Many critical-safety domains require operators to continuously monitor and process a large number of variables. In this digital age, it might seem surprising that many of the interfaces in complex environments (e.g., flight cockpits, air traffic control centers, power plants) continue to observe a “single-sensor-single-indicator” (SSSI; as termed by Goodstein [ 18 ]) design guideline for data visualization—whereby low-level readings from environment sensors are directly and independently communicated to the operator via dedicated instruments. The rationale for SSSI’s continued use is that separate information channels allows for transparency and for individual data elements to be flexibly combined by trained operators to generate appropriate responses for all possible operational purposes, even for situations that might not be anticipated by the interface designers [ 14 ]. Moreover, complex systems are often composed of multiple sub-systems that are highly-coupled (see Methods for an example; a fixed wing landing task). Thus, the display of individual data elements not only communicates the data per se, but also allows the operator to monitor the changing relationship between sub-groups of multiple variables.

The limitations to SSSI are intuitively apparent. A busy array of instruments requires operators to sequentially seek out data with eye-movements, which then has to be integrated. In other words, visual scanning and the cognitive load on working memory are believed to place a burden on operators, which might be obviated with information visualization designs that supported decision-making instead of “data availability” [ 46 ]. This belief has motivated displays, such as those termed “ecological interface designs”, that seek to maximize “information extraction” by exploiting our seemingly limitless capacity for pattern recognition [ 6, 36 ]. In an early example, the data of 100 sensor values of a nuclear power plant, which hitherto had to be individually monitored, were mapped into a single centralized octagon display [ 47 ]. In this example, growing distortions in the octagon’s symmetry indicated a developing abnormality. To fully appreciate (and justify the practical implementation of) ecological interface designs, especially in safety-critical domains, it is necessary to demonstrate that relying on SSSIs is indeed effortful and vulnerable to variable human factors.

Operators often utilize such displays in demanding and stressful situations. Stress is an interactive process whereby a demand is placed on an operator and the response is determined by a combination of the details and appraisal of the stressor, along with perceived coping resources. Where stressors outweigh perceived coping abilities, state anxiety is likely to be invoked, which is an acute negative emotion related to a specific event and is characterized by “consciously perceived feelings of tension and apprehension” ([ 34 ], p. 17). Relatedly, trait anxiety is a general disposition where individuals respond to stressful situations with high levels of state anxiety [ 45 ]. In the present paper, we are specifically interested in examining the effects of working memory load and state anxiety during SSSI use.

Visual scanning behavior is influenced by multiple factors, which interact to determine our ability to acquire just-in-time and task-relevant information from the environment. Attentional control theory (ACT; [ 16 ]) offers a comprehensive framework that explicitly considers the relationships between attention, the working memory system, and anxiety. Thus, it can serve to help us understand how access to visual displays with more than one region of interest might be susceptible to user states. ACT is based around previously delineated attentional sub-systems, a goal-directed system and a stimulus-driven system (see [ 10 ]). The goal-directed system controls attention based on current or future goals, past experience, and predictions. Whereas the stimulus driven system directs attention based on the saliency and expectancy of sensory events. In the context of aviation, purposeful eye movements across different instruments are associated with higher proficiency and random eye movements with worse proficiency [ 8 ].

ACT postulates that anxiety can lead to a modification in the balance between the attentional sub-systems presented previously. Specifically, it is suggested that anxiety leads to decreased prioritisation of the goal-directed system, with the stimulus driven system gaining increased control over the allocation of attention. This change in prioritisation decreases the likelihood of attention being efficiently directed toward goal-relevant information. The reprioritisation is underpinned, according to ACT, by anxiety-induced changes to the functioning of specific (see [ 25 ]) working memory functions, namely: inhibition, shifting and updating. It is predicted that anxiety can lead to reduced efficiency in inhibiting inappropriate prepotent responses, and maintaining attention on task relevant information. It is also predicted that anxiety can impair the ability to switch between tasks, and update and monitor the information in working memory. Thus, in SSSI based tasks, impairments in the ability to seek task-relevant information, switch between sub-tasks, and monitor and updated information in working memory, seem very likely to be detrimental for performance.

Changes to gaze behavior have been identified in tasks performed under anxious conditions, with results providing support for ACT’s predicted influence of anxiety on attentional control [ 3, 7, 40 ]. Anxiety has been shown to increase the frequency of fixations on goal-irrelevant stimuli [ 41 ] and reduce the duration of ordinarily long target-focused fixations [ 7, 26 ]. In the context of SSSI based tasks, Allsop and Gray [ 1 ] successfully used ego-threatening instructions and monetary incentives to induce anxiety in participants with extensive practice in instrument scanning for performing a flight landing task. Anxiety led to decreased percentage dwell time on the instruments, and increased time on the external world. Scanning entropy, which is indicative of the randomness of visual scanning, also increased. Interestingly, changes in anxiety positively correlated with changes in scanning randomness. In a partially analogous context, Vine and colleagues [ 37 ] investigated the effects of stress on gaze behavior in commercial pilots as they encountered a simulated emergency situation, during an important periodic proficiency exam. Perceiving the exam to be more threatening (defined by subjectively rating the task to be demanding, along with low coping evaluations) was associated with higher search rates and more fixations on unimportant instrument locations. Such evaluations were also marginally related to increases in scanning entropy.

Turning to the effects of cognitive load on gaze behavior in tasks requiring visual scanning. In a hazard perception task, cognitive load has been shown to lead to a longer duration to first fixate on the hazard and, interestingly, also reduce the average hazard fixation durations for individuals with lower working memory capacities [ 43 ]. Cognitive load has also been shown to lead to more spatially concentrated gaze behavior during real-world driving [ 30 ]. With relation to SSSI displays, cognitive load has been shown to increase the average dwell time on instruments [ 35 ].

Cognitive load may exacerbate the effects of anxiety on gaze behavior, as, like other interference theories of anxiety [ 33 ], ACT suggests that anxiety consumes a limited pool of working memory resources. Therefore, when working memory demands converge on working memory limits, anxiety-induced attentional changes may be more likely to occur [ 4 ]. For instance, individuals with lower working memory capacities exhibit stronger negative relationships between anxiety and simple, processpure, measures of attentional control [ 31, 21 ]. Of more central interest, studies have also directly manipulated demands on working memory, and investigated its interaction with anxiety. Increasing cognitive load has been shown to compound the effects of anxiety in simple, tests measuring specific aspects of attentional control [ 5, 28 ].

A limited number of studies have examined the combined influence of anxiety and working memory on gaze behavior [ 27, 39 ]. Findings from these studies are less homogenous. Some studies have not found an interactive effect of anxiety, with cognitive load [ 27 ] or working memory capacity [ 44 ], whereas others have [ 39 ].

In sum, the present paper aims to elucidate the influence of anxiety and cognitive load on information seeking behavior during a task adhering to SSSI visualisation design philosophy [ 12 ], namely an instrument flight task.

The cognitive anxiety, heart rate, n-back and performance results will be presented first. Then the following eye movement analyses will be presented: percentage dwell time, transition frequency and scanning entropy.

This study aimed to demonstrate the effects of anxiety and cognitive load on information seeking behavior in a control scenario adhering to the SSSI visualisation design philosophy. It was framed within attentional control theory (ACT; Eysenck et al., [ 16 ]), which provides an account for how anxiety and the working memory system interacts to influence the control of attention. Participants first undertook training to perform an instrument landing task where information had to be acquired from discrete cockpit instruments in order to complete the task accurately. Then, during testing, both anxiety and cognitive load were manipulated.

The effectiveness of the anxiety manipulation was validated by increases in self-reported cognitive anxiety and objective heart rate between the neutral and anxiety conditions. This offers support for the use of evaluative instructions and monetary incentives for this purpose, with the average increase in heart rate being comparable to results from previous studies using similar manipulations (e.g., Cooke et al., [ 9 ]; Moore et al.,[ 26 ]). Self-reported anxiety was also higher in the high versus low cognitive load conditions, suggesting that participants had more concerns over their ability to perform the task under high cognitive load.

Cognitive load was successfully manipulated using an auditory n-back task, with more incorrect responses in high, compared to low, cognitive load conditions. Importantly, response times remained the same across these two conditions, which quells concerns over a speed-accuracy tradeoff. Interestingly, anxiety led to decreased reaction time, while accuracy was maintained. A likely explanation for this finding is that anxiety was accompanied by a compensatory increase in effort and auxiliary processing resources [ 16 ]. This behavioral finding is supported by increases in self-reported effort that have been previously shown to accompany anxiety [ 9, 42 ], and offers an explanation for flight task performance being maintained in anxious conditions in the current study. This is also in line with ACT, which makes an important distinction between performance effectiveness and processing efficiency. Performance effectiveness is the overall performance outcome, whereas processing efficiency refers to the effort or resources invested in order to achieve a performance outcome, with processing efficiency more readily being impacted than performance effectiveness. In the present study, maintaining flight performance in anxious conditions was achieved at the cost of reduced processing efficiency, as evidenced by both reduced n-back response times (i.e., more resources) and the anxiety-induced changes to gaze behavior detailed below.

Gaze behavior was significantly impacted by cognitive load, with a reduction in the number of transitions between areas of interest (e.g., instruments) being found. This indirectly supports previous studies showing cognitive load to increase average dwell time on SSSIs [ 35 ] and decrease the variability of gaze location [ 31 ], as both these gaze changes will consequently likely lead to less transitions. It is probable that this decrease in transition frequency led to instruments being inadequately sampled and thus was responsible for the observed impairment of flight task performance in high cognitive load conditions.

In anxious conditions, there was a tendency for participants to look more towards the outside world as opposed to the instrument panel. Although this finding was marginally significant (p = .06), it is qualitatively similar to Allsop & Gray’s [ 1 ] findings. In accordance with ACT and previous work, we suggest that anxiety led to a decreased influence of the goal-directed attentional system, which inturn led participant’s to be more likely to orientate attention away from goal-relevant information (i.e., the instruments crucial for performance of this task).

Partial evidence was found for anxiety leading to an increase in the randomness of gaze behavior. Specifically, whilst the ANOVA main effects for anxiety, or interactive effects of anxiety and cognitive load, on scanning entropy were not found, it was evident that the variation in response to the anxiety manipulation was most likely responsible for these null effects. Planned correlational analyses revealed that change in anxiety from the neutral to anxiety conditions correlated with change in scanning entropy, but interestingly, only when cognitive load was high. This result offers some support for the suggested (e.g., [ 4 ]) interaction between anxiety and working memory demands and is somewhat analogous to findings in simpler tasks [ 25, 26 ]. The ability to effectively seek important information in tasks adhering to an SSSI design philosophy may therefore be most impaired when both anxiety and cognitive load is high.

To elucidate support for the effects of anxiety on scanning randomness, we conducted a supplementary, alternative analysis, where participants were categorized into low- and high-anxiety manipulation responders. This analysis again revealed that entropy was higher for participants who had larger increases cognitive anxiety from neutral to anxiety conditions. When taken together, these results concur with previous studies [ 1, 37 ], and offers support for the suggestion that an individual’s reaction to a potentially anxiety-inducing situation is crucial in determining the effects on the predictability of gaze behavior. The interesting next step would be to determine exactly what underpins these anxiety-induced changes in scanning randomness. It is possible that the impairment to certain working memory functions (e.g., shifting, updating, inhibition) contributes to these changes.

The current study offers a number of interesting findings, however there are a number of limitations that should be considered when interpreting the results. Firstly, the participants in the current task were trained novices, therefore whether the various results generalize to other populations (i.e., true experts) requires further examination. Secondly, whilst the employed anxiety manipulation is readily used in perceptual-motor experiments, it cannot compare to real-world anxiety-inducing situations and is comparatively relatively weak. Thirdly, the sample size was relatively small, meaning that certain effects may not have been detected. Thirdly, adherence to the scan pattern specified during training may have somewhat dampened the effects of the independent variables on gaze behavior. Fourthly, subjective cognitive anxiety was measured after completion of each trial, potentially introducing retrospective bias. Specifically, participants may report higher anxiety after performing poorly.

While the study reported here is based on a flight control scenario, it holds broad implications for information visualization design. Visualizations can be dichotomized into those that are designed for “data availability” and those for “information extraction” [ 46 ]. It is often assumed that those that are designed to maximize “data availability” (e.g., SSSIs) burden the operator with the task of seeking out relevant data, maintaining this data in memory, while integrating these data to generate appropriate responses/decisions. The current results show that this is indeed the case. Visual scanning behavior across multiple and separate channels of information is an effortful process that is further compromised by operators’ states of anxiety and high working memory load. A shift to more integrated displays, such as ‘glass cockpits’, is not necessarily the solution, as they still present information in separate, albeit spatially closer, display regions. Indeed, there is some evidence to suggest that they can lead to poorer flight performance in novices [ 48 ]. It would be interesting for future research to explicitly examine whether differing data visualisation philosophies (e.g., SSSI compared to ecological interface design) differ in their capability for operators to use them in high anxiety, high cognitive load conditions.