Below are the first 10 and last 10 pages of uncorrected machine-read text (when available) of this chapter, followed by the top 30 algorithmically extracted key phrases from the chapter as a whole.
Intended to provide our own search engines and external engines with highly rich, chapter-representative searchable text on the opening pages of each chapter. Because it is UNCORRECTED material, please consider the following text as a useful but insufficient proxy for the authoritative book pages.
Do not use for reproduction, copying, pasting, or reading; exclusively for search engines.
OCR for page 112
Flight to the Future: Human Factors in Air Traffic Control 6 Workload and Vigilance In the previous chapter, the underlying cognitive characteristics of the tasks that air traffic controllers carry out were described in some detail. Many of these cognitive operations impose demands on the controller's mental workload. It has already been noted that the projected increase in air traffic over the next decade threatens to overwhelm the capacity of the air transportation system. If safety is not to be compromised, it is vital that individual controllers are not subjected to overload due to high traffic density and complexity. Accordingly, in this chapter we examine the characteristics of the mental workload of air traffic controllers and its relationship to overall system performance. It should be noted at the outset that the entire range of controller workload, from low to high, needs to be considered in air traffic control operations. It is most natural to think of high levels, or overload, when considering workload. Considerable evidence exists to indicate that human operators who experience high levels of workload can be susceptible to errors or performance breakdown. A study by Endsley and Rodgers (1996), for example, appeared to demonstrate a positive correlation between workload and operational errors, at least for high levels of workload. Indeed, high workload was identified as one of the contributing causes to the accident at the Los Angeles airport in 1991, in which a departing commuter aircraft had been positioned on the runway in the path of a landing USAir 737 (National Transportation Safety Board, 1991). However, underload can be equally pernicious. Hopkin (1995) suggested that the extensive research on overload in air traffic control has led to a relative neglect of underload; as we discuss later, operational errors have also been reported under conditions of low
OCR for page 113
Flight to the Future: Human Factors in Air Traffic Control to moderate traffic complexity (Stager, 1991; Stager and Hameluck, 1990). Thus, it is important to understand both underload and overload, including the ways in which situation awareness mediates the relationship between workload and errors. Accordingly, in this chapter we treat the load on the controller as falling along a continuum from low to high, examining both workload and vigilance. Another point worth noting is that there is no typical controller workload profile or characteristic style of vigilance that is representative of air traffic control in general. Workload patterns and the quality of vigilant monitoring are likely to differ between en route, TRACON, and tower controllers, between control centers of different levels, between radar and nonradar control, between different sectors, and so on. Ultimately any comprehensive examination of the workload of air traffic control must be stratified by these and other job- and system-related factors. We provide here a general analysis of mental workload and vigilance for en route and TRACON controllers, recognizing that there is likely to be considerable diversity within these categories. MENTAL WORKLOAD History and Definitions The study of mental workload has occupied a prominent position in human factors research and practice over the past four decades. Workload assessment studies have been conducted since the 1950s and early 1960s (Brown and Poulton, 1961), and air traffic control was an early area of application (Kalsbeek, 1965; Leplat and Sperandio, 1967). However, much of the theoretical development of the field can be traced to a 1977 conference of the North Atlantic Treaty Organization and subsequently published book, Mental Workload (Moray, 1979). Since that seminal volume, several thousand studies have been conducted on the theoretical underpinnings, assessment techniques, and practical implications of mental workload in a variety of domains. A partial bibliography created a decade ago had over 500 listings (Hancock et al., 1988). Even reviews of this work number in the dozens, and only a few are mentioned here: Damos (1991), Hancock and Meshkati (1988), Huey and Wickens (1993), Kantowitz and Campbell (1996), Lysaght et al. (1989), Moray (1988), O'Donnell and Eggermeier (1986), Warm et al. (1996), and Wickens (1992a). The number of publications attests to the importance accorded the concept of mental workload in the human factors research community. The pace of research has slowed somewhat in recent years, having been replaced by studies of situation awareness (Flach, 1994; Gilson et al., 1994; Wickens, 1992b). Nevertheless, an understanding of the factors influencing human mental workload is likely to be crucial to the design of air traffic control and other systems (Andre and Hancock, 1995). This need will persist in the future, as systems become more automated. What is workload? First, the term generally refers to mental workload, that
OCR for page 114
Flight to the Future: Human Factors in Air Traffic Control is, the load associated with the mental (including cognitive and affective) processes of the human operator, rather than (or in addition to) physical workload. This emphasis on mental workload is appropriate because the job of air traffic control, in common with most other modern work settings, is primarily cognitive and information-intensive, rather than physical and labor-intensive. Accordingly, most of what is discussed in this chapter deals with mental workload.1 The term mental workload has an immediate intuitive meaning, yet it has resisted precise definition. Various authors have conceived of workload as the objective task demands imposed on the human operator, the mental effort exerted by the operator to meet these demands, the performance of the operator, the psychophysiological state of the operator, and the operator's subjective perception of expended effort. Many definitions assume that mental workload is an intervening construct that reflects the relationship between the environmental demands imposed on the human operator and the capabilities of the operator to meet those demands. Workload may be driven by the objective load imposed on the controller from external environmental sources (airspace factors, displays, tasks, procedures, other controllers, and supervisors), but not inevitably, because workload is also mediated by the controller's response to the load and by his or her skill level, task management strategies, and other individual characteristics. Although there is no agreed-on definition of workload, theories of workload based on the concept of attentional capacity or resources have been proposed (Kahneman, 1973; Kantowitz and Casper, 1988; Wickens, 1984). Each of these theories assumes that tasks (except those that can be performed automatically) require the allocation of the operator's attentional resources for efficient execution and that operator workload reflects the overall level of demand for resources. The theories differ in assuming either a single pool of resources that can be flexibly allocated to different activities (Moray, 1967) or multiple resources that differ qualitatively according to such features as input and output modalities, stages of information processing, and response requirements (Wickens, 1984). Workload and Air Traffic Control Performance What are the relationships between traffic factors, workload, and performance? At a global level, the controller's workload is related to the capacity to manage traffic. The more aircraft that have to be handled, the greater is the 1 It is worthwhile to remember that consideration of physical workload may still be necessary on occasion. Even in the most information-intensive job, the human operator must interact physically with devices to exchange information. The placement and control features of these input and output devices, if poorly designed, may not only lead to injury (e.g., carpal tunnel syndrome) but also induce discomfort and fatigue. Furthermore, to the extent that the physical demands imposed on controllers (e.g., keyboard entry, movement of flight strips, reaching, and other manual behaviors) interact with cognitive activities and therefore contribute indirectly to mental workload, consideration of the physical workload is important.
OCR for page 115
Flight to the Future: Human Factors in Air Traffic Control workload until, at some point, ''things give." The job of the air traffic controller is often seen as the quintessential example of a high-workload, stress-inducing occupation. However, although the workload can be high, there is little support for the view that the job is uniquely stressful (Costa, 1993; Hopkin, 1992). Studies show that controllers experience peaks and troughs of workload during operations. High workload can be a problem, not solely because it may impact negatively on controller performance (and hence potentially on safety), but also because it can set an upper limit to traffic-handling capacity. Decreasing sector size or increasing the number of controllers does not necessarily solve this problem because of the consequent increase in intersector and intercontroller coordination and communication. Moreover, decreasing sector size reduces the amount of time spent with each aircraft, so that the controller has less time to build up the picture of the traffic; as discussed further below, this may increase workload. Low traffic load may result in boredom and reduced alertness, with consequent implications for handling emergencies. Although factors conducive to high and low workload are prevalent in air traffic control, this does not necessarily mean that all controllers experience extremes of workload. All successful controllers use various adaptive strategies to manage their performance and subjective perceptions of task involvement. Sperandio (1971) first showed that controllers handled an unexpected increase in traffic load adaptively by decreasing the amount of time they spent processing each aircraft. A controller may also stop doing less important, peripheral tasks, thus leaving more time for active control; or increase spacing, stack aircraft, or prevent them from entering the sector (hence reducing airspace capacity). Because air traffic control is a team activity, another possibility is that controllers may ask a colleague to take over a particular task. In general, controllers may use a variety of strategies to manage workload and regulate their performance: if they do not use any of these adaptive strategies, further increases in traffic load may result in errors. These considerations suggest that one needs to distinguish between workload drivers (i.e., factors in the environment external to the controller), controller workload, controller strategies, and performance consequences. Figure 6.1 presents a schematic view of the interrelationships of these factors. The influence of environmental drivers can be modeled, but it must be supplemented by assessment of the controller to measure the actual workload experienced. The controller uses various strategies to cope with the external drivers. Controller performance represents the joint consequences of the effects of task drivers on workload and the mediating influence of controller strategies. The remainder of this chapter examines each of these workload factors in turn. In the following sections, we: (1) model the effects of workload drivers, (2) discuss the effects of workload drivers on controller workload, (3) examine the relationship of workload to performance, (4) evaluate the role of controller vigilance, and (5) consider the influence of low traffic load and sleep loss or sleep
OCR for page 116
Flight to the Future: Human Factors in Air Traffic Control FIGURE 6.1 Interrelationships between workload drivers, workload, and performance. disruption on performance. Note that, from a systems view, workload assessment and modeling are not necessarily of interest in themselves. The goal is to predict system performance under varying conditions of workload. To achieve this goal, one needs first to measure workload, predict the effects of different drivers on workload, and then attempt to predict controller performance in response to very high or very low workload. A more detailed discussion of workload assessment appears in Chapter 10, on research methods. MODELING WORKLOAD There is often a need for predictive workload assessment, not only for new systems but also for systems in which new hardware or software capabilities are to be introduced. Another factor in the thrust to develop predictive models is that federal agencies such as the FAA and the Air Force require workload certification of new aircraft before they can be acquired (Federal Aviation Administration, 1994). A number of such predictive workload models have been proposed, and a few are briefly discussed here.2 2 Only models that assume that controllers are capable of concurrent time sharing of tasks are described. Several models that assume only serial processeing of tasks and view multiple-task work-load as a problem of scheduling (e.g., the SAINT model of Wortman et al., 1978) are not discussed here.
OCR for page 117
Flight to the Future: Human Factors in Air Traffic Control Some predictive models are based on the results of empirical studies, whereas others have been developed from first principles. One of the most commonly used models is time-line analysis, in which workload is modeled as a function of the proportion of time spent in performing a task relative to the total time available (Parks and Boucek, 1989). Levels of workload during a specified time interval can then be determined by summing the time lines of each task performed during that period and dividing by the time interval. Conventional time-line analysis makes the rather overarching assumption that workload during a given interval is 100 percent if the controller is fully occupied with a particular task during that interval, so that the introduction of other tasks during that same interval would lead to workload greater than 100 percent, or to overload. In contrast, as discussed previously, resource theories of multitask performance predict, and studies have shown, that overload will result only if each task competes for the same resources (or the same input and output channels) and if the total resource demand exceeds the operator's capacity. For example, a controller can easily talk to a pilot while scanning the radar display, but cannot easily talk while listening to another controller. To accommodate these findings, the W/Index predictive model includes metrics of task and resource conflict in estimating workload (North and Riley, 1989; Sarno and Wickens, 1995), and variants of this model have been employed to predict workload in a number of aviation settings, including at least one in an air traffic control setting (Burbank, 1994). The effective time available for completing a task is also a feature of workload models proposed by Hancock and Chignell (1988) and by Laudeman and Palmer (1995). In the former, workload estimates are a function of time of task completion but are also modulated by predicted estimates of the skill level of and degree of mental effort expended by the human operator. Laudeman and Palmer (1995) extended conventional time-line analysis by assuming a linear function of increasing workload during the time available for completing a task. In their model, the function begins at zero before the task is attempted and returns to zero after task completion. By summing together the workload functions for each task, a predicted workload profile over time is obtained. The area under this overall workload function was proposed as an index of workload. Finally, Rouse et al. (1993) described a state-space, predictive model of subjective workload. In their model, subjective workload is a lagged function of the operator's actions, performance, the system state, and the operator's previous subjective experience. Subjective workload does not simply parallel operator performance, and workload is predicted to be dependent on adaptive changes that the operator initiates to moderate the impact of increases in imposed task load. Human operators often use such changing strategies and variations in operating procedures to maintain performance at some preferred level that is acceptable but not necessarily optimal or perfect (Hart and Wickens, 1990). However, whereas the model proposed by Rouse et al. (1993) implies that operators use adaptive strategies to keep subjective
OCR for page 118
Flight to the Future: Human Factors in Air Traffic Control workload within acceptable bounds, including accepting lower levels of performance, the empirical evidence suggests that this is not the only adaptive strategy. Operators, including controllers (e.g., Sperandio, 1971) may change behaviors (e.g., operating procedures) not to minimize subjective workload, but to keep performance within acceptable limits. Workload models have the unique advantage that they provide the only means of assessing workload ahead of time. The predictions can then be put to empirical test. Hence workload modeling is useful in design and in prototyping. A potential disadvantage is that this method is only as good as the underlying model. Furthermore, not all models have been experimentally validated. Nevertheless, on balance, workload modeling holds considerable promise for evaluating evolving systems and is likely to play an important role in the assessment of future automation concepts in air traffic control. Workload Drivers There are many workload drivers in the air traffic control environment. As stated earlier, however, it is important to note that task load should not be assumed to elicit a passive, fully predictable response from the human operator. Different controllers may respond differently to the same load factor (e.g., an increase in traffic density). The same controller may respond variably on two different occasions by using preplanning, task shedding, or other coping strategies to minimize mental workload on one occasion but not on the other. Skill and training also influence the response to workload drivers. We present here some of the more important sources of workload in air traffic control. Airspace Load A starting point is to examine those aspects of the air traffic control environment that contribute to task loading. In essence, this approach attempts to analyze the intrinsic load of air traffic control operations, as a prelude to assessing (or predicting) the load on the controller. At the simplest level of such an analysis, for example, the number of aircraft being handled by the controller could be defined as an important load factor. This variable is clearly insufficient on its own, however, because the demand imposed by a certain number of aircraft on the controller would also depend on other factors, such as traffic complexity, aircraft mix, weather, etc. Hence, one way to proceed would be to enumerate all potential variables and categorize them, deduce interrelationships between variables, and compute derived load factors from the raw airspace variables. Using essentially this approach, Arad (1964) conducted a series of assessments of the objective job difficulty of different air traffic control operations. The goal of the studies was to use the results as a basis for sector design as well as other issues, such as planning of staff levels. Arad divided the drivers of load
OCR for page 119
Flight to the Future: Human Factors in Air Traffic Control Routine Control Load Airspace Load Number of controlled aircraft Number of controlled aircraft Sector flight time Sector flow organization coefficient Proportion of standard aircraft Mean airspeed Proportion of non-standard aircraft Sector area Proportion of terminal area hand-offs Mean aircraft separation Proportion transitioning Proportion of VFR to IFR "pop-ups" FIGURE 6.2 Some airspace load variables in air traffic control. Source: Adapted from Arad (1964). into three general categories: background load, routine load, and airspace load. These load factors were defined by equations including such variables as the number of aircraft under control, sector flight time, sector area, mean aircraft separation, etc. Figure 6.2 lists the routine load and airspace load variables in Arad's (1964) scheme. As mentioned at the outset of this chapter, controller workload is likely to vary among the different air traffic control positions. Airspace load factors also differ among the TRACON, en route, and oceanic control environments. Such differences must be taken into account in any comprehensive evaluation of controller workload. Hurst and Rose (1978) followed up Arad's (1964) research with systematic observations of controllers on 47 radar sectors in the Boston and New York areas. Using modified versions of Arad's (1964) load factors, they examined the relationship between load and observer ratings of the activity level of controllers. These behavioral ratings of busyness were not related to the derived control load factors but were significantly correlated with peak traffic counts. Bruce et al. (1993) carried out a similar study on 65 sectors in 7 en route centers and found a significant relationship between traffic complexity (as assessed by traffic load factors) and the level of controller activity (e.g., verbalizations and manual activities). Unfortunately, the number of overt behaviors engaged in by controllers may or may not be accompanied by increased mental workload. As noted previously, behavioral ratings may be insensitive to the covert demands on the information processing of load factors. Hopkin (1971) pointed out that even more sophisticated behavioral measures (such as the use of keyboards and of communications equipment) may be largely insensitive to the load associated with problem solving and decision making by the controller. Hurst and Rose (1978) were aware of this limitation and suggested that physiological measures might provide additional validation of the impact of task load on controller mental workload. A more serious weakness of their approach, however, is the notion implicit in their study design and data analysis procedures that controller workload is a direct, open-loop function of task load, as in the stress-strain relationship of mechanical
OCR for page 120
Flight to the Future: Human Factors in Air Traffic Control structures. Definition and quantification of airspace load factors are insufficient by themselves because of the multiple, closed-loop nature of the air traffic control operations. Air traffic control represents a dynamic system in which the controller's behaviors affect some of the same control load variables that are thought to impact on mental workload (e.g., airspeed).3 Furthermore, as noted earlier, Sperandio (1971) has shown that controller workload does not necessarily increase proportionately with increases in airspace load, because controllers use strategies and vary operating procedures to achieve acceptable levels of performance. The role of airspace factors in driving controller workload is being addressed by ongoing efforts to establish objective measures of sector complexity (Rodgers et al., 1995; Pawlak et al., 1996) and could also be informed by the recent development of the SATORI (situation assessment through re-creation of incidents) tool (Rodgers and Duke, 1994). SATORI allows for the graphical recreation of all radar, weather, and communications data recorded at an en route center. SATORI contains within it several airspace variables in addition to those analyzed by Arad (1964), Hurst and Rose (1978), and Bruce et al. (1993). This software tool can be used to extract such variables as the number of way points, the volume of airspace, the number of navigation aids, military operations areas, as well as other sector and weather-related information that may be relevant as workload drivers. Display Factors The design of visual displays is a traditional area of concern in aviation human factors (Stokes and Wickens, 1988), and air traffic control is no exception. Early radar displays such as the plan position indicator (PPI) suffered from poor signal-to-noise ratios, glare, low contrast, and other factors that impeded quick and accurate detection of targets (Baker, 1962). With the development of signal preprocessing, alphanumeric displays, high-resolution graphics, and color-coded displays, the sensory detection problem was largely eliminated. But perceptual and cognitive processing remain important display design issues. The new display capabilities allow much information to be displayed. For example, the plan view display (PVD) can provide such information as the aircraft call sign, aircraft type, TCAS-equipped aircraft, reported altitude, assigned altitude, speed, time, target track, track history, and other information. Of course, the controller can 3 Jorna (1991) makes the related point that the common view taht air traffic control is a paced task over which the controller has little or no control is not entirely correct. Controllers may choose on occasion to divert aircraft into holding patterns if they have indications of high pacing of incoming aircraft, so as to not compromise safety (although delays occur) while managing their workload.
OCR for page 121
Flight to the Future: Human Factors in Air Traffic Control choose not to have all the information displayed, but nevertheless display clutter and potential increased workload are possible. Other factors that influence display-related workload include type size, luminance, contrast, color, and visual coding of alphanumeric symbology. Standard human engineering guidelines and databases (Sanders and McCormick, 1992; Van Cott and Kinkade, 1972) can be consulted for appropriate design choices for each of these factors. New displays for aiding the controller in prediction and extrapolation of flight paths have become available in recent years. As we noted in Chapter 5, the need to project aircraft paths into the future imposes a high demand on the controller and thus is likely to be a source of workload. Algorithms that allow for the accurate prediction of flight paths, with appropriate display of these predictions, will therefore considerably reduce the controller's workload in this phase. Efforts to examine three-dimensional display technology to facilitate controller visualization of the vertical dimension are also being initiated (May et al., 1995). Controller-Pilot and Controller-Controller Communications The primary means of communication between controller and pilots is verbal, through the use of radio telephony (RT). Commands and clearances from the controller allow the pilot to navigate through crowded terminal areas with the required amounts of separation. Controller-pilot communications are also vital for exchanging information about weather, traffic flying under visual flight rules, runway hazards, etc. Communications between controllers are also required for efficient handoffs between sectors, planning, scheduling, and other activities. Consequently, attention has focused on the nature of verbal communications and its role in the overall workload of the controller (Cardosi, 1993; Kanki and Prinzo, 1995; Prinzo and Britton, 1993). The analysis of the controller's verbal behavior as an embedded task index of workload has already been mentioned (Leplat and Browaeys, 1965). High levels of communications may not only increase controller workload but may also impact negatively on the controller's ability to get the big picture. Jorna (1991) stated that, when controllers spend more than half their time communicating with pilots, they report that their traffic awareness becomes disturbed. When this occurs, the effect of any task factor or workload driver (such as a visual flight rules to instrument flight rules pop-up) that normally has only a small impact on mental workload and performance may loom larger. Finally, different aspects of controller-pilot communications (e.g., message length and composition) also have an impact on pilot workload (Morrow, in press; Morrow et al., 1993) and, to the extent that this leads to communication delays or misunderstandings between pilot and controller, the workload of the controller can also be indirectly affected. Morrow et al. (1993) have outlined some principles for improving collaborative communication between controllers and pilots.
OCR for page 122
Flight to the Future: Human Factors in Air Traffic Control Workload and Current Automation Although automation in air traffic control has been limited in scope to date (as is discussed more thoroughly in Chapter 12), implications of workload studies for current automation can be drawn. Automation has traditionally been introduced in many systems partially in an attempt to reduce or regulate the operator's required level of mental workload at times of high task load. This is the standard engineering solution to operator error (or substandard performance). In some instances, this may be the correct solution. The simplest view of the effect of introducing automation is that cognitive resources are freed for performance of other manual tasks. Workload research and theories of attention provide some basis for predicting the resulting impact on workload and performance. Clearly, automation of a previously manual task will have an effect only to the extent that the task is resource sensitive (Norman and Bobrow, 1975). Moreover, the multiple-resource theory of Wickens (1992a) predicts that the required resources must overlap with those required to perform the manual tasks. Alternatively, automation of a task will benefit performance and workload if it frees input or response channels that would otherwise be tied up (Navon, 1984). Numerous dual-task and multitask studies have shown that removing a task from the operator's control can benefit performance and workload if these requirements are met (Damos, 1991; Wickens, 1992a). Tsang and Johnson (1989) found that lateral-hold automation in a flight control task reduced subjective workload, both when the flight task was performed alone and when it was combined with other cockpit tasks. In these and other multitask studies, the automated task was removed from manual control from the outset and remained so throughout the study. Given that more flexible automation is current in many systems (e.g., the cockpit flight management system, which has several modes; see Chapter 12), whether the workload benefits of automation also accrue when automation is invoked in a dynamic and flexible manner needs to be examined. In such cases, as opposed to when a task is permanently allocated to automation, the operator is likely to monitor the automation from time to time to ensure its proper functioning or to use its outputs, as in the case of decision-aiding automation. Thus automation of a task is not the same as removal of the task, so that the assumption that automation frees up cognitive resources and reduces workload may not hold (Wiener, 1988). Parasuraman (1993) found that periodic, dynamic automation of flight-related tasks enhanced performance on other flight tasks performed manually and reduced subjective workload. These effects were not simply the result of task subtraction (e.g., doing two tasks instead of three) as in multiple-task studies (e.g., Tsang and Johnson, 1989), because subjects were required to supervise the automated task and were able to do this satisfactorily (as assessed by post-session tests). These studies suggest that, in principle, automation of tasks can be beneficial in air traffic control to the extent that the automated tasks are resource demanding
OCR for page 124
Flight to the Future: Human Factors in Air Traffic Control straightforward. Theory predicts that controllers may expand attentional resources in response to an increase in task load (Kahneman, 1973). The controller may experience increased mental workload, but performance is maintained. Although attentional resource theories predict that performance will decline if the upper limit of the controller's capacity is exceeded, demonstrating this in operational air traffic control settings has proved somewhat elusive. Increases in task load may also increase controller mental workload but not change performance because of the use of compensatory or regulatory methods discussed previously (e.g., Sperandio, 1971). Nevertheless this would suggest that the controller has less spare capacity for dealing with unusual circumstances or emergencies, leading one to suspect that operational errors might increase. However, there is little direct evidence for such a scenario. In an analysis of operational errors in Canada, Stager and colleagues (1989) found that operating irregularities and incidents were not uniquely associated with high workload, but rather with low to moderate workload and moderate pace levels (intermediate values on the airspace load variables described earlier). Another study of air traffic control errors by the Canadian Aviation Safety Board (1990) found that, of 217 incidents selected from 437 occurrences, 60 percent of the system errors were attributed to planning, judgment, or attention lapses on the part of controllers. However, most of the operational errors occurred during conditions of low traffic complexity. Similar results were obtained for U.S. air traffic controllers (Rodgers, 1993). Stager (1991) considers a number of human error taxonomies that might be examined to better understand this apparent paradox. One way to resolve it is to distinguish between task load and controller mental workload. Low traffic load does not necessarily lead to low controller mental workload. As discussed in more detail in the next section, recent findings indicate that maintaining vigilance under low task load requires considerable mental workload. Hopkin (1988) noted that the emphasis in air traffic control research on high task load and stress has led to a comparative neglect of low task load and boredom. If maintaining vigilance is boring but demanding, and if, as has been argued, boredom is itself a stressor (Thackray, 1981), then the neglect of these factors becomes doubly serious. Hence, both very low and very high levels of task load (e.g., number of aircraft) can lead to substandard performance. This explanation does not account for the failure of existing studies to show a relationship between operational errors and high task load, unless one assumes that adaptive strategies are sufficient to limit the risk of error at high workload. However, it may be dangerous to assume that such a relationship does not exist merely because it has not been demonstrated empirically to date. A conservative conclusion would be that operational performance in air traffic control can be compromised by both very high and very low task load. Having discussed the upper levels of workload, we now turn to a discussion of the lower end of the continuum, or vigilance.
OCR for page 125
Flight to the Future: Human Factors in Air Traffic Control VIGILANCE History and Definitions Many activities require sustained attention for successful completion. When the activity needs to be continued for a long period of time without interruption, the ability to maintain vigilance for the occurrence of unpredictable but critical events may be compromised. Mackworth (1957) provided an early definition of vigilance: ''a state of readiness to detect and respond to certain small changes occurring at random time intervals in the environment" (pp. 389–390). Although this definition is still used by most researchers today, the emphasis on detection may not be relevant to some modern systems, including the current air traffic control environment; as discussed earlier, the controller sensory detection problem has largely been eliminated because of improvements in sensor and display technology. Instead, vigilance for the discrimination or diagnosis of unusual conditions is required. Sometime such conditions may be missed altogether ("I didn't see it on the scope"), but more often they are not understood and responded to speedily. An expanded view of vigilance extends the concept beyond detection to discrimination, recognition, or diagnosis, and the measure of vigilant performance to include both accuracy as well as speed of response. The decline in detection performance overtime in vigilance tasks, or the vigilance decrement, has been confirmed in a large number of investigations (Davies and Parasuraman, 1982). The vigilance decrement refers equally to the decline in detection rate and the increase in response time over the duration of the watch. Several studies have shown that most of the decrement occurs within 30 minutes (Teichner, 1974), although for very perceptually demanding visual targets it can appear within the first five minutes (Nuechterlein et al., 1983). The cardinal features associated with vigilance decrement are the temporal uncertainty and low probability of occurrence of targets. Since Mackworth's pioneering experiments, many studies of vigilance have been carried out. For reviews of this large corpus of work, see Craig (1985), Davies and Parasuraman (1982), Huey and Wickens (1993), Parasuraman (1986), and Warm (1984). Task Factors Influencing Vigilance Numerous task factors affect detection performance in vigilance tasks. These include psychophysical parameters, such as target intensity and duration, as well as temporal and spatial characteristics of the vigilance task, such as target frequency, regularity, number of stimulus sources, background event rate, and so on. For reviews of the effects of these factors on the vigilance decrement and on the overall level of vigilance performance, see Davies and Parasuraman (1982) and Warm and Jerison (1984). In general, vigilance is high for targets that are
OCR for page 126
Flight to the Future: Human Factors in Air Traffic Control highly salient, temporally and spatially predictable, and occur frequently in the context of a low background event rate. Unfortunately, many real-world targets possess the opposite of some of these attributes: they occur very infrequently, and, although modern signal processing techniques can ensure that target intensity and duration are above threshold (although not always, as in the case of passive sonar targets; see Mackie et al., 1994), the temporal and spatial unpredictability of targets poses a considerable challenge to the controller. The vigilance tasks that have been studied in laboratory experiments are quite varied in their characteristics. Despite this diversity, it is possible to describe vigilance tasks along some common dimensions. On the basis of a review of the literature and their own experiments, Parasuraman (1979; Parasuraman and Davies, 1977) suggested that many vigilance tasks can be classified according to a four-fold taxonomy: target discrimination type (successive or simultaneous), background event rate (low or high), sensory modality (visual or auditory), and target source complexity (single or multiple source of targets). Signal detection theory (Green and Swets, 1966) suggests two possible sources of the vigilance decrement: a decrement in perceptual sensitivity (d') and an increment in response bias (b) overtime. The vigilance taxonomy was first applied to define the conditions under which each of these outcomes is likely. The increment in b over time indicates that operators become increasingly conservative over time in calling an event a target. This finding is ubiquitous in vigilance studies, suggesting that appropriate training to regulate the subject's response criterion can reduce the vigilance decrement. Training studies have found the decrement in detection rate can be reduced, although not eliminated completely, and that the response criterion can be moved in the direction of optimality (Craig, 1985; Davies and Parasuraman, 1982). More generally, training the human operator's response criterion can enhance performance in many detection tasks, as shown by Parasuraman (1985) in studies of chest x-ray inspection by radiologists and by Bisseret (1981), who examined the detection of aircraft course conflicts by controllers. In this latter study, expert controllers had lower values of ß than trainees—that is, they were more willing to call for a correction of a detected conflict. Bisseret (1981) suggested that trainees were less willing to respond because of their greater uncertainty, and therefore that training should emphasize appropriate adjustment of the response criterion. The Workload of Vigilance Vigilance tasks are boring and have traditionally been thought of as undemanding. This view follows from the traditional arousal theory of vigilance, which views the vigilance environment as an unstimulating one. In European scientific circles the words vigilance (French) or vigilanz (German) are often used synonymously with arousal or alertness, and reduced vigilance and lowered arousal are thought to be closely related. Considerable research exists to indicate,
OCR for page 127
Flight to the Future: Human Factors in Air Traffic Control however, that although vigilance is influenced by arousal (as are many perceptual and cognitive functions), the vigilance decrement is not inevitably a consequence of reduced arousal (Parasuraman, 1984). Moreover, more recent research shows that, although maintaining vigilance can be boring, it imposes considerable mental workload on the operator. This finding is consistent with newer multidimensional conceptions of arousal that make reference to attentional resource theory (Matthews et al., 1990) or psychophysiological adaptation (Hancock and Warm, 1989). Recent studies indicate that even superficially simple vigilance tasks can impose considerable mental workload, of the level associated with such tasks as problem solving and decision making (Warm et al., 1996). As noted earlier, the work of Warm and colleagues has established that subjective mental workload in vigilance is high and is sensitive to numerous task and environmental factors that influence task performance. The workload of vigilance does not simply arise from the operator's efforts to combat the tedium of having to perform a dull task (Sawin and Scerbo, 1994; Thackray, 1981). Using a simulated air traffic control display, Warm et al. (1996) showed that advanced notification of a conflict reduced rather than increased subjective workload, even though such decision aiding should increase boredom because it leaves the operator with little to do. These results support the view that the workload of vigilance is directly task-related, rather than a by-product of boredom. Vigilance and Air Traffic Control Maintaining vigilance for critical events such as loss of separation, altitude deviations, VFR pop-ups, incorrect pilot readbacks, and other infrequent events is an important component of the controller's task. However, despite the importance of controller vigilance to the safety of air traffic control operations, there are comparatively few studies of vigilance during simulated air traffic control. Studies in the operational setting are very rare. Thackray and colleagues (Thackray et al., 1979; Thackray and Touchstone, 1989a, 1989b) have conducted a series of studies using task conditions that more closely simulate current radar displays in air traffic control. The results of a representative study are described here (Thackray and Touchstone, 1989b). In their task, which was presented on a console that closely resembled an actual air traffic control radar workstation, subjects (university students, not controllers) were presented with two diagonal, nonintersecting flight paths on a graphics display. Aircraft were identified by a data block giving the call sign, altitude, and ground speed. The aircraft could move in either direction along the paths, and the data blocks were updated every 6 seconds. The number of aircraft under control was 16. Subjects were required to detect one of three types of critical event: (1) a change in the altitude part of the data block to "XXX," simulating a transponder malfunction; and two aircraft at the same altitude and either moving (2) toward
OCR for page 128
Flight to the Future: Human Factors in Air Traffic Control each other (conflict) or (3) away from each other (nonconflict) on the same flight path. Nine critical events, three of each type, were presented in random order during each 30-minute segment of a 2-hour vigil. Subjects detected all the transponder malfunction targets and showed little change in speed of detection overtime on task (mean detection time averaged about 9 seconds). For the same-altitude targets, however (conflict or nonconflict), about 4 percent of the targets were missed during the first hour and 13 percent during the second hour. Moreover, the latency of detected targets increased from about 19 seconds to about 28 seconds over the course of the watch. The results of the study by Thackray and Touchstone (1989b) are consistent with the findings of classical vigilance studies using simpler, artificial stimuli and targets: a vigilance decrement over time was observed, both in the number of targets detected and in the speed of detection. However, subjects had to monitor a relatively large number of targets (16), and no decrement was found for the simpler targets (transponder malfunctions). In earlier studies with simpler critical events (e.g., altitude deviations) and lower numbers of aircraft, Thackray and colleagues found detection speed showed very little increase with time on task. Drawing on the vigilance taxonomy proposed by Parasuraman and Davies (1977), Byrne (1993) pointed out that the transponder malfunction target used by Thackray and Touchstone (1989b) was of the simultaneous type, whereas the altitude targets were of the high-event rate/successive type. He suggested that the greater demand for controlled processing imposed by the altitude targets was the reason why only these targets were associated with vigilance decrement. These findings suggest that the greater the information processing demands imposed by airspace load factors and target type, the greater the likelihood of a vigilance failure occurring during extended watches. The studies by Thackray and colleagues could be criticized for their use of students as subjects. Furthermore, subjects were required only to monitor targets, without any of the other activities, such as communications and keyboard entry, that controllers engage in routinely. It is difficult to predict exactly what influence these factors would have on the pattern of results. One could argue, for example, that vigilance failures might be exacerbated with the additional demands imposed by these other activities. This would follow from theoretical and empirical vigilance studies indicating that the vigilance decrement is a function of the information processing demands of target detection, so that depletion of resources by other tasks would increase the decrement (Parasuraman et al., 1987). However, exactly the opposite might also be predicted, on the grounds that controllers are more vigilant when they have more to do than when they are bored (Sawin and Scerbo, 1994). Studies using more closely simulated air traffic control, or field studies may need to be conducted to resolve this issue. Whatever the outcome of such studies, as discussed previously, it should not be concluded that vigilance problems cannot occur in real operations. The results of an important recent study by Pigeau and colleagues (1995) of North American
OCR for page 129
Flight to the Future: Human Factors in Air Traffic Control Aerospace Defense (NORAD) operators warn that such a conclusion would be premature. The subjects were 16 experienced surveillance operators who used normal operating procedures while they worked at actual NORAD consoles (which present fused, correlated data from several radar sites). Subjects had to identify either beacon tracks of aircraft with transponders or search tracks of aircraft without transponders (e.g., light general aviation aircraft) that were detected with search radar. Both simulated and live traffic were used, but detection performance was assessed only for the simulated tracks, which Pigeau and colleagues stated were indistinguishable from actual tracks. A number of task conditions were manipulated, including sector size, watch length, and shift time. A vigilance decrement in detection speed overtime was obtained for the search tracks (which imposed a memory load associated with a successive discrimination type of target) but not for the beacon targets, which required simultaneous discrimination. However, the decrement was restricted to a particular sector size and occurred only during the night shift, so that the vigilance decrement was not as ubiquitous as found in laboratory studies. Vigilance and Current Automation Sheridan (1970) pointed out many years ago that automated systems change the role of the operator from a controller to a supervisor. Although various forms of automation have been implemented in current air traffic control systems, the controller still maintains fairly direct control over aircraft. Hence, controllers are very much in the loop in current air traffic control systems. To the extent that current automation has the aim of allocating certain routine data gathering and manipulation tasks to computers, leaving intact the controller's decision making and planning duties, automation may not harm controller vigilance. However, if automation does encroach on these higher-order task functions, there is the attendant danger that vigilant monitoring may be negatively impacted. Parasuraman and colleagues have carried out several studies indicating that the monitoring of failures in the automated control of a task is poorer than manual monitoring when operators are engaged simultaneously in other tasks (Molloy and Parasuraman, in press; Parasuraman et al., 1993, 1994, 1996). Another factor relevant to automation concerns the workload of vigilance. Lowering the information-processing demands of the task environment can promote better vigilance. However, the danger in this approach is that it can be counterproductive if carried too far. The notion that an operator will have less to do, thereby allowing more time for vigilant monitoring, has often provided a rationale for implementing automation in systems. Vigilance itself has been seen as a low-workload task. As noted earlier, many studies have exploded these myths, but they still persist today in some quarters. In fact, in some cases, the human operator may be faced with greater monitoring workload levels with an automated system than existed prior to the automation, despite the fact that the
OCR for page 130
Flight to the Future: Human Factors in Air Traffic Control automation was intended to reduce workload. The paradox (Bainbridge, 1983) is that implementing automation in an attempt to reduce workload may actually result in increased workload, because of the cognitive workload associated with monitoring the automation. WORK-REST SCHEDULES, SHIFT WORK, AND SLEEP DISRUPTION Consideration of performance issues at the lower end of the workload scale would not be complete without considering the related implications of work-rest schedules, shift work, sleep loss, sleep disruption, and fatigue on controller performance. Current work-rest schedules for controllers in the United States call for an 8-hour shift (10-hour maximum with overtime), distributed into 7 hours on duty and 1 hour of breaks, 2-hour maximum time at position, and a minimum of 8 hours between shifts. Vigilance research suggests that performance decline can occur after about 30 minutes spent continuously at a task. However, as discussed previously, to date there is no evidence for any significant decrement in controller vigilance performance within the normal time-at-position limit of 2 hours. Given that breaks totaling a maximum of 1 hour are taken at periodic intervals throughout an 8-hour shift, performance is also unlikely to decline during the course of the shift (Swanson et al., 1989), although subjective feelings of fatigue may increase progressively with time (Rosa, in press). There is little evidence for any significant loss in the performance of tasks by controllers over the course of a normal 8-hour shift (Stager and Hameluck, 1988), although some studies find decreases in alertness and psychomotor ability on selected tests within performance assessment batteries. Rhodes et al. (1994), for example, reported a reduction in accuracy in the Wilkinson serial-reaction time test from the start to the end of a shift, especially during the midnight shift. However, Costa (1993) found no change in controller reaction time or in critical flicker fusion frequency after a 7-hour shift. Although Rhodes et al. (1994) stated that their test battery was "representative of some of the fundamental elements of the air-traffic controller's job," this needs to be verified, and in the absence of such validation the interpretation of such test changes remains unclear. Although there is no systematic trend toward poorer controller performance toward the end of a shift, this is not say that there is no variation in performance. Stager and Hameluck (1988) analyzed operational errors at several Canadian centers as a function of both time on position and time on shift. Of 265 operating irregularities investigated, approximately 40 percent occurred within the first 30 minutes on position, 70 percent within the first hour, and 85 percent within the first 90 minutes. It is possible that the greater incidence of errors when they first assume a position could be because they are in the process of forming the picture of the traffic, although Stager and Hameluck (1988) found no direct support for
OCR for page 131
Flight to the Future: Human Factors in Air Traffic Control this view. With respect to time on shift, there was no significant trend toward greater incidents toward the end of an 8-hour shift. Instead, many of the incidents occurred during the first 1 or 2 hours of the shift. In a cross-tabulation of time on shift and time on position for 101 controllers in 63 two-controller incidents, over 33 percent occurred within the first hour of work and the first hour on position. Of course, one source of difficulty in analyzing performance variations overtime is the variation in traffic over the course of the shift. These results indicate that current work-rest scheduling practices per se are not associated with increased operational errors. However, there is more to work-rest scheduling than time on position and time on shift. Although controllers must rest for a minimum of 8 hours between shifts, the type of rotation between shifts and its impact on circadian rhythms and sleep patterns can potentially impact negatively on performance. The effects of sleep loss on performance in many industrial and military systems have been amply documented (Wilkinson, 1992). Unlike some task environments having constraints that impose severe sleep deprivation (e.g., combat situations, medical practice; see Huey and Wickens, 1993), the air traffic control task imposes no such constraints. However, sleep-related disruptions in performance efficiency do remain a direct concern in air traffic control. In considering the relevance of sleep disruption to controller performance, we have already noted Stager et al.'s (1989) report of substantial operational errors at modest and low task load periods, associated with late night and early morning hours. Such a finding is consistent with the well-documented fluctuations in performance efficiency associated with human circadian rhythms, which fall to a low point in the late night, early morning hours (Huey and Wickens, 1993; Horne, 1988). Furthermore, survey data from air traffic controllers has revealed that night work caused increases in the subjective rating of sleepiness and a reduction of total amount of sleep during the work week (Melton, 1985; Melton et al., 1973, 1975; Smith et al., 1971; McAdaragh, 1995). Nevertheless, it is apparent that night performance in air traffic control is inevitable, so long as night operations continue in the national airspace. At issue is whether there are ways to ameliorate their negative effects. There is evidence that permanent assignments of some workers to a night shift never produces a full adaptation of circadian rhythms (and therefore restored performance efficiency) to the inverted day-night cycle (Huey and Wickens, 1993), and such a solution is discouraged in any case because of concerns about any disruption of family life for controllers assigned to a permanent night shift. If then, shifts are to be rotated, two issues arise. How often should such rotations occur and, if they occur frequently, whether they should be phase advanced or phase delayed (see Figure 6.3). Regarding the first issue, the evidence is equivocal. Folkard (1980) has argued that for cognitive/memory tasks, such as those of the air traffic controller, relatively rapid rotation is more advantageous; whereas more recently, Wilkinson (1992) has argued that it is better to change
OCR for page 132
Flight to the Future: Human Factors in Air Traffic Control FIGURE 6.3 Extreme examples of phase delay and phase advanced schedules. The hashed box represents an 8-hour work period positioned within each 24-hour day. shifts intermittently (i.e., once a week or longer), than continuously (i.e., working a different shift each 24-hour period). In any case, no matter what the frequency of shift change may be, there is now fairly uniform evidence that phase-delayed shifts, such as that shown in Figure 6.3a, are less disruptive than phase-advanced shifts (Figure 6.3b) (Barton and Folkard, 1993), in the same manner that recovery from westbound transoceanic flights, which expand the day, is more rapid than from eastbound flights, which contract the day (Graeber, 1988). There are at least two reasons for the advantage of the phase-delayed schedule. First, it appears that human circadian rhythms have a "preferred" cycle that is slightly longer than 24 hours, and hence it is easier for those rhythms to adapt to a temporary lengthening of the cycle than to a temporary shortening. Second, the phase-delayed schedule distributes the work week over a longer period of time, with greater and more regular sleep opportunities between work time. Unfortunately, the second reason is quite precisely the reason why air traffic controllers have generally preferred to opt for the phase-advanced schedule, as it is one that allows compression of 40 hours of duty time into a 4-day work week, allowing longer nonwork weekends (McAdaragh, 1995). In fact, a recent survey of 997 air traffic control specialists at 12 different facilities revealed that none had adopted the "performance preferred" phase-delayed schedule (McAdaragh, 1995). Thus, the current shift work preferences of controllers can degrade performance due to circadian rhythm disruption, sleep deficit, and accumulated fatigue. In conclusion, the relation of shift work to air traffic control performance remains an important one, particularly as sleep-related disruptions appear to be most prevalent in low-load, vigilance-like monitoring tasks (Huey and Wickens,
OCR for page 133
Flight to the Future: Human Factors in Air Traffic Control 1993). As we note in Chapter 12, it is precisely these kinds of tasks that may be come more prevalent in the more automated controller workstation of the future. CONCLUSIONS Projected increases in future air traffic threaten to pose substantial demands on the capacity, and potentially the safety, of the air traffic system. Controllers may experience peaks and troughs of workload. If safety is not to be compromised, individual controllers should not be subjected to overload due to high traffic density. High workload can lower performance and set an upper limit to traffic-handling capacity. Decreasing sector size or increasing the number of controllers does not necessarily solve this problem, because of the consequent increase in intersector and intercontroller coordination and communication. Low workload may result in boredom and reduced alertness, with consequent implications for handling emergencies. However, although factors conducive to high and low workload are prevalent in the air traffic control environment, this does not necessarily mean that all controllers experience extremes of workload. Most controllers use various adaptive strategies to manage their performance and subjective perceptions of task involvement. Various factors that influence mental workload in the air traffic control environment have been identified. These include airspace variables, display factors, and controller-pilot communications. Although studies examining each of these factors have been conducted, the precise relationships between these variables and workload, and the interrelationships of these variables, remain incompletely characterized. Moreover, the relationships between task load variables, controller mental workload, controller performance, and system performance are complex and not amenable to simple generalizations. Evidence linking operational errors to performance and workload has found that errors occur under low task load conditions. Such conditions may increase demands on controller monitoring and vigilance. Vigilance declines as the information-processing demands of target identification increase (e.g., high memory load, high event rate, high spatial uncertainty). The mental workload of maintaining vigilance is also high, contrary to the belief that boring tasks are undemanding. Studies examining vigilance during simulated air traffic control have shown that performance can be good, but it declines under high task load conditions. Current work-rest schedules do not appear to have a negative impact on controller performance, although subjective complaints of fatigue may occur. However, shift work and the consequent disruption of circadian rhythms and sleep loss continue to be a major source of concern. Current shift-work patterns (e.g., phase-advanced shifts and compressed work weeks) may result in degraded performance. There are examples of current and past automation that have led to a reduction
OCR for page 134
Flight to the Future: Human Factors in Air Traffic Control of controller workload. However, more generally, automation changes the pattern of the controller's workload. Automation can also increase workload due to demands on vigilance. These findings suggest that further implementation of automation in air traffic control systems must be preceded by systematic analysis of the impact of new technologies on controller workload and vigilance.
Representative terms from entire chapter: