfaces (Hancock, Hancock, &
Warm, 2009).
Another potentially powerful application of individual differ-
ences research may lie in the development of better training
procedures for operators (Szalma, 2008, 2009). Individuals who
are high or low on a particular trait may benefit from some degree
of customized training for how to cope with the stress and work-
load imposed by different features of the automation. If individuals
have specific trait-based vulnerabilities to particular components
of an automated task, training procedures could be developed to
help them compensate for their vulnerability to a particular task
demand by learning to cope more effectively when it is encoun-
tered.
Summary and Conclusions
The current study confirmed that personality traits predict
differential response to automation. Clearly, independent exam-
ination of person and task characteristics provides a limited
view of the factors that influence operator response. Some
outcome measures were dominated by task factors (e.g., agree-
ment with automation, response time), others were dominated
by person factors (e.g., coping), and still others by interactive
effects (e.g., workload, stress, and for Neuroticism, perfor-
mance accuracy). Future research should therefore consider
both person and task characteristics, and how these interact. For
instance, the negative impact of reliable automation on the
workload and stress experienced by more extraverted operators
deserves more attention, as does the role of performance goals
in influencing the relationship of Conscientiousness to work-
load. In addition, the relationships between automation and
person characteristics should be investigated using tasks other
than threat detection.
Although the results of the current study indicate that there is a
relationship between personality and response to automation, only
two levels of reliability and the workload-adaptiveness of automa-
tion were investigated. How these traits relate to response across
different dimensions of automation (e.g., type, level, triggering
criteria; Parasuraman et al., 2000) remains to be determined. In
addition, there are traits other than the Five Factors that may be
relevant to human interactions with automated agents. Program-
matic consideration of how automation and human characteristics
Figure 12. Incorporation of individual differences into the Parasuraman et al. (2000) model of automation,
incorporating the guidelines described by Szalma (2009). Adapted from Parasuraman et al. (2000); Thropp
(2006), and Szalma (2009).
94 SZALMA AND TAYLOR
interact to influence system performance will serve to refine and
extend theoretical models of automation as well as the perfor-
mance and well-being of the human operators who use automated
systems.
References
Ackerman, P. L., & Heggestad, E. D. (1997). Intelligence, personality,
and interests: Evidence for overlapping traits. Psychological Bulletin,
121, 219–245.
Arbuckle, J. L. (2009). AMOS (version 18.0) [Computer Program]
Chicago: SPSS.
Barrick, M. R., Mount, M. K., & Judge, T. A. (2001). Personality and
performance at the beginning of the new millennium: What do we
know and where do we go next? Personality and Performance, 9,
9–29.
Block, J. (2001). Millennial contrarianism: The five-factor approach to
personality description 5 years later. Journal of Research in Person-
ality, 35, 98–107.
Breen, L. J., Endler, N. S., Prociuk, T. J., & Okada, M. (1978).
Person ⫻situation interaction in personality prediction: Some spe-
cifics of the person factor. Journal of Consulting and Clinical Psy-
chology, 46, 567–568.
Brown, T. A. (2006). Confirmatory factor analysis for applied re-
search. New York: Guilford Press.
Costa, P. T., & McCrae, R. R. (1992). NEO PI-R professional manual.
Lutz, FL: Psychological Assessment Resources.
Couch, L. L., Adams, J. M., & Jones, W. H. (1996). The assessment
of trust orientation. Journal of Personality Assessment, 67, 305–
323.
Cox-Fuenzalida, L. E., Swickert, R., & Hittner, J. B. (2004). Effects of
neuroticism and workload history on performance. Personality and
Individual Differences, 36, 447–456.
Evans, A. M., & Revelle, W. (2008). Survey and behavioral measure-
ments of interpersonal trust. Journal of Research in Personality, 42,
1585–1593.
Eysenck, H. J., & Eysenck, M. W. (1985). Personality and individual
differences: A natural science approach. New York: Plenum Press.
Funke, G., Matthews, G., Warm, J. S., & Emo, A. K. (2007). Vehicle
automation: A remedy for driver stress. Ergonomics, 50, 1302–
1323.
Goldberg, L. R., Johnson, J. A., Eber, H. W., Hogan, R., Ashton, M. C.,
Cloninger, C. R., & Gough, H. G. (2006). The international person-
ality item pool and the future of public-domain personality measures.
Journal of Research in Personality, 40, 84–96.
Hancock, P. A., & Chignell, M. H. (1988). Mental workload dynamics
in adaptive interface design. IEEE Transactions on Systems, Man and
Cybernetics, 18, 647–658.
Hancock, P. A., Hancock, G. M., & Warm, J. S. (2009). Individuation:
The N⫽1 revolution. Theoretical Issues in Ergonomics Science, 10,
481–488.
Hancock, P. A., & Warm, J. S. (1989). A dynamic model of stress and
sustained attention. Human Factors, 31, 519–537.
Hart, S. G., & Staveland, L. E. (1988). Development of NASA-TLX
(Task Load Index): Results of empirical and theoretical research. In:
P. A. Hancock and N. Meshkati (Eds.), Human mental workload (pp.
139–183). Amsterdam: Elsevier.
Hockey, G. R. J. (1997). Compensatory control in the regulation of
human performance under stress and high workload: A cognitive–
energetical framework. Biological Psychology, 45, 73–93.
Kline, R. B. (2005). Principles and practice of structural equation
modeling, 2nd ed. New York: Guilford Press.
Lazarus, R. S. (1999). Stress and emotion: A new synthesis. New York:
Springer.
Lee, J., & Moray, N. (1992). Trust, control strategies and allocation of
function in human-machine systems. Ergonomics, 35, 1243–1270.
Lee, J., & See, K. A. (2004). Trust in automation: Designing for
appropriate reliance. Human Factors, 46, 50–80.
Matthews, G. (2008). Personality and information processing: A
cognitive-adaptive theory. In: G. J. Boyle, G. Matthews, and D. H.
Saklofske (Eds.), Handbook of personality theory and assessment:
Vol. 1: Personality theories and models (pp. 56–79). Thousand Oaks,
CA: Sage.
Matthews, G., & Campbell, S. E. (1998). Task-induced stress and
individual differences in coping. Proceedings of the Human Factors
and Ergonomics Society, 42, 821–825.
Matthews, G., Campbell, S. E., Falconer, S., Joyner, L. A., Huggins, J.,
Gilliland, K., . . . Warm, J. S. (2002). Fundamental dimensions of
subjective state in performance settings: Task engagement, distress,
and worry. Emotion, 2, 315–340.
Matthews, G., Deary, I. J., & Whiteman, M. C. (2003). Personality
traits, 2nd ed. Cambridge: Cambridge University Press.
Matthews, G., Joyner, L., Gilliland, K., Campbell, S., Falconer, S., &
Huggins, J. (1999). Validation of a comprehensive stress state ques-
tionnaire: Towards a state “big three”? In: I. Mervielde, I. J. Deary,
F. De Fruyt, and F. Ostendorf (Eds.), Personality psychology in
Europe vol 7. (pp. 335–350). Tilburg, the Netherlands: Tilburg
University Press.
Maxwell, S. E., & Delaney, H. D. (2004). Designing experiments and
analyzing data: A model comparison perspective, 2nd ed. Mahwah,
NJ: Erlbaum.
Merritt, S. M., & Ilgen, D. R. (2008). Not all trust is created equal:
Dispositional and history-based trust in human-automation interac-
tions. Human Factors, 50, 194–210.
Meyer, J., & Biton, Y. (2002). Why better operators receive worse warn-
ings. Human Factors, 44, 343–353.
Mooradian, T., Renzl, B., & Matzler, K. (2006). Who trusts? Person-
ality, trust, and knowledge sharing. Management Learning, 37, 523–
540.
Oron-Gilad, T., Szalma, J. L., Thropp, J. E., & Hancock, P. A. (2005).
Incorporating individual differences into the adaptive automation
paradigm. In: P. Carayon, M. Robertson, E. Kleiner, and P. L. T.
Hoonakker (Eds.), Human factors in organizational design and man-
agement VIII (pp. 581–586). IEA Press: Santa Monica, CA.
Parasuraman, R., & Hancock, P. A. (2001). Adaptive control of mental
workload. In: P. A. Hancock and P. A. Desmond (Eds.), Stress,
workload, and fatigue (pp. 305–320). Mahwah, NJ: Erlbaum.
Parasuraman, R., & Manzey, D. H. (2010). Complacency and bias in
human use of automation: An attentional integration. Human Fac-
tors, 52, 381–410.
Parasuraman, R., Mouloua, M., & Hilburn, B. (1999). Adaptive aiding
and adaptive task allocation enhance human-machine interaction. In:
M. Scerbo and M. Mouloua (Eds.), Automation technology and
human performance: Current research and trends (pp. 119–123).
Mahwah, NJ: Erlbaum.
Parasuraman, R., Sheridan, T. B., & Wickens, C. D. (2000). A model of
types and levels of human interaction with automation. IEEE Trans-
actions on Systems, Man, and Cybernetics—Part A: Systems and
Humans, 30, 286–297.
Penley, J. A., & Tomaka, J. (2002). Associations among the big five,
emotional responses, and coping with acute stress. Personality and
Individual Differences, 32, 1215–1228.
Prinzel, L. J., Freeman, F. G., & Prinzel, H. D. (2005). Individual
differences in complacency and monitoring for automation failures.
Individual Differences Research, 3, 27–49.
Rice, S. (2009). Examining single- and multiple-process theories of
trust in automation. The Journal of General Psychology, 136, 303–
319.
95
PERSONALITY AND ADAPTIVE AUTOMATION
Rotter, J. B. (1980). Interpersonal trust, trustworthiness, and gullibility.
American Psychologist, 35, 1–7.
Singh, I. L., Molloy, R., & Parasuraman, R. (1993a). Automation-
induced “complacency”: Development of the complacency-potential
rating scale. The International Journal of Aviation Psychology, 3,
111–122.
Singh, I. L., Molloy, R., & Parasuraman, R. (1993b). Individual differ-
ences in monitoring failures of automation. The Journal of General
Psychology, 120, 357–373.
Szalma, J. L. (2008). Individual differences in stress reaction. In: P. A.
Hancock and J. L. Szalma (Eds.), Performance under stress (pp.
323–357). Hampshire, UK: Ashgate.
Szalma, J. L. (2009). Individual differences in human-technology in-
teraction: Incorporating variation in human characteristics into hu-
man factors and ergonomics research and design. Theoretical Issues
in Ergonomics Science, 10, 381–397.
Szalma, J. L., Warm, J. S., Matthews, G., Dember, W. N., Weiler,
E. M., Meier, A., & Eggemeier, F. T. (2004). Effects of sensory
modality and task duration on performance, workload, and stress in
sustained attention. Human Factors, 46, 219–233.
Thropp, J. E. (2006). Individual preferences using automation. Unpub-
lished doctoral dissertation, University of Central Florida, Orlando,
FL.
Warm, J. S., Parasuraman, R., & Matthews, G. (2008). Vigilance
requires hard mental work and is stressful. Human Factors, 50,
433–441.
Wickens, C. D., & Dixon, S. R. (2007). The benefits of imperfect diag-
nostic automation: A synthesis of the literature. Theoretical Issues in
Ergonomics Science, 8, 201–212.
Young, M. S., & Stanton, N. A. (2004). Taking the load off: Investi-
gations of how adaptive cruise control affects mental workload.
Ergonomics, 47, 1014–1035.
Received December 2, 2009
Revision received January 18, 2011
Accepted February 14, 2011 䡲
96 SZALMA AND TAYLOR
A preview of this full-text is provided by American Psychological Association.
Content available from Journal of Experimental Psychology Applied
This content is subject to copyright. Terms and conditions apply.
