modeling and predicting mental workload in en route air ... · placed on the human operator and the...

INTRODUCTION

With the rapid advancement of technology,complex work systems have evolved in which op-erators must adapt their decision making and per-formance in the face of dynamic, ever-changingenvironments, concurrent task demands, timepressure, and tactical constraints (Moray, 1997;Sheridan, 2002). The assessment and predictionof the mental workload associated with operatingsuch complex systems has long been recognizedas an important issue (e.g., Gopher & Donchin,1986; Moray, 1979). Mental workload – or justworkload – is the general term used to describethe mental cost of accomplishing task require-ments (Hart & Wickens, 1990; Wickens, 1992).Workload varies as a function of task demandsplaced on the human operator and the capacity ofthe operator to meet those demands (Gopher &Donchin, 1986; Hopkin, 1995). High levels of

workload occur when task demands exceed oper-ator capacity.

Research efforts in complex work systems suchas piloting (e.g., Wilson, 2002), unmanned aerialvehicle control (e.g., Dixon, Wickens, & Chang,2005), anesthesiology (e.g., Leedal & Smith, 2005),railway signaling (e.g., Pickup et al., 2005), andautomobile driving (e.g., Recarte & Nunes, 2003)have focused on identifying factors that influencemental workload and techniques for measuring it.In contrast, in the current paper we develop a the-oretical model of operator strategic behavior andworkload management, within the context of enroute air traffic control (ATC), through which task-related workload can be predicted within com-plex work systems. The model we present takesinto account the changing task priorities and man-agement of resources by operators as well as thefeedback that operators receive in response to theirinput. In the next section, we explain the nature of

Modeling and Predicting Mental Workload in En Route AirTraffic Control: Critical Review and Broader Implications

Shayne Loft, Penelope Sanderson, Andrew Neal, and Martijn Mooij, The University ofQueensland, Brisbane, Australia

Objective: We perform a critical review of research on mental workload in en routeair traffic control (ATC). We present a model of operator strategic behavior and work-load management through which workload can be predicted within ATC and othercomplex work systems. Background: Air traffic volume is increasing worldwide. Ifair traffic management organizations are to meet future demand safely, better modelsof controller workload are needed. Method: We present the theoretical model and thenreview investigations of how effectively traffic factors, airspace factors, and opera-tional constraints predict controller workload. Results: Although task demand has astrong relationship with workload, evidence suggests that the relationship dependson the capacity of the controllers to select priorities, manage their cognitive resources,and regulate their own performance. We review research on strategies employed bycontrollers to minimize the control activity and information-processing requirementsof control tasks. Conclusion: Controller workload will not be effectively modeled untilcontrollers’ strategies for regulating the cognitive impact of task demand have beenmodeled. Application: Actual and potential applications of our conclusions includea reorientation of workload modeling in complex work systems to capture the dynam-ic and adaptive nature of the operator’s work. Models based around workload regula-tion may be more useful in helping management organizations adapt to future controlregimens in complex work systems.

Address correspondence to Shayne Loft, ARC Key Centre for Human Factors and Applied Cognitive Psychology, Universityof Queensland, Brisbane 4072, Australia; [email protected]. HUMAN FACTORS, Vol. 49, No. 3, June 2007, pp. 376–399.DOI 10.1518/001872007X197017. Copyright © 2007, Human Factors and Ergonomics Society. All rights reserved.

PREDICTING MENTAL WORKLOAD 377

the work performed by en route air traffic con-trollers (ATCos) and provide an overview of ourapproach.

OVERVIEW

Controlled airspace is divided into sectors. Anen route sector is a region of airspace that is typ-ically situated at least 30 miles (~48 km) from anairport for which an associated ATCo has respon-sibility. ATCos have to accept aircraft into theirsector; check aircraft; issue instructions, clear-ances, and advice to pilots; and hand aircraft offto adjacent sectors or to airports. The radar screendisplays characteristics of the sector (e.g., bound-aries and airways), the spatial position of aircraft,and vital flight information (identifiers, altitude,speed, flight destination). When the aircraft leavesthe airspace assigned to the ATCo, control of theaircraft passes on to ATCo controlling the nextsector (or to the tower ATCo). As is typical inmany real-world complex systems, this environ-ment imposes multiple concurrent demands on theoperator. As Gronlund, Ohrt, Dougherty, Perry,and Manning (1998) described it, “In the en routeair traffic control environment (involving the high-speed and high-altitude cruise between takeoffand landing), the system that confronts the air traf-fic controller comprises a large number of aircraftcoming from a variety of directions, at diversespeeds and altitudes, heading to different desti-nations. Like most complex, dynamic systems,this one cannot be periodically halted while thecontroller takes a brief respite” (p. 263).

ATCos have two main goals. The primary goalis to ensure that aircraft under jurisdiction adhereto International Civil Aviation Organization(ICAO) mandated separation standards. For ex-ample, one of the most common separation stan-dards requires that aircraft under radar control beseparated by at least 1,000 feet vertically (2,000feet above 29,000 feet, unless reduced verticalseparation minima apply) and 5 nautical mileshorizontally. The secondary goal is to ensure thataircraft reach their destinations in an orderly andexpeditious manner. These goals require the ATCoto perform a variety of tasks, including monitoringair traffic, anticipating loss of separation (i.e., con-flicts) between aircraft, and intervening to resolveconflicts and minimize disruption to flow. (For anextensive compilation of the tasks and goals of enroute control, see Rodgers & Drechsler, 1993.)

Total world airline scheduled passenger trafficin terms of passenger-kilometers is projected togrow at an annual average rate of 4.4% over theperiod 2002 to 2015, according to forecasts pre-pared by the ICAO (2004). In the United Statesalone, the number of aircraft handled by ATCcenters is expected to increase from 46.2 millionin 2004 to more than 60.2 million in 2016 (Feder-al Aviation Administration, 2005). To accommo-date predicted traffic growth there is a need toincrease en route airspace capacity through theintroduction of new air traffic management sys-tems (e.g., free flight) or the adaptation of existingairspace designs (e.g., sector boundaries), con-troller tools (e.g., conflict resolution), and proce-dures (e.g., reduced separation minima). Theconsensus among research and operational com-munities is that it is important to understand thefactors that drive mental workload if they are toimprove airspace capacity (Christien, Benkouar,Chaboud, & Loubieres, 2003; Majumdar, Ochieng,McAuley, Lenzi, & Lepadatu, 2004).

Most research has focused on identifying char-acteristics of the air traffic picture that create taskdemand for ATCos (e.g., Grossberg,1989; Kirwan,Scaife, & Kennedy, 2001; Manning, Mills, Fox,& Pfleiderer, 2001). These characteristics includethe number of aircraft in transition though a sec-tor, the number of aircraft changing altitude, andthe number of potential conflicts. Several researchgroups have attempted to predict mental workloadon a moment-to-moment basis by using linearcombinations of task demand factors as predic-tors. The resulting sets of task demand factors areknown as dynamic density metrics. Studies haveshown that the dynamic density of the airspace ata given moment accounts for approximately halfthe variance in workload at that point in time(e.g., Kopardekar & Magyarits, 2003; Laudeman,Shelden, Branstrom, & Brasil, 1998). In psycho-logical terms, this represents relatively strong pre-diction. In practical terms, however, a significantproportion of variance remains unaccounted for.Furthermore, one goal of workload modeling isto allow ATC providers to predict workload levelsahead of time in order to allow them to put work-load management strategies in place. For example,this may include splitting a sector or introducingflow restrictions. However, to date, dynamic den-sity metrics have been unable to accurately pre-dict ATCo workload ahead of time (Kopardekar& Magyarits, 2003; Majumdar & Ochieng, 2002;

378 June 2007 – Human Factors

Masalonis, Callaham, & Wanke, 2003). Many re-searchers argue that these limitations stem fromthe fact that there is no simple linear relationshipbetween task demand and workload (e.g., Athènes,Averty, Puechmorel, Delahaye, & Collet, 2002;Chatterji & Sridhar, 2001). Moreover, these re-searchers view workload as an emergent propertyof the complex interaction between the ATCo andthe air traffic situation, rather than as a simpleoutcome of task demand inputs at a single pointin time.

The approach to predicting ATCo mental work-load that is presented in the current paper is in linewith these views. According to Sperandio (1971),workload is not something imposed upon a pas-sive ATCo but, rather, is something the ATCoactively manages. He proposed a model in whichchanges in strategy (primarily resource manage-ment) allow ATCos to regulate how task demandsare transformed into workload, thus keepingworkload within acceptable limits. In a paper thatdeserves to be better known, Rouse, Edwards, andHammer (1993) took a similar view, modelingworkload as a feedback control process driven bysubjective mental workload. Several current re-search groups agree with Sperandio’s (1971) viewthat a relationship between task demand and work-load can be better understood by considering howATCos use strategies to manage their resourcesand regulate their workload (Athènes et al., 2002;Averty, Collet, Dittmar, Vernet-Maury, & Athènes,2004; Cullen, 1999; Hilburn, 2004; Histon &Hansman, 2002; Majumdar et al., 2004). Anotherkey aspect of Sperandio’s (1971) approach is thatthe effect of ATCo control actions on the systemis fed back to the ATCo, such that future task de-mands are actively regulated by the ATCo (Pawlak,Brinton, Crouch, & Lancaster, 1996).

In the current paper, we present a model ofmental workload that puts ATCos in the loop withair traffic events, reacting to the consequences oftheir own proactive behavior. Without denying thevalidity of the task demand approach, we arguethat the link between task demand and workloadis largely connected to the manner in which ATCosmanage their resources. We begin by outlining ageneral model of workload in ATC and contrastit with previous approaches. We then review taskdemand research in ATC. A characteristic of thisresearch that limits its interpretability is the ex-tremely large list of methods and task demandfactors that have been reported. (For exhaustive

reviews, see Hilburn, 2004; and Mogford, Gutt-man, Morrow, & Kopardekar, 1995.) The modelpresented in this paper provides a framework forintegrating this literature and studying its poten-tial strengths and shortcomings. We then turn ourattention to the smaller body of research that hasfocused on ATCo control strategies (e.g., Amaldi& Leroux; 1995; Histon & Hansman, 2002) andto task models that have been built to simulate theperformance of ATCos (e.g., Callantine, 2002;Leiden, Kopardekar, & Green, 2003). Finally, wereview models in which researchers have attempt-ed to integrate task demands with human perfor-mance models in order to predict workload (e.g.,Averty et al., 2004; Cullen, 1999).

MENTAL WORKLOAD MODELINGARCHITECTURES

In this section we outline different modelingarchitectures that have been used to understandATCo mental workload. At the end, we present thearchitecture that guides our review of the litera-ture that follows.

Common Architectures

Aprevalent model in ATC mental workload re-search is that different properties, or task demands,of the air traffic situation will pose problems ofdifferent levels of complexity to the ATCo and,depending on the ATCo’s skill, experience, andstrategy, will produce different levels of subjectiveworkload. This is effectively an open-loop model,as shown in Figure 1. Variants of this general open-loop architecture have been used to model sourcesof ATCo workload. For example, Hilburn andJorna’s (2001) model shows system factors com-bining to create task demand, which, in accordancewith operator factors such as skill, strategy, andexperience, will lead to some degree of workload.Similarly, Mogford et al.’s (1995) model depictsa relationship between source factors (objectivecomplexity, air traffic patterns, sector character-istics) and workload being mediated by qualityof equipment, individual differences, and ATCostrategies. Both these models acknowledge thatATCo strategy can influence workload. Neverthe-less, researchers using these models as frame-works have tended to focus on whether individualaspects of the ATC environment affect workload.We argue that the tendency to seek input-outputrelations (“does increasing the number of aircraft


increase workload?” or “how does strategy influ-ence the impact of the number of aircraft on work-load?”) fails to take into account fully the goalsand management of resources by the ATCo andfeedback that the ATCo receives from the systemin response to his or her input.

Sperandio’s Architecture

It is interesting to contrast the aforementionedmodels with that of Sperandio (1971), shown inFigure 2. Again, task demand is on the left andmental workload on the right. Consistent with theprevious two models, Sperandio (1971) proposedthat ATCo strategy is an intervening variable be-tween task demand and the work achieved and thatthe ATCo selects strategies to keep mental work-load within acceptable limits. However, in contrastto the models mentioned in the previous para-graph, the Sperandio (1971) model includes twofeedback control loops. First, variation in mentalworkload resulting from work methods has,through feedback, a regulating effect on the choiceof work methods (Feedback Loop 1). Second, thework method used in response to perceived taskdemands regulates the task demand encountered

in the future (Feedback Loop 2). Sperandio (1971)emphasized that it is the change in workload, notthe change in task demand, that explains thechange in strategy. The change in strategy changeswhat information is extracted from the airspaceand thus affects workload. The relationship amongtask demand, strategy, and workload is adaptiveand so can be shown only in a feedback controldiagram, as shown in Figure 2. This position isconsistent with that of Rouse et al. (1993) but isin contrast to much subsequent research in whichmental workload is predicted from objectivelymeasured characteristics of the airspace, tuned bystrategy and other factors.

A Systems Approach to Modeling MentalWorkload

The basis of our approach is that the ATCo isin a continuous relationship with a dynamic worldand is an adaptive element in that world (Athèneset al., 2002; Pawlak et al., 1996; Sperandio, 1971).As a result, mental workload cannot be a functionsolely of task demands; it is also a function of thestrategy the ATCo uses to manage traffic andwhether the strategy, once invoked, has provided

Air traffic factors

Mediating factors

Mental workload

Task demandAir traffic

factorsMediating factors

Mental workload

Task demand

Figure 1. Generic form of an open-loop model of mental workload implicitly adopted in many studies of ATCo men-tal workload.

Input:Work

assigned

Choice of work method

Output:Workdone

Feedback loop 1

Feedback loop 2

Task demand

Input: Output:

Feedback loop 1

Feedback loop 2

Mental workload

Task demand

Figure 2. Sperandio’s (1971) closed-loop model of ATCo mental workload in which actual work done influenceschoice of work methods and amount of work accepted. Task demand is equivalent to work to be done. Variation ofoperator’s strategies and regulating effects on workload, J. C. Sperandio, Ergonomics, (1971), Taylor & Francis Ltd,adapted by permission of the publisher (Taylor & Francis Ltd, http://www.tandf.co.uk/journals).


a comfortable level of control over task demands.In the next section, we work through the ATC lit-erature with the help of the model that is shown inFigure 3. In an adaptation of Sperandio’s (1971)model, our model represents two control loops thatgovern ATCo activity: first, the management ofworkload by the internal reorganization of prior-ities leading to a different strategy; and second,the management of workload by explicit controlof the airspace. The work of the ATCo occupiesthe shaded part in the centre of Figure 3, where-as the world that the ATCo controls is shown inthe perimeter.

The model in Figure 3 simplifies the worldthat the ATCo controls. It shows two aircraft,which is the minimum needed to indicate that theATCo is concerned with managing relationshipsamong aircraft rather than controlling single air-craft in isolation. One aircraft is at the top of Fig-ure 3 and the other at the bottom. Each aircraftreceives instructions from the ATCo (see plussigns [+] on links into adder symbols at top rightand bottom right of Figure 3). The pilot considersthe difference between the instruction and the air-

craft’s current flight profile (see minus sign [–]on feedback loops coming into rightmost addersymbols) and makes an appropriate adjustmentto the aircraft’s flight profile. The aircraft’s newflight profile is combined with the flight profile ofall other aircraft (see two + signs entering adderat left) and becomes the task demand fed back tothe ATCo. The ATCo can take action to change fu-ture task demand fed back through the system(e.g., by accepting aircraft early or by putting air-craft in a holding pattern) or he or she can changefuture task demand with cognitive strategies (e.g.,by considering a set of aircraft as one for purpos-es of control).

The ATCo remains aware of work to be doneby monitoring present task demands (see left ofFigure 3). The work to be done is translated intoactual work done through the control activitiesperformed. As Figure 3 suggests, a set of controlactivities can be classified as a strategy. A strategycan be described as a specific class of air trafficmanagement that achieves one or more objectives(e.g., safety, orderliness, expeditiousness) with acertain investment of time and effort. Selection

A/C 2

+

+

PilotA/C 1

Pilot A/C 2

Control activities/strategies

Prioritization

Meta cognition

+

-

-

demands

A/C 2

PilotA/C 1

PilotA/C 2

ATCo

Control activities/strategies

Prioritization

Meta cognition

Feedforward Feedback

Task

Task demands

Work tobe done

Actual workdone

+

+

-

Figure 3. Model of ATCo activity in which strategy is controlled through feedback and feedforward information aboutmental workload. Work to be done represents how objective task demands are mentally represented by the ATCo.


among strategies is driven by the relative priorityof the ATCo’s objectives as the work to be doneevolves over time (Kallus, Van Damme, & Ditt-man, 1999; Kirwan & Flynn, 2002; Niessen,Eyferth, & Bierwagen, 1999). The strategy chosenwill lead to a certain quality of control over thesituation, which will often reflect subjective timepressure. For example, the contextual control mod-el (COCOM) of Hollnagel (2002) and Hollnageland Woods (2005) distinguishes strategic, tactical,opportunistic, and scrambled control as the resultof an operator’s subjective judgment of the rela-tionship between time available for action (Ta) andthe time required to evaluate the situation (Te),select a response (Ts), and perform the response(Tr). In their COCOM model, strategic and tacti-cal control are mostly proactive, whereas oppor-tunistic and scrambled control are mostly reactive.An ATCo will work to achieve strategic controland avoid scrambled control as much as possible.

Figure 3 indicates that prioritization drivescontrol activities/strategies. Prioritization refersto the professional set of values that guide controlof air movements at any point, such as safety,orderliness, and expeditiousness. Prioritization,in turn, is driven by metacognitive factors such asawareness of time available to perform tasks,anticipation of future difficulties, and the ATCo’sknowledge of his or her capacity. However, we donot assume that this is a conscious process. In-stead, prioritization is a consideration satisfieddirectly or indirectly through control action. Thestrategy may be a learned response that is not opento introspection, and it may not require explicitconsideration of safety, orderliness, or expedi-tiousness. However, this does not remove the factthat strategies will always reflect some balance ofpriority among safety, orderliness, and expedi-tiousness. The selection of strategy will be logi-cally guided by the appropriate priority even if thepriority is not consciously accessed.

Focusing in particular on time, Schmidt (1978)predicted ATCo mental workload with a queuingtheory model based on the relationship betweenthe frequency of observable tasks that require de-cisions/actions and the time required to make thesedecisions/actions. More recently, Hendy, Liao,and Milgram (1997) used a simulated ATC task andmodeled overall workload as a univariate functionof time pressure. Time pressure, in turn, was mod-eled as the ratio of time available to time required,also expressible as the ratio of the information-

processing rate demanded by a task and the max-imum information-processing capacity of theATCo. Performance data were well matched bythe model. Most recently, Rantanen and Levinthal(2005) demonstrated that an ATCo’s time to firstintervention in resolving conflicts was faster whenthe ratio of time to act to the duration of a windowof opportunity for action was small – in otherwords, when there was less discretionary time.

The dashed lines within the shaded ATCo partof Figure 3 indicate that metacognition can beinfluenced by both feedforward and feedback sig-nals. Looking at feedforward (at left), the ATComay be aware that a large number of aircraft areabout to enter the sector and thus adjusts his or herstrategy for handling traffic already on frequency.Looking at feedback (at right), the ATCo may beaware that the quality of actual work done mayhave been compromised by time pressure and thusadjust priorities toward achieving safety at thepossible expense of expeditiousness (e.g., by rear-ranging the trajectories of aircraft in order to min-imize monitoring and coordination requirements).The adder to the right of metacognition in Figure3 shows that the ATCo may notice differencesbetween the work to be done and the actual workthat is getting done, which will also trigger a meta-cognitive response. For example, the ATCo mayrealize that a heavy communication load is mak-ing him or her fall behind in dealing with work tobe done, yet an even heavier communication loadis anticipated. A rearrangement of priorities mayoffer a control strategy that has a less intense com-munications load. In these ways the model in Fig-ure 3 represents the fact that ATCos behave andreact to the consequences of their behavior – inrespect both to the perceived discrepancy betweencurrent goals and system state and to the ATCo’sunderstanding of his or her own capacity – andthat these processes drive mental workload.

In the following sections we review researchon the relationships between task demand andmental workload and between operator capacityand mental workload. Then we review researchthat combines the two in ways that are at least par-tially consistent with the model in Figure 3.

TASK DEMANDS AND MENTAL WORKLOAD

Research concerning the relationship betweentask demand and mental workload has had a long


history, dating back more than 40 years (Arad,1964; Couluris & Schmidt, 1973; Davis, Dana-her, & Fischl, 1963; Hurst & Rose, 1978; Schmidt,1976). This research has focused on uncoveringproperties of the air traffic environment that con-tribute to cognitive complexity and, via tuning fac-tors such as skill, strategy and experience, resultin workload (see Figure 1). The task demand liter-ature, in its own right, provides a solid basis fromwhich to model the complexity of task demands(work to be done) in the ATC system. Researchershave expended great effort in developing predic-tive models based on task demand, and the factthat these models have been able to account forsignificant variance in ATCo workload warrantsa state-of-the-art review and synthesis. We main-tain, however, that focusing only on task demandoverlooks the reciprocal interactions presented inFigure 3. Subsequently, we examine whether thetask demand literature explicitly or implicitly mod-els these aspects of control, and we outline con-tradictions in the literature that result from thefailure to take a systems view.

Our review of task demand research is sum-marized in Table 1. The material is drawn fromgovernment and contractor technical reports, op-erational reviews, journal articles, and book chap-ters. Much material originates in the United Statesand Europe, primarily from the Federal AviationAdministration, the National Aeronautics andSpace Administration (NASA), and the EuropeanOrganisation for the Safety of Air Navigation(EUROCONTROL). During the review process,we found that several aspects of task demandresearch limited its interpretability. Two valuablesurveys of research relating to task demand (Hil-burn, 2004; Mogford et al., 1995) do not explicit-ly address these issues. These constraints, and howwe dealt with them, will be briefly discussed next.

First, systematic comparison among studieswas complicated by the wide variety of researchmethodologies reported. As presented in Table 1,these methodologies include knowledge elicita-tion techniques such as verbal protocol analysis(e.g., Pawlak et al., 1996), experiments in whichresearchers made predictions a priori about howmental workload will vary with systematic manip-ulation of task demand (e.g., Boag, Neal, Loft, &Halford, 2006), and correlational studies in whichresearchers extracted values for task demand fac-tors from flight data and correlated these valueswith workload on a post hoc basis (e.g., Kopar-

dekar & Magyarits, 2003). A second characteris-tic of task demand research that limits cross-studycomparison is the wide variety of workload mea-sures used. An evaluation of the advantages anddisadvantages of each approach is beyond thescope of this paper (see Farmer & Brownson,2003; Hilburn & Jorna, 2001). However, Table 1categorizes studies according to the workload cri-terion employed. As is evident from Table 1, themeasurement of workload is far from uniform.

Table 1 reveals that most studies have focusedon identifying traffic factors. Traffic factors re-flect the instantaneous distribution of air traffic ina sector in terms of both the number of aircraft andthe complexity of their relationships. However,ATCos can actively regulate the mental workloadassociated with traffic factors by using economi-cal control strategies. Researchers such as Hilburn(2004) and Histon and Hansman (2002) identifiedtwo further factors that influence the choice andeffectiveness of ATCo control strategies and whichare generally independent from traffic factors.Airspace factors reflect the underlying structuralproperties of the airspace (e.g., number of cross-ing altitude profiles). Airspace factors constrainthe relationship between traffic factors and work-load by shaping the evolution of air traffic andcreating predictable air traffic patterns that can beexploited by ATCos. Operational constraints re-flect operational requirements (e.g., restrictions ofavailable airspace) that place restrictions on ATCostrategy and control action.

Traffic Factors Predicting MentalWorkload

Task demand research has typically focused onexplicit properties of the distribution of aircraftthat predict mental workload and are computed inreal time using radar track data or derivationsthereof. Of all the traffic factors, the aircraft count,or the number of aircraft under control, is the mostpowerful predictor of workload (e.g., Hurst &Rose, 1978; Kopardekar & Magyarits, 2003; Man-ning et al., 2001). High aircraft count leads to anincrease in workload because it increases the mon-itoring, communication and coordination requiredto handle aircraft in a safe, orderly, and expedi-tious manner.

Density factors are derivatives of traffic countthat measure the horizontal and vertical distancesbetween aircraft and the way in which these dis-tances change with time. Various measures of


traffic density have been developed, including theaverage horizontal (or lateral) separation distancesbetween aircraft (Chatterji & Sridhar, 2001) andaircraft counts divided by sector volumes (Kopar-dekar & Magyarits, 2003). These density measuresassume, perhaps erroneously, that all aircraft inclose proximity are noticed by the ATCo and there-fore exert some influence on mental workload. Inreality, the influence of aircraft density on work-load will depend on what the aircraft are doing inrelation to each other – for example, whether theyare converging or diverging. For this reason, den-sity factors based on the minimum separations be-tween pairs of aircraft are more sensitive (Chatterji& Sridhar, 2001). Close future minimum separa-tion between an aircraft pair will undoubtedly cap-ture ATCo attention because of the likelihood ofa separation violation, reducing the ATCo’s capac-ity to attend to other control tasks.

Although traffic count and density adequatelyreflect the number of routine aircraft-associatedtasks that an ATCo has to perform within a certaintime frame, the complexity of air traffic is impor-tant in determining the difficulty of the tasks orevents handled and thus the resulting mentalworkload. ATCos report that they can handle rel-atively large volumes of traffic if the aircraft areflying on regular routes and the flow is orderly(e.g., Amaldi & Leroux, 1995; Mogford et al.,1995). In contrast, small volumes of traffic canlead to overload if aircraft interact in complexways (Kallus, Van Damme, & Dittman, 1999;Mogford et al., 1995). Complexity factors fallinto two categories. The first are commonly re-ferred to as aircraft transition factors and capturechanges in an aircraft’s state in any of the threeaxes of altitude (e.g., Lamoureux, 1999), speed(e.g., Kopardekar & Magyarits, 2003), or heading(e.g., Laudeman et al., 1998). Performance mixof aircraft is also an important aircraft transitionfactor (Schaefer, Meckiff, Magill, Pirard, & Aligne,2001). In order to calculate the minimum separa-tion between two aircraft in altitude transition,the ATCo needs to know how fast the aircraft willclimb (or descend) relative to each other. Pre-sumably, this job would be made harder by in-creased variability in performance profiles.

The second set of complexity factors relates tothe number and nature of potential conflicts with-in a sector. Potential conflicts emerge from thecombination of density and transition factors pre-sent at any time. The influence of potential con-

flicts on mental workload depends on their spe-cific properties. For example, Boag et al. (2006)developed a “transitions metric” for assessing thedifficulty of judging whether a pair of aircraft willbe in lateral and vertical conflict at the same time.If two aircraft are on converging flight paths, andare both maintaining the same level, then theATCo simply needs to assess whether they willviolate the lateral separation standard. This prob-lem is relatively simple because the ATCo needconsider only one transition (the transition into lat-eral conflict). However, if the aircraft are chang-ing levels, then the ATCo must assess when theaircraft will violate and regain separation in onedimension and also whether the aircraft will be inconflict in the other dimension at the same time.Up to four transitions (into and out of lateral andvertical conflict) may be possible. The Boag et al.(2006) transitions metric accounted for signifi-cant amounts of variance in ATCos’ ratings ofcomplexity and workload. Imminent violationsof separation also increase workload (Chatterji &Sridhar, 2001). The time available for an ATCo todetect and respond to a potential conflict affectshow difficult the conflict is to resolve. A conflictthat develops quickly gives the ATCo only a lim-ited time to act, creating significant time pressure.Conflicts in close proximity to sector boundaries(e.g., Pawlak et al., 1996) and/or high numbers ofsurrounding aircraft (e.g., Kopardekar & Magyar-its, 2003) also constrain how conflicts can beresolved, reducing the number of options formaneuvering.

Combining Task Demand Factors IntoDynamic Density Metrics

Several research groups (Kopardekar & Mag-yarits, 2003; Laudeman et al., 1998; Masaloniset al., 2003) have used regression models to iden-tify sets of factors that best predict mental work-load and have created algorithms by weightingthese different factors according to their predictivepower. The resulting algorithms are commonlyreferred to as dynamic density metrics. Dynamicdensity has been defined as “the collective effortof all factors, or variables, that contribute to sector-level air traffic control complexity or difficulty atany point in time” (Kopardekar & Magyarits,2003, p. 1). The Laudeman et al. (1998) metric isperhaps the best known, describing dynamic den-sity as the sum of the density of traffic weightedby the number of changes in speed, heading, and

384

TAB

LE 1

:Stu

die

s P

red

ictin

g A

TCo

Men

tal W

ork

load

: Met

hod

s o

f M

easu

rem

ent

and

Sum

mar

ies

of

Res

ults

Aut

hors

Met

hod

Mea

sure

sTa

sk D

eman

d F

acto

rsSy

stem

s V

iew

Ara

d (1

964)

Co

rrel

atio

nal

ATC

o a

ctiv

ity

Traf

fic, a

irsp

ace,

and

Sect

or

bo

und

arie

s al

igne

d w

ith

stan

dar

d fl

ow

s re

duc

e o

per

atio

nal c

ons

trai

nts

the

tim

e p

ress

ure

of

cont

rol a

ctiv

itie

s.

Bo

ag e

t al

.E

xper

imen

t: p

art-

task

MW

rat

ing

sTr

affic

The

num

ber

of

confl

ict

bo

und

ary

tran

siti

ons

incr

ease

s (2

006)

sim

ulat

ion

Co

mp

lexi

ty r

atin

gs

MW

bec

ause

ATC

os

cann

ot

red

uce

info

rmat

ion

set.

Buc

kley

, DeB

arys

he,

Exp

erim

ent:

ful

l tas

kA

TCo

act

ivit

yTr

affic

and

air

spac

eA

irsp

ace

fact

ors

inte

ract

wit

h tr

affic

fac

tors

in d

eter

-H

itch

ner,

& K

ohn

sim

ulat

ion

min

ing

MW

; ATC

os

use

airs

pac

e st

ruct

ure

to s

imp

lify

(198

3)ai

r tr

affic

.

Cha

tter

ji &

Sri

dha

rC

orr

elat

iona

l: d

ynam

icM

W r

atin

gs

Traf

ficM

od

el in

corp

ora

tes

cog

niti

ve a

spec

ts (e

.g.,

tim

e p

res-

(200

1)d

ensi

tysu

re) i

nto

cho

ice

of

traf

fic c

om

ple

xity

fac

tors

.

Co

ulur

is &

Sch

mid

tE

xper

imen

t: F

ull t

ask

ATC

o a

ctiv

ity

Air

spac

eU

sing

ATC

o ac

tivity

as

mea

sure

of M

W fa

ils t

o ta

ke in

to

(197

3)si

mul

atio

n ac

coun

t th

e in

tent

of

cont

rol a

ctio

n.

Dav

is e

t al

. (19

63)

Exp

erim

ent:

Ful

l tas

kA

TCo

act

ivit

yTr

affic

Usi

ng A

TCo

activ

ity a

s m

easu

re o

f MW

fails

to

take

into

si

mul

atio

n ac

coun

t th

e in

tent

of

cont

rol a

ctio

n.

Gal

ster

et

al. (

2001

)E

xper

imen

t: P

art-

task

MW

rat

ing

sTr

affic

Del

ays

in c

onfl

ict

det

ecti

on

imp

ose

co

nstr

aint

s o

n si

mul

atio

n Se

cond

ary

task

co

nflic

t re

solu

tio

n.

His

ton

& H

ansm

anK

now

led

ge

elic

itat

ion:

live

Co

mp

lexi

ty r

atin

gs

Traf

fic, a

irsp

ace,

and

Air

spac

e fa

cto

rs in

tera

ct w

ith

traf

fic f

acto

rs in

det

er-

(200

2)o

bse

rvat

ion

and

inte

rvie

w

op

erat

iona

l co

nstr

aint

sm

inin

g M

W; A

TCo

s us

e ai

rsp

ace

stru

ctur

e to

sim

plif

y ai

r tr

affic

.

Hur

st &

Ros

e (1

978)

Co

rrel

atio

nal:

full

task

ATC

o a

ctiv

ity

Traf

ficU

sing

ATC

o a

ctiv

ity

as m

easu

re o

f M

W f

ails

to

tak

e si

mul

atio

n in

to a

cco

unt

the

inte

nt o

f co

ntro

l act

ion.

Kir

wan

et

al. (

2001

)K

now

led

ge

elic

itat

ion:

Co

mp

lexi

ty r

atin

gs

Traf

fic, a

irsp

ace,

and

ATC

os

use

airs

pac

e st

ruct

ure

to s

imp

lify

air

traf

fic.

gro

up ju

dg

men

t o

per

atio

nal c

ons

trai

nts

Ko

par

dek

ar &

Co

rrel

atio

nal:

unifi

edC

om

ple

xity

rat

ing

sTr

affic

and

air

spac

eA

irsp

ace

fact

ors

pla

y an

imp

ort

ant

role

in p

red

icti

ng

Mag

yarit

s (2

003)

dyn

amic

den

sity

MW

acr

oss

sec

tors

.

Lam

our

eux

(199

9)E

xper

imen

t: p

art-

task

MW

rat

ing

sTr

affic

M

W is

link

ed t

o th

e m

enta

l cal

cula

tions

and

pro

ject

ions

si

mul

atio

n in

volv

ed in

man

agin

g g

roup

s o

f A

/C.

Laud

eman

et

al.

Kno

wle

dg

e el

icit

atio

n:A

TCo

act

ivit

y Tr

affic

Ack

now

led

ged

tha

t va

riat

ion

in t

he in

tent

of

ATC

o

(198

8)in

terv

iew

cont

rol a

ctio

n ca

n in

fluen

ce t

he r

elat

ions

hip

bet

wee

n C

orr

elat

iona

l: d

ynam

icA

TCo

act

ivit

y an

d M

W.

den

sity

385

Man

ning

et

al.

Cor

rela

tiona

l:A

TCo

activ

ityM

W r

atin

gs

Traf

fic

ATC

o a

ctiv

ity

do

es n

ot

sig

nific

antl

y p

red

ict

MW

ove

r (2

001)

fro

m r

out

ine

flig

ht d

ata

and

ab

ove

tra

ffic

coun

t; f

ails

to

tak

e in

to a

cco

unt

the

inte

nt o

f co

ntro

l act

ion.

Man

ning

, Mill

s, F

ox,

Cor

rela

tiona

l:A

TCo

activ

ityM

W r

atin

gs

Traf

fic

No

. and

typ

e o

f A

TCo

co

mm

unic

atio

ns d

o n

ot

sig

nifi-

Pfle

ider

er, &

fro

m r

out

ine

flig

ht d

ata

cant

ly p

red

ict

MW

; fai

ls t

o t

ake

into

acc

oun

t th

e in

tent

M

og

ilka

(200

2)o

f A

TCo

co

mm

unic

atio

ns.

Mas

alo

nis

et a

l.K

now

led

ge

elic

itat

ion:

Co

mp

lexi

ty r

atin

gs

Traf

fic, a

irsp

ace,

and

Met

ric

unab

le t

o a

ccur

atel

y p

red

ict

MW

ahe

ad o

f (2

003)

inte

rvie

wo

per

atio

nal c

ons

trai

nts

tim

e; f

ails

to

tak

e in

to a

cco

unt

the

influ

ence

of

ATC

o

Co

rrel

atio

nal:

dyn

amic

cont

rol s

trat

egy/

acti

vity

on

futu

re t

ask

dem

and

tha

t is

d

ensi

tyfe

d b

ack

thro

ugh

syst

em.

Met

zger

& P

aras

ur-

Exp

erim

ent:

par

t-ta

skM

W r

atin

gs

Traf

fic

Del

ays

in c

onfl

ict

det

ecti

on

imp

ose

co

nstr

aint

s o

n am

an (2

001)

sim

ulat

ion

Seco

ndar

y ta

sk

confl

ict

reso

luti

on.

Mo

gfo

rd e

t al

. K

now

led

ge

elic

itat

ion:

Co

mp

lexi

ty r

atin

gs

Traf

fic, a

irsp

ace,

and

Airs

pac

e fa

cto

rs a

nd o

per

atio

nal c

ons

trai

nts

pla

y a

role

(1

993)

rati

ng a

nd r

anki

ng

op

erat

iona

l co

nstr

aint

sin

sha

pin

g t

he r

elat

ions

hip

bet

wee

n tr

affic

fac

tors

and

M

W.

Paw

lak

et a

l. (1

996)

Kno

wle

dg

e el

icit

atio

n:M

W r

atin

gs

Traf

fic, a

irsp

ace,

and

Em

pha

sis

on

fact

ors

tha

t im

pac

t o

n th

e co

gni

tive

verb

al p

roto

col t

echn

ique

Co

mp

lexi

ty r

atin

gs

oper

atio

nal c

onst

rain

tsac

tivi

ty o

f A

TCo

s, a

s o

pp

ose

d t

o b

ehav

iora

l ind

icat

ors

(e

.g.,

ATC

o a

ctiv

ity)

.

Scha

efer

et

al.

Kno

wle

dg

e el

icit

atio

n:C

om

ple

xity

rat

ing

sTr

affic

, air

spac

e, a

ndA

irsp

ace

fact

ors

inte

ract

wit

h tr

affic

fac

tors

in d

eter

-(2

001)

inte

rvie

wop

erat

iona

l con

stra

ints

min

ing

MW

.

Stei

n (1

985)

Exp

erim

ent:

full

task

MW

rat

ing

sTr

affic

No

. of

hand

off

s (in

bo

und

and

out

bo

und

) pre

dic

t M

W;

sim

ulat

ion

how

ever

, hig

hly

corr

elat

ed w

ith

A/C

co

unt.

Wyn

dem

ere,

Inc.

Kno

wle

dg

e el

icit

atio

n:M

W r

atin

gs

Traf

fic a

nd a

irsp

ace

Met

ric

incl

uded

sev

eral

air

spac

e fa

cto

rs.

(199

6)cr

itic

al d

ecis

ion

Co

rrel

atio

nal:

dyn

amic

den

sity

No

te.M

W =

men

tal w

ork

load

; A/C

= a

ircra

ft.


altitude; the proximity of aircraft; and the time un-til predicted conflicts. This metric accounted for22% of the variance in ATCo activity (a proxy forworkload) not predicted by aircraft count. Kopar-dekar and Magyarits (2003) incorporated 23 fac-tors from four published dynamic density metricsto form a composite metric. This unified dynam-ic density accounted for 39% of the variance inATCo complexity ratings, significantly more thanaircraft count alone.

Airspace Factors and OperationalConstraints Predicting Mental Workload

Airspace factors and operational constraintsare key contributors to ATCo task demand andmental workload (Histon & Hansman, 2002; Kir-wan et al., 2001). Airspace factors refer to the un-derlying structural properties of the airspace,whereas operational constraints refer to temporaryvariations in operational conditions within the air-space. Airspace factors constrain the relationshipbetween traffic factors and workload by shapingthe evolution of air traffic and creating predictableair traffic patterns. Knowledge of these patternslets the ATCo use information-processing stra-tegies that simplify air traffic management. Fur-thermore, temporary variations in operationalconditions, such as communications limitations(Mogford, Murphy, Yastrop, Guttman, & Roske-Hofstrand, 1993), can restrict ATCo control actionand strategy.

Studies examining airspace factors have foundthat the size of a sector can influence the mentalworkload imposed by traffic factors (Arad, 1964;Histon & Hansman, 2002). On the one hand, alarger sector size will typically increase the num-ber of aircraft in the sector and the number of po-tential events that require attention. On the otherhand, events evolve faster in smaller sectors, andlimited space in a sector can reduce the optionsfor conflict resolution. Increased number of avail-able flight levels can reduce workload becausethey allow ATCos to maintain separation usingvertical separation (Histon & Hansman, 2002;Kirwan et al., 2001). Traffic events occurring closeto the outside of sector boundaries are importantfurther determinants of workload (Couluris &Schmidt, 1973; Histon & Hansman, 2002) be-cause they can cause the ATCo’s “area of regard”to be greater than the official dimensions of thesector. Aircraft events occurring outside sectorboundaries require attention because they can

affect aircraft currently in the sector, increasingthe complexity of coordination (handoffs, pointouts) with adjacent ATCos (e.g., Kirwan et al.,2001; Mogford et al., 1993). However, the pres-ence of well-defined ingress and egress points(Histon & Hansman, 2002) lets ATCos anticipateproblems, thus reducing workload.

The number, orientation, and complexity ofstandard flows also influence the mental workloadimposed by traffic factors (Histon & Hansman,2002; Schaefer et al., 2001). Standard flows areaircraft flow patterns that emerge from underly-ing airway structure, standardized procedures,and other regular constraints such as ingress andegress points. ATCos use their knowledge of stan-dard flows to create important structure-basedabstractions that simplify the management of airtraffic (Histon & Hansman, 2002). For example,ATCos can simplify the search process involved inconflict detection by focusing on known crossingpoints and/or known crossing altitude profiles be-tween standard flows (Histon & Hansman, 2002;Pawlak et al., 1996). Knowledge of these “hotspots” can reduce the workload associated withmanaging air traffic (Amaldi & Leroux, 1995;Kallus, Van Damme, & Dittman, 1999).

Several operational constraints place restric-tions on ATCo control action. For example, re-strictions on available airspace can result fromconvective weather, activation of special-use air-space, or aircraft in holding patterns (Kirwan etal., 2001; Mogford et al., 1993). Restrictions onavailable airspace increase the likelihood of sep-aration violations. In addition, restricted airspacerequires more precisely planned conflict resolutionstrategies. Procedural restrictions, such as miles-in-trail spacing, can also constrain traffic flow andconflict resolution strategies (Histon & Hansman,2002; Pawlak et al., 1996).

Because of practical constraints, weightings fortask demand factors are typically validated againstone or only a few sectors (Histon & Hansman,2002), limiting the variation observed in airspacefactors and operational constraints. As a result, dy-namic density metrics developed using specificsectors perform less effectively when extended toother sectors. However, two research groups haverecently incorporated airspace factors into theirmetrics (Kopardekar & Magyarits, 2003; Masa-lonis et al., 2003). For example, the Kopardekarand Magyarits (2003) unified metric was devel-oped across four sectors, and the metric performs

differently across them. Nevertheless, comparisonsacross the different sectors revealed the contri-bution of airspace factors. Factors with significantpredictive value included the altitude level of thesector (high vs. low), the structure of the airspace,and the size of the sector. It seems that if research-ers wish to predict mental workload across sectors,then airspace factors and operational constraintswill be important because they mediate the effectof traffic factors on workload. However, if re-searchers wish to predict workload within sectors,then the airspace factors and operational con-straints become less important.

Summary and Assessment

Research focusing on task demand factors hasshown that task demand accounts for a significantamount of variance in mental workload. We soughtto develop our model of workload by investigatinghow different task demand factors might affectATCos’selection of strategies for control and thusaffect workload. However, this was made difficultby the lack of research examining how airspacestructure and aircraft configuration relate to theselection of strategies for control, as depicted bythe model in Figure 3. We argue that a significantlimitation of the task demand approach is that itviews the ATCo as a passive recipient of task de-mand. It does not explicitly take into account thefact that ATCos can actively take steps that changetask demand, so as to keep workload at an accept-able level. Without denying the importance oftask demands, we believe that workload might bemore strongly connected to the ATCo’s ability tomanage his or her cognitive capacity, as describedin the next section.

It is difficult to assess which aspects of taskdemand are most closely causally related to men-tal workload. One reason is multicollinearity.Ideally each traffic factor in a predictive modelshould contribute to workload relatively indepen-dently of other traffic factors, but this is seldom so.For example, in the Korpardekar and Magyarits(2003) unified metric, traffic count appears in sev-eral forms, such as sector volume (allowing moreaircraft), number of aircraft, and aircraft countsquared. However, the causal connection betweentraffic count and workload might be mediated bya number of complexity factors. For example, fac-tors such as traffic density, number of speed tran-sitions, number of conflicts, and number of aircraftnear sector boundaries all depend on traffic count.

In addition, under some conditions traffic countmay carry the key causal connection because ofthe increase in low-level activities needed, where-as under other conditions complexity propertiesemerging from traffic count may carry the keycausal connection because of the need to resolvecomplex traffic situations. In addition, many fac-tors measuring the complexity of traffic situationsare closely related. For example, the number of po-tential conflicts may depend on the number ofspeed, heading, and altitude variations. Overall,problems of multicollinearity make it difficult todetermine how task demand affects workload.The problem with taking each possible task de-mand predictor and putting it in a regression equa-tion is that the relative importance of each predictordepends on what other predictors have been in-cluded in the equation.

Even if the problem of multicollinearity can beresolved, task demand is still insufficient to ac-count for mental workload. First, combinations oftask demand factors rarely account for more thanhalf the variance in workload or complexity rat-ings (Kopardekar & Magyarits, 2003; Majumdar& Ochieng, 2002). Although there may be objec-tive and measurable features of sectors and aircraftflow, the difficulty of controlling traffic is theATCo’s subjective experience. If ATCos have al-ternative work methods for meeting increases intask demand, there will not necessarily be a lin-ear relation between task demand and workload(Chatterji & Sridhar, 2001; Hilburn, 2004). Fur-thermore, task demand approaches do not takeinto account the ATCo’s intent when extractingtask demand predictors from radar track data. Forexample, changes in aircraft altitude are weightedso that they vary directly with workload. However,an ATCo could have various reasons for issuing achange in flight level, many of which may actual-ly reduce workload (e.g., by ensuring separation).

A second reason that task demand is an insuf-ficient basis for modeling mental workload is theneed to predict mental workload ahead of time. Itwould be helpful to be able to predict probableaircraft trajectories from their flight plans andestimate the workload that the probable trajecto-ries will impose on the ATCo. The problem is thattask demands change dynamically; ATCos changethe trajectories of aircraft when they intervene toensure separation and establish arrival sequences.Furthermore, ATCos can reduce task demands indownstream sectors by carrying out tasks that


would otherwise have to be done in that sector(giving descent clearances, implementing speedcontrol, etc.). As has been found by Kopardekarand Magyarits (2003) and Masalonis et al. (2003),a workload model that does not take ATCo activ-ity into account may not be able to predict work-load accurately in the near future, such as 1 hrahead.

A third reason for concern with task demand-driven models of mental workload is that suchmodels are intended to predict whether workloadwill interfere with the performance or effective-ness of ATCos. The term performance refers to theability of ATCos to carry out their tasks (maintainsituation awareness, resolve conflicts, managedeparture flows, etc.), whereas the term effective-ness refers to the outcomes that the ATCo achieves(i.e., the safety, orderliness, and efficiency of traf-fic flows: Neal, Griffin, Neale, Bamford, & Boag,1998). Performance and effectiveness do not al-ways decline as workload increases. Observationsof ATCos in an operational environment suggestthat whereas their ability to maintain an orderlyand efficient flow of traffic does decrease asworkload increases, their ability to perform taskssuch as detecting and resolving conflicts and man-aging departures flows does not decline (Griffin,Neal, & Neale, 2000). It appears that the rela-tionships among workload, performance, andeffectiveness are complex and, possibly, contex-tually specific.

Human performance models provide a way ofaddressing these important issues. By building ahuman performance model that simulates how theATCo carries out control tasks, it may be possibleto generate more accurate predictions of aircrafttrajectories and, hence, of future task demands.Furthermore, by taking into account strategies thatATCos use to minimize the amount of control ac-tivity required to meet their objectives, one canmore accurately predict the effects of demands onboth mental workload and performance.

OPERATOR CAPACITY, STRATEGIES,AND MENTAL WORKLOAD

As noted previously, and as Figure 3 suggests,mental workload emerges not only from task de-mands but also from how control activity is as-sembled to meet task demands. Understandingworkload therefore involves understanding thestrategies that ATCos use to meet task demands.

Research in this area focuses on identifying cog-nitive tasks, eliciting controller strategies, and at-tempting to build computational models of ATCoactivity. In the next section we provide a briefreview of human performance models in ATC andnote how ATCos manage time pressure. We focuson three main control tasks identified in cogni-tive task analyses: the higher level control task ofmaintaining situation awareness and the controlsubtasks of detecting conflicts and resolving con-flicts (Kallus, Van Damme, & Dittman, 1999; Nealet al., 1998; Rodgers & Drechsler, 1993). Researchthat has examined ATCo strategies is summa-rized in Table 2.

Human Performance Models

Several models of ATCo performance havebeen developed over the past decade (Callantine,2002; Kallus, Van Damme, & Barbarino, 1999;Leiden et al., 2003; Niessen et al., 1999). In gen-eral, these models identify the control tasks thatATCos perform, the order in which they carrythem out, and the time required and time availableto do so. Such models offer insight into sources ofmental workload.

Some human performance models are fairlyhigh level, providing a verbal description of be-havior (Kallus, Van Damme, & Barbarino, 1999).Others are based on formal architectures such asAdaptive Control of Thought-Rational (Anderson,1993) and have been validated with empirical data(Leiden et al., 2003; Niessen et al., 1999). For ex-ample, the human performance model developedby Leiden et al. (2003) specifically focuses onarrivals streams. Leiden et al. (2003) modeledmental workload using the concepts of “task uti-lization” and “idle utilization.” They identifiedATCo tasks, obtained estimates for how long eachtask took to perform, and predicted how muchtime sets of tasks would take under different lev-els of traffic load. Task utilization, therefore, re-flects the time required to accept aircraft, resolveconflicts, provide metering, issue descent clear-ances, handoff aircraft, and transfer communica-tions. Idle utilization is all remaining time. Leidenet al. (2003) used idle utilization as an indicatorof available capacity. To our knowledge, howev-er, no studies have directly compared the validityof workload predictions generated by such humanperformance models with that of predictions gen-erated by dynamic density metrics. A further dif-ficulty with the Leiden et al. (2003) approach is


389

TAB

LE 2

:Sum

mar

y o

f E

mp

irica

l Stu

die

s E

xam

inin

g A

TCo

Str

ateg

y fo

r C

ont

rol T

asks

Aut

hors

Con

trol

Tas

ksM

etho

dSu

mm

ary

Resu

lts

Am

aldi

& L

erou

x (1

995)

Mai

ntai

n SA

Inte

rvie

wG

roup

A/C

into

str

eam

s of

tra

ffic,

sel

ectiv

ely

extr

act

A/C

dat

a fo

r co

nflic

t de

tect

ion

(alti

tude

firs

t);

Con

flict

det

ectio

nat

tend

to

criti

cal p

oint

s w

here

con

flict

s ha

ve p

revi

ousl

y oc

curr

ed; d

ecis

ion

to in

terv

ene

or n

ot, a

nd

Con

flict

reso

lutio

ntim

ing,

dep

ends

on

judg

men

t of

con

flict

ris

k.

Bis

sere

t (1

971)

Mai

ntai

n SA

Inte

rvie

wRe

gula

te a

tten

tion

allo

cate

d to

indi

vidu

al A

/C (b

ased

on

confl

ict

stat

us);

sele

ctiv

ely

extr

act

A/C

C

onfli

ct d

etec

tion

Expe

rimen

tda

ta fo

r co

nflic

t de

tect

ion

(alti

tude

firs

t).

Bou

des

et a

l. (1

997)

Con

flict

det

ectio

nIn

terv

iew

Sele

ctiv

ely

extr

act

airc

raft

A/C

dat

a fo

r co

nflic

t de

tect

ion.

Gro

nlun

d et

al.

(199

8)M

aint

ain

SAEx

perim

ent

Sele

ctiv

ely

atte

nd t

o A

/C (b

ased

on

spat

ial p

ositi

on);

sele

ctiv

ely

extr

act

A/C

dat

a fo

r co

nflic

t C

onfli

ct d

etec

tion

dete

ctio

n (a

ltitu

de fi

rst)

.

His

ton

& H

ansm

an (2

002)

Mai

ntai

n SA

Inte

rvie

wC

lass

ify A

/C in

to s

tand

ard

and

nons

tand

ard

flow

s; s

elec

tivel

y ex

trac

t A

/C d

ata

for

confl

ict

dete

ctio

n C

onfli

ct d

etec

tion

Expe

rimen

t(a

ltitu

de fi

rst)

; att

end

criti

cal p

oint

s w

here

con

flict

s ha

ve p

revi

ousl

y oc

curr

ed.

Con

flict

reso

lutio

n

Kal

lus,

Van

Dam

me,

Con

flict

det

ectio

nIn

terv

iew

A

tten

d cr

itica

l poi

nts

whe

re c

onfli

cts

have

pre

viou

sly

occu

rred

; ref

er t

o pr

evio

usly

use

d co

nflic

t &

Ditt

man

(199

9)C

onfli

ct re

solu

tion

reso

lutio

n st

rate

gies

; und

er h

igh

wor

kloa

d re

solv

e co

nflic

ts im

med

iate

ly (m

ake

safe

ty fi

rst

prio

rity)

.

Kirw

an &

Fly

nn (2

002)

Con

flict

reso

lutio

nIn

terv

iew

ATC

o us

es h

euris

tics

for

reso

lvin

g co

nflic

ts t

hat

min

imiz

e co

ntro

l act

ivity

.

Lepl

at &

Bis

sere

t (1

966)

Con

flict

det

ectio

nEx

perim

ent

Sele

ctiv

ely

extr

act

A/C

dat

a fo

r co

nflic

t de

tect

ion

(alti

tude

firs

t).

Mea

ns e

t al

. (19

88)

Mai

ntai

n SA

Expe

rimen

tRe

gula

te a

tten

tion

allo

cate

d to

indi

vidu

al A

/C (b

ased

on

amou

nt o

f con

trol

pre

viou

sly

exer

cise

d on

A/C

).

Nea

l et

al. (

1998

)C

onfli

ct d

etec

tion

Inte

rvie

w

Att

end

criti

cal p

oint

s w

here

con

flict

s ha

ve p

revi

ousl

y oc

curr

ed; r

efer

to

prev

ious

ly u

sed

confl

ict

Con

flict

reso

lutio

nre

solu

tion

stra

tegi

es.

Paw

lak

et a

l. (1

996)

Mai

ntai

n SA

Inte

rvie

wG

roup

A/C

into

str

eam

s of

tra

ffic.

Redd

ing

et a

l. (1

991)

Mai

ntai

n SA

Inte

rvie

wG

roup

A/C

into

str

eam

s of

tra

ffic.

Rant

anen

& N

unes

(200

5)C

onfli

ct d

etec

tion

Expe

rimen

tSe

lect

ivel

y ex

trac

t A

/C d

ata

for

confl

ict

dete

ctio

n (a

ltitu

de fi

rst,

hea

ding

sec

ond,

spe

ed t

hird

).

Rosk

e-H

ofst

rand

&M

aint

ain

SAIn

terv

iew

Gro

up A

/C in

to s

trea

ms

of t

raffi

c; s

elec

tivel

y ex

trac

t A

/C d

ata

for

confl

ict

dete

ctio

n (a

ltitu

de fi

rst)

.M

urph

y (1

998)

Con

flict

det

ectio

n

Seam

ster

et

al. (

1993

)C

onfli

ct d

etec

tion

Inte

rvie

w

Cla

ssify

A/C

into

sta

ndar

d an

d no

nsta

ndar

d flo

ws;

att

end

criti

cal p

oint

s w

here

con

flict

s ha

ve

Con

flict

reso

lutio

npr

evio

usly

occ

urre

d; re

fer

to p

revi

ousl

y us

ed c

onfli

ct s

trat

egie

s.

Sper

andi

o (1

971)

Mai

ntai

n SA

Expe

rimen

tRe

gula

te a

tten

tion

allo

cate

d to

indi

vidu

al A

/C; g

roup

A/C

into

str

eam

s of

tra

ffic.

Sper

andi

o (1

978)

Mai

ntai

n SA

Expe

rimen

tRe

gula

te a

tten

tion

allo

cate

d to

indi

vidu

al A

/C; g

roup

A/C

into

str

eam

s of

tra

ffic.

Wei

tzm

an (1

993)

Con

flict

reso

lutio

nIn

terv

iew

Cho

ice

of c

onfli

ct d

etec

tion

stra

tegy

dep

ends

on

tem

pora

l pro

xim

ity o

f the

con

flict

, the

geo

met

ryof

the

con

flict

, and

the

cer

tain

ty o

f the

con

flict

.

Will

ems

et a

l. (1

999)

Con

flict

det

ectio

nEx

perim

ent

Regu

late

att

entio

n al

loca

ted

to in

divi

dual

A/C

; sel

ectiv

ely

extr

act

A/C

dat

a fo

r co

nflic

t de

tect

ion

(alti

tude

firs

t).

No

te.A

/C =

airc

raft

; SA

= s

itua

tio

nal a

war

enes

s.


that it does not estimate the time that ATCos spendbuilding and maintaining situation awareness.Such cognitive activity contributes to workload,but it is unobservable.

Using a human performance model similar tothat of Leiden et al. (2003), Callantine (2002)emulated ATCo cognitive activity with a simpleset of heuristics for planning and decision mak-ing. At the beginning of a processing cycle, theCallantine (2002) model scans the environmentto detect tasks that need to be carried out. This isreferred to as “maintain situation awareness.”The model then selects the task with the highestpriority and carries it out. Possible tasks includecarrying out a plan that has been developed previ-ously, developing a plan to resolve a conflict, andissuing a descent clearance. Aset of rules describeshow each of these tasks is carried out. A prelim-inary evaluation of the Callantine (2002) modeldemonstrated that it could handle relatively sim-ple spacing problems in high-altitude sectors butperformed less well when handling merging traf-fic in low-altitude sectors. Nonetheless, the resultsare encouraging because they show that a simplemodel can control traffic in a plausible manner.The next step is to operationalize the concept ofmental workload within this type of model, butthat step has yet to be taken.

Time Pressure

Cognitive task analyses demonstrate thatATCos are required to complete many tasks, manyof which must be time-shared (e.g., Cox, 1994;Rodgers & Drechsler, 1993). According to aninformation-processing model developed byHendy et al. (1997), subjective estimates of men-tal workload are driven by the ratio of (a) the timeneeded to process the information necessary tomake a decision to (b) the time available before thedecision has to be put into action. The most impor-tant human constraint, then, is the maximum rateat which work can be done. The competent ATCogenerally knows the rate at which he or she cancomplete tasks, and this knowledge is activelymanaged by the ATCo to avoid overload. ATCosmaintain acceptable levels of workload underheavy task demand by seeking control strategiesthat minimize the amount of control activity (e.g.,planning, monitoring, coordinating) required tomeet their objectives and, if necessary, reorderingwork priorities. According to Hendy et al.’s (1997)model, these changes help to reduce the amount

of information that has to be processed. In the nextsections we examine how ATCos manage themental workload associated with maintaining sit-uational awareness.

Maintaining Situational Awareness

ATCos work to maintain a valid mental repre-sentation of the current air traffic situation, whichis commonly referred to as situational awareness(SA; Endsley, 1995; Endsley & Smolensky,1998). As Dailey (1984) stated, “The central skillof the controller seems to be the ability to respondto a variety of quantitative inputs about severalaircraft simultaneously and to form a continuous-ly changing mental picture to be used as the basisfor planning and controlling the courses of theaircraft” (p. 134).

SAis usually understood to involve (a) the con-tinuous perception of information in the environ-ment, (b) the integration of this information withprior knowledge to form a coherent understandingor “mental picture” of the current situation, and (c)the use of this mental picture to direct visual search,guide perception, anticipate the future state of airtraffic, and plan required actions (Endsley, 1995).

SAis maintained through monitoring, which isthe continuous or intermittent comparison of ananticipated versus an actual traffic situation. Mon-itoring involves directing attention to externalsources of information (e.g., sector maps, a radarscreen, or flight plans) in order to determine if tra-jectories of future aircraft movement and positionsare consistent with the mental picture. As long asthe mental picture remains consistent with actualevents, SAis maintained. Research indicates thatmany operational errors can be attributed to SAproblems (e.g., Jones & Endsley, 1996) and, con-versely, that scores on measures of SAcan predictperformance (e.g., Durso, Hackworth, Truitt,Crutchfield, Nikolic, & Manning, 1998). Underheavy task demand ATCos can be so busy dealingwith traffic events that they do not have time toupdate their mental picture, cannot plan ahead, andare forced to work reactively. ATCos refer to thisas “losing the picture.” However, as will be dis-cussed shortly, ATCos have strategies that helpavoid this situation.

To assess ATCo SA, query techniques arecommonly used that tap ATCos’ ability to recallinformation about the air traffic situation (Adams,Tenney, & Pew,1995). Findings suggest that ATCoscan reduce the mental workload associated with


monitoring by regulating the amount of attentionthey give to individual aircraft (Bisseret, 1971;Gronlund et al., 1998; Sperandio, 1971). Earlystudies indicated that ATCos could recall moreabout the positions and flight data of aircraft onwhich they had performed control actions (Meanset al., 1988) or of aircraft that were in potentialconflict (Bisseret, 1971; Sperandio, 1971), ascompared with aircraft on which they had notperformed control actions or which were not inpotential conflict. More recently, Gronlund et al.(1998) found that ATCos appeared to classify air-craft into two categories – important versus non-important – on the basis of how soon they wouldlose separation with other aircraft. Althoughthere was no difference in ATCos’ability to recallthe two-dimensional position or heading of im-portant versus nonimportant aircraft, ATCos weremore likely to recall the altitude and ground speedof important aircraft. Gronlund et al. (1998) con-cluded that ATCos assigned importance to aircrafton the basis of relative spatial position to otheraircraft, and that information not presented spa-tially (e.g., altitude, speed) was selectively attend-ed on the basis of this importance weighting.

As noted previously, the underlying structureof the airspace can become the basis for abstrac-tions that simplify the ATCo’s cognitive work(Histon & Hansman, 2002; Seamster, Redding,Cannon, Ryder, & Purcell, 1993). Field observa-tions conducted by Histon and Hansman (2002)indicated that standard flows are one of the mostimportant structure-based abstractions. ATCosclassify aircraft into standard and nonstandardclasses according to their match with standardflow, which would include the aircraft’s futurerouting, ingress and engress points, coordinationrequirements, and crossing routes/altitude pro-files (Histon & Hansman, 2002; Seamster et al.,1993). These factors allow the ATCo to form ageneral expectation of how aircraft will movethrough the sector, significantly reducing thecomplexity of control. Aircraft classified as non-standard increase the complexity of control be-cause their trajectory and interactions with otheraircraft are more difficult to predict a priori.

In addition, structured interviews reveal thatATCos process aircraft in groups to reduce theinformation-processing requirements associatedwith monitoring air traffic (Amaldi & Leroux,1995; Histon & Hansman, 2002; Redding, Ryder,Seamster, Purcell, & Cannon, 1991). For exam-

ple, if four aircraft are heading southbound and sixaircraft northbound, the ATCo might process andmonitor the four southbound aircraft as one streamand the six northbound aircraft as another stream,rather than monitor each individual aircraft.ATCos report that such strategies let them focuson the intersection of the two streams, rather thanrequiring them to assess the conflict status of eachpossible aircraft pair (Pawlak et al., 1996). Aircraftseparation within streams is maintained by speedcontrol (Sperandio, 1978), and separation betweenstreams at common intersection points is main-tained by altitude control. Overall, establishingstreams simplifies the process of maintaining SA,letting the ATCo work with more aircraft simul-taneously and use fewer control actions.

Conflict Detection

One aspect of SA that is particularly critical isconflict detection. Conflict detection research hastended to focus on (a) factors that increase thecomplexity of detecting conflicts or (b) strategiesthat ATCos use to minimize this complexity. Theconnection with mental workload is seldom ex-plicitly addressed, although increases in the timetaken to detect conflicts have been associated withgreater mental workload ratings (e.g., Galster,Duley, Masalonis, & Parasuraman, 2001; Metzger& Parasuraman, 2001). In an operational context,a delay in conflict detection imposes constraintson the ATCo because it decreases the time avail-able to intervene and ensure separation. Areducedtime to implement a conflict resolution plan canforce an ATCo into a situation in which he or shehas to create a disorderly flow of traffic, which hasthe potential to cause problems in the future (Cox,1994; Kallus, Van Damme, & Dittman, 1999). Incontextual control terms (e.g., Hollnagel, 2002),this could represent one aspect of the shift fromtactical to a reactive control.

Several studies have examined how taskdemand affects the accuracy and timeliness ofconflict detection (Boag et al., 2006; Galster et al.,2001; Leplat & Bisseret, 1966; Metzger & Para-suraman, 2001; Nunes & Scholl, 2004; Rantanen& Nunes, 2005; Remington, Johnston, Ruthruff,Gold, & Romera, 2000). For example, Remingtonet al. (2000) found that conflict detection accura-cy decreased, and conflict detection latency in-creased, with higher traffic count and increasingangles of convergence. Conflict detection latencyalso increased as time to conflict increased. Higher


traffic count presumably affects conflict detectionby increasing visual search requirements and re-ducing the time available to make conflict statusdecisions (Hendy et al. 1997). Boag et al. (2006)showed that the number of aircraft transitions inand out of conflict positively predicted detectiontime. Similarly, Leplat and Bisseret (1966) foundthat ATCos took longer to detect conflicts whenthree variables needed to be processed to deter-mine conflict status (e.g., altitude, heading, andspeed) than when only one variable needed to beprocessed.

Nevertheless, ATCos may change their strate-gy for conflict detection in response to these taskdemands and anticipated mental workload. For ex-ample, ATCos appear to prefer using altitude infor-mation to heading and speed information whendetermining the likelihood of aircraft conflict(Amaldi & Leroux, 1995; Leplat & Bisseret, 1966;Willems, Allen, & Stein, 1999). This finding issupported by cognitive task analyses (Kallus, VanDamme, & Dittman, 1999; Neal et al., 1998;Seamster et al., 1993), interviews with controllers(Amaldi & Leroux, 1995; Boudes, Amaldi, &Cellier,1997), anecdotal reports (Roske-Hofstrand& Murphy, 1998; Wickens, Mavor, & McGee,1997), and experiments (Bisseret, 1971; Gronlundet al., 1998; Rantanen & Nunes, 2005). Takentogether, this body of research suggests that ATCossave attentional resources by extrapolating aircrafttrajectories for lateral separation only in circum-stances where vertical separation is questionable.Furthermore, ATCos prefer to use altitude andheading information over speed information (Bis-seret, 1971; Leplat & Bisseret, 1966; Rantanen& Nunes, 2005; Willems et al., 1999). Presumably,this is attributable to the greater cognitive effortassociated with processing speed information(see Law et al., 1993).

Afurther strategy that ATCos use to detect con-flicts, already touched upon, is to exploit airspacestructure and standard flows (Amaldi & Leroux,1995; Histon & Hansman, 2002; Kallus, VanDamme,&Dittman,1999; Neal et al.,1998; Roske-Hofstrand & Murphy, 1998; Seamster et al., 1993).Specific aircraft events occur routinely at specificlocations in a sector, and ATCos learn to recognizespecific air traffic configurations. Memory for pastexperiences lets ATCos anticipate where conflictsmay occur (sector “hot spots” or “critical points”),simplifying the cognitive work of detecting con-flicts. By focusing on a finite number of critical

crossing points, ATCos do not need to evaluatethe likelihood of conflict between all aircraft pairsin the sector, thus reducing task demands and,presumably, mental workload.

Conflict Resolution

The ATC literature has provided some usefulinsights into the strategies that ATCos use toresolve conflicts under different levels of taskdemand and thus regulate their mental workload.On the basis of detailed interviews with ATCos,several research groups have argued that theworkload associated with resolving conflicts de-pends largely on sector-specific knowledge thatATCos acquire over time (Kallus, Van Damme,& Dittman, 1999; Neal et al., 1998; Seamster etal., 1993). ATCos often report that they have theequivalent of a “conflict resolution library” ofsolutions to particular configurations of air traffic(Kallus, Van Damme, & Dittman, 1999). When aconflict is detected, ATCos “access” their libraryto find a previous solution and then adapt andapply that previous solution to the new case. If noappropriate solution is found, ATCos must eitherreview their solution library for a suitable plan oruse problem-solving techniques to develop a newsolution. Retrieving a solution from memory re-duces cognitive work and the associated workload.

ATCos also regulate mental workload by beingselective about when they intervene to ensureseparation. ATCos interviewed by Kallus, VanDamme, and Dittman (1999) reported that underlow workload they tend not to solve conflicts im-mediately because it can reduce the efficiency ofaircraft movement. They prefer to monitor the sit-uation. Under high-workload conditions, howev-er, they are reluctant to let potential conflicts rununless intervening tasks leave them with enoughcapacity to monitor the potential conflict. Underhigh workload they tend to solve problems by tak-ing immediate action in order to conserve atten-tional resources. This strategy also reduces thelikelihood that they will forget to return to the un-resolved situation because of distraction fromcompeting tasks (Loft, Humphreys, & Neal, 2003).

Interviews with ATCos conducted by Amaldiand Leroux (1995; also see Weitzman, 1993) indi-cate that a further factor determining when ATCosintervene is their judgment of the probability thatthe aircraft pair will violate separation. If they re-spond to every possible conflict they are proactivebut overloaded, whereas if they respond only to


definite conflicts they are reactive and may losethe picture. Under high mental workload condi-tions ATCos shift their criterion for classifyingconflicts, becoming more conservative so thatthey intervene to ensure separation if there is anyuncertainty regarding future separation betweenaircraft. Although such a change in conflict detec-tion criterion may temporarily increase ATCocontrol activity, it will significantly reduce theamount of control activity required in the longerterm. Moreover, when intervening to ensure sep-aration under conditions of high workload, ATCoschoose solutions that require minimum monitor-ing and coordination. In this way, ATCos maintainresilient control. For example, Kirwan and Flynn(2002) identified various heuristics ATCos use toresolve conflicts, such as (a) using as few controlactions as possible, (b) giving aircraft initial levelchanges early and fine-tuning later, (c) using so-lutions that require less coordination, (d) usingvertical separation for complex conflicts, and (e)keeping solutions simple and safe.

Summary and Assessment

The key characteristic of the ATCo mentalworkload model presented in Figure 3 is that theATCos do not passively react to events but, in-stead, actively control workload by selecting stra-tegies that have different demands on cognitiveresources. Operator capacity is therefore not a sta-tic property of the ATCo but a dynamic one. In thissection we reviewed efforts to identify the con-trol tasks that ATCos perform and the strategiesATCos use to minimize the control activity asso-ciated with these control tasks, in order to under-stand workload.

Human performance models have been dev-eloped that integrate ATCo control tasks intocomputational frameworks, but such models gen-erally do not include sophisticated sets of strate-gies, model mental workload, or demonstrate howmental workload might drive the selection ofstrategies. Research on time pressure suggests thatone way workload might be modeled is as a func-tion of the ratio of the time to perform a task tothe time available before the task must be com-pleted. Task timing information for specific con-trol tasks could be estimated via empirical data(e.g., Cardosi, 1993), interviews (e.g., Amaldi &Leroux,1995), cognitive task analyses (e.g., Kallus,Van Damme, & Dittman, 1999) or observation(e.g., Histon & Hansman, 2002). If mappings be-

tween workload – however it is measured – andselection of strategies could be operationalizedwithin human performance models, a better under-standing could probably be gained of the adaptivenature of ATCo work and of workload itself.

Research on three principal ATCo activities –maintaining SA, detecting conflicts, and resolvingconflicts – provides abundant evidence that ATCosseek ways of minimizing mental workload. Theworkload of maintaining SA can be reduced byusing standard flows and by grouping aircraft sothat many aircraft can be handled in one operation.Moreover, by focusing on altitude information andseeking resolution on that basis before consideringheading and speed, ATCos seek to reduce work-load. The workload associated with the subtask ofconflict detection is also amenable to strategic con-trol. For example, sector structure produces loca-tions where conflicts are more likely or less likely.The workload of resolving conflicts can be reducedthrough development and use of a repertoire ofsolutions mapped to the imminence, probability,and geometry of potential conflicts, as well as thetiming of interventions. Overall, ATCos attend tocues that give them the smallest amount of infor-mation necessary for effective decision makingaccording to the priorities chosen for perfor-mance.

All the these factors can in principle be mod-eled computationally, leading to a better under-standing of the relation among task demands,ATCo activity, and ATCo mental workload, butoperationalization can be difficult. In the next sec-tion we review models that reflect such modeling.

MODELS INTEGRATING TASK DEMANDSWITH OPERATOR ACTIVITY

This review has focused on two broad deter-minants of mental workload: task demands (theamount and complexity of work) and operatorcapacity (the resources the ATCo can marshal tomeet demand, including strategies). There havebeen some relatively recent efforts by researchersto develop techniques that integrate task demandwith operator capacity in order to predict workload(Averty, Athènes, Collet, & Dittmar, 2002; Avertyet al., 2004; Chaboud, Hunter, Hustache, Mahlich,& Tullett, 2000; Cullen, 1999; Stamp, 1992).

For example, Chaboud et al. (2000) developeda model that describes the mental workload cor-responding to different control tasks. First, the


model describes workload for routine tasks, suchas flight data management, coordination, and radiocommunications. The workload value is based onthe number of aircraft sector entries and is mul-tiplied by the estimated time it would take ATCosto complete the task (task duration). Second, avalue is assigned for the workload associated withclimbing and descending aircraft, defined as thenumber of aircraft with a 6,000-foot vertical evo-lution multiplied by task duration. Third, a valueis assigned based on the workload associatedwith monitoring conflicts, defined as the numberof conflicts multiplied by task duration. This work-load metric correlated well with ATCo activitymeasures but was not validated with an indepen-dent workload measure. Two significant limita-tions of the work of Chaboud et al. (2000) are that(a) the workload weights assigned to different con-trol tasks were fixed and (b) the durations assignedto tasks were fixed. These methods do not takeinto account the fact that the workload associatedwith different control tasks can be modulated bythe use of strategy.

Averty et al. (2002, 2004) developed a metriccalled the traffic load index (TLI), which takes intoaccount the fact that, through their actions, ATCoswill regulate their own mental workload. The TLIis calculated by assigning each aircraft underjurisdiction a weight that contributes toward theTLI for the sector. Aircraft under jurisdiction ofthe ATCo are assigned a base weight of 1. Any air-craft that need additional monitoring – for exam-ple, because they are anticipated to be involved ina conflict or will need vectoring on their descentprofile into an airport – accrue additional weight-ing. The weighting is thus a measure of the amountof monitoring each aircraft requires. This firstpart of the TLI provides a simple measure of taskdemand. The second part of the TLI assesses theeffects of ATCo activity. If the ATCo sees a poten-tial conflict but waits a long time before resolvingit, then the so-called maturing time (MT) beforeacting is protracted and workload associated withmonitoring is increased. In contrast, if the ATCoresolves the potential conflict immediately, MT isshort and monitoring workload is removed. Evenif action is taken immediately, in some cases sep-aration may not be ensured for quite a long time,leading to a long MT because of the remaininguncertainty. Acting earlier removes time pressureto act but leaves some uncertainty as to whethera conflict will be resolved as desired. In contrast,

acting later imposes time pressure when one doescome to act, but it provides greater certainty as towhether the conflict will be resolved as desired.

Averty et al. (2002, 2004) have shown that thecorrelation between the NASA Task Load Index(NASA-TLX; Hart & Staveland, 1988) and TLIwas significantly higher than that between thetraffic count and NASA-TLX. This indicates thatTLI is a better predictor of ratings of mental work-load than is traffic count. In addition, the TLI hadhigher correlations with all physiological work-load measures than did both traffic count and theNASA-TLX. By modeling the timing of ATCos’interventions, the TLI measure goes some way incapturing the effect of different ATCo strategieson workload.

In a further effort, Cullen (1999) built a mentalworkload model in which she attempted to quan-tify the sequences and durations of ATCo tasks.Factors influencing the sequence and durations oftasks included (a) environmental conditions thatinitiated each task, (b) priorities assigned to tasks(urgent, high, low), (c) rules specifying whichtasks could be interrupted by tasks of higher pri-ority, and (d) rules governing task selection andsequence. The task model soundly predicted taskdurations. However, the model’s ability to predicttask sequence and workload was poor. In gener-al, observed workload was higher than predictedworkload. Cullen (1999) concluded that work-load was poorly predicted because the task modelcould not accurately predict sequences of ATCoactivity. Acloser inspection of the model indicatesthat it did not take into account the effects ofworkload management strategies on the sequencesand durations of task behavior. In addition, themodel did not account for the fact that multipletasks are often completed concurrently or thatATCos often monitor the progress of aircraft afterissuing instructions to ensure pilot complianceand adequate separation. Overall, the predictivevalidity of Cullen’s (1999) model was disappoint-ing. Notably, the factors missing from the modelincluded factors modeled in Figures 2 and 3 anddiscussed in preceding sections of this review.

A further and very important model that inte-grates task demands with operator activity is theMan-Machine Interaction Design and AnalysisSystem (MIDAS; Corker & Smith, 1993). MIDASis not a mental workload model per se but a hu-man performance model used for the analysis ofhuman-machine systems design issues. MIDAS


is a first-principles model of human performance,built upon 35 primitive operator tasks such as vi-sual monitoring, typing, grasping, and comput-ing. Each of these tasks is assigned a workloadweight on each of the six channels from multipleresource theory (Wickens, 1984): visual and audi-tory input, spatial and verbal cognitive process-ing, and manual and voice output. Additionally,MIDAS has a dynamic mechanism for switchingbetween control strategies that is based on Holl-nagel’s (2002) COCOM model. The model switch-es among strategic, tactical, opportunistic, andscrambled control depending on four control para-meters: the event horizon (a measure of success ofprevious control actions taken by the operator), anestimate of the time available for the control activ-ity, an estimate of the time required to completethe control activity, and an estimate of competen-cy (measured in the number of goals). The specif-ic values of these control variables that will makeMIDAS switch between different control strategiesare domain dependent. For the ATCo model in AirMIDAS, the aviation-specific version of MIDAS,the event horizon is measured in number of aircraftunder control and the complexity of maneuversthe aircraft have to perform (Corker, 2003). Eval-uations of Air MIDAS have been performed, butdetailed reports are still forthcoming.

SUMMARY AND CONCLUSION

The most prevalent approach to studying themental workload of ATCos is to investigate trafficfactors that produce or influence task demand, onthe assumption that there is a relationship betweentask demand and mental workload that is mediat-ed by control strategy (see Figure 1). In this paperwe reviewed this and other modeling architecturesthat have been used to understand workload inATC. Influenced by Sperandio’s (1971) emphasison the central role of alternative work methods,or strategies, in controlling workload (Figure 2),we developed a model expanding on the ATCo’srole in selecting an appropriate strategy for meet-ing task demands (Figure 3). The model showsstrategy being driven on a proactive/feedforwardbasis as well as on a reactive/feedback basis. Themodel also shows strategy being driven not onlyby task demands but also by judgments aboutwork priorities that is, by a hierarchy of standardsthat ATCos aim to preserve.

In the body of this review, we organized thevast literature on ATCo performance around two

major themes: research investigating the relation-ship between task demands and mental workload,and research investigating the relationship be-tween ATCo capacity and mental workload. Muchof the latter work emphasizes the crucial role ofATCo strategies in handling task demand. Towardthe end of the review we examined models inwhich researchers have attempted to integratethese two themes. The overall impression is thatthere is still a long way to go before ATCo work-load is fully understood, let alone modeled com-putationally to a level that would support robustorganizational decision making about allocationof ATCo work through sector sizes, rostering, and so on.

Much research still seeks relationships betweentask demands and ATCo mental workload in theopen-loop manner illustrated in Figure 1. Howev-er, simply “integrating” task demand and opera-tor capacity in closed-loop models is unlikely tohelp in modeling workload. Information is alsoneeded about strategies, performance priorities,and an appropriate architecture to link all theseelements. Figure 3 represents an attempt to do so.This approach may be more useful in understand-ing ATCo workload than attempting to model therelationship between isolated traffic factors andATCo speed or accuracy of responding. However,an even more appropriate architecture may be onethat models – far more explicitly even than that inFigure 3 – the fact that ATCos regulate their work-load by selecting control strategies that meet taskdemands, driven by the relative priority of theirobjectives. Figure 4 shows the simplest form ofsuch a workload regulation model. In Figure 4, adeviation from the desired level of workload leadsto an adjustment of control strategy, and the con-trol strategy shapes the task demands that produceactual workload. Actual workload is comparedwith the desired level of workload, and the looprepeats.

One strength of the mental workload modelspresented in Figures 3 and 4 is that they use relatively domain-independent system-level para-meters (e.g., task demand, metacognition, prior-itization, strategy). Consequently, as outlined inthe introduction to this paper, operational modelsof workload could be developed for a variety ofcomplex work systems (e.g., piloting, unmannedaerial vehicle control, anesthesiology, railwaysignaling, and automobile driving) using the con-ceptual models presented in Figures 3 and 4.


Moreover, it suggests quite a different way ofinvestigating workload in complex work systems.First, instead of investigating the linear relation-ship between task demand and workload at spe-cific moments in time, one might investigatedynamic properties of workload that could showworkload to lead or lag events (Rouse et al., 1993).Second, one might investigate workload signals,rather than task demand signals, that lead to aswitch in strategy. Third, one might investigatehow operators learn signals relevant for manag-ing workload and developing essential skills (e.g.,estimating time needed to execute plans, estimat-ing when events will occur), and examine therelationship between the two. Fourth, one mightinvestigate far more thoroughly than before howstrategies create specific patterns of task demands.Fifth, one might study the impact of operators’persistence with inappropriate strategies on bothworkload and control quality and how any suchproblems might be alleviated. This list is notexhaustive – there are many further possible im-plications of taking such an approach to concep-tualizing both ATCo workload and workload inother complex work systems.

Finally, taking a mental workload-centeredview rather than a task demand-centered viewmay have longer term benefits. Globally, this is anera of rapid changes in air traffic management(EUROCONTROL, 1999; Federal Aviation Ad-ministration, 2005). For example, “free flight”refers to a wide variety of ATC regimens that allshare the following characteristics: an increase inairspace capacity through new decision supporttools, increased automation to aid the ATCo, andincreased flexibility for airline and aircraft oper-ations. Similarly, in modern aircraft, pilots areprovided with host of automated flight control sys-tems that automate tasks such as status monitoringand flight mode (e.g., climb, cruise descend;

Kantowitz, 1994; Wickens, 2002). Aviation pro-viders and regulatory bodies need tools that arecapable of assessing the workload experienced byoperators under these proposed systems. Modelsthat focus primarily on the link between task de-mand and workload are likely to have difficultygeneralizing to new automation systems becauseoperator control strategies will change. More fullydeveloped models that show how control strate-gies regulate task demand and thereby workloadwould be extremely useful because they would letresearchers explore the consequences of new con-trol arrangements. Good progress has been madein describing the empirical relationship betweentask demand and workload in many complex worksystems. The next step is to develop and test dy-namic models that explain the relationship amongworkload, task demands, and strategy-driven ac-tivity within current and future systems.

ACKNOWLEDGMENTS

This research was supported by Linkage GrantLP0453978 from the Australian Research Counciland Airservices Australia and a postdoctoral re-search fellowship awarded to Shayne Loft fromthe Australian Research Council.

REFERENCES

Adams, M. J., Tenney, Y. J., & Pew, R. W. (1995). Situational aware-ness and the cognitive management of complex systems. HumanFactors, 37, 85–104.

Amaldi, P., & Leroux, M. (1995). Selecting relevant information in acomplex environment: The case of air traffic control. In L. Norros(Ed.), 5th European Conference on Cognitive Science Approachesin Process Control (pp. 89–98). Espoo, Finland: VTTAutomation.

Anderson, J. R. (1993). Rules of the mind. Hillsdale, NJ: Erlbaum.Arad, B. A. (1964, May). The controller load and sector design. Journal

of Air Traffic Control, 12–31.Athènes, S., Averty, P., Puechmorel, S., Delahaye, D., & Collet, C.

(2002). Complexity and controller workload: Trying to bridge thegap. In Proceedings of the 2002 International Conference onHuman-Computer Interaction in Aeronautics (pp. 56–60).Cambridge: Massachusetts Institute of Technology.

Strategy Task demands

Desired workload

Actual workload

+-

Figure 4. Model of the regulation of mental workload by adapting strategy to task demands.


Averty, P., Athènes, S., Collet, C., & Dittmar, A. (2002, October).Evaluating a new index of mental workload in real ATC situationusing psychophysiological measures. Paper presented at the 21stDigital Avionics System Conference, Irvine, CA.

Averty, P., Collet, C., Dittmar, A., Vernet-Maury, E., & Athènes, S.(2004). Mental workload in air traffic control: An index constructedfrom field tests. Aviation, Space, and Environmental Medicine, 75,333–341.

Bisseret, A. (1971). An analysis of mental model processes involved inair traffic control. Ergonomics, 14, 565–570.

Boag, C., Neal, A., Loft, S., & Halford, G. (2006). An analysis of rela-tional complexity in an air traffic control conflict detection task.Ergonomics, 49, 1508–1526.

Boudes, N., Amaldi, P., & Cellier, J. M. (1997). Memory performanceand diagnostic processes in air traffic control. In S. Bagnara, E.Hollnagel, M. Mariani, & L. Norros (Eds.), Time and space inprocess control: Proceedings of the 6th European Conference onCognitive Science Approaches to Process Control (pp. 204–208).Espoo, Finland: VTT Automation.

Buckley, E. P., DeBaryshe, B. D., Hitchner, N., & Kohn, P. (1983).Methods and measurements in real time air traffic control systemsimulation (Rep. No. DOT/FAA/CT-83/26). Washington, DC: U.S.Department of Transportation, Federal Aviation Administration.

Callantine, T. J. (2002). CATS-based air traffic controller agents(Rep. No. NASA/CR-2002-211856). Moffett Field, CA: NASAAmes Research Center.

Cardosi, K. M. (1993). Time required for transmissions of time-criticalair traffic control messages in an en route environment. InternationalJournal of Aviation Psychology, 3, 303–313.

Chaboud, T., Hunter, R., Hustache, J. C., Mahlich, S., & Tullett, P.(2000). Investigating air traffic complexity: Potential impacts onworkload and costs (European Experimental Centre Note No.11/00). Brussels, Belgium: EUROCONTROL.

Chatterji, G. B., & Sridhar, B. (2001). Measures for air traffic controllerworkload prediction. In Proceedings of the 1st AIAA Aircraft, Tech-nology, Integration, and Operations Forum [CD-ROM]. Reston,VA: American Institute of Aeronautics and Astronautics.

Christien, R., Benkouar, A., Chaboud, T., & Loubieres, P. (2003, June).Air traffic complexity indicators and ATC sectors classification.Paper presented at the 5th USA/Europe Air Traffic Management R& D Seminar, Budapest, Hungary.

Corker, K. M. (2003). Acognitive framework for operation of advancedaerospace technologies. In E. Hollnagel (Ed.), Handbook of cog-nitive task design. (pp. 417–446). Mahwah, NJ: Erlbaum.

Corker, K. M., & Smith, B. R. (1993, September). An architecture andmodel for cognitive engineering simulation analysis: Applicationto advanced aviation automation. Paper presented at the AIAAComputing in Aerospace 9 Conference, San Diego, CA.

Couluris, G. J., & Schmidt, D. K. (1973). Air traffic control jurisdic-tions of responsibility and airspace structure. Paper presented atthe 1973 Institute of Electrical and Electronics Engineers (IEEE)Conference on Decision and Control held December 5–7 in SanDiego, CA.

Cox, M. (1994). Task analysis of selected operating positions withinUK air traffic control (IAM Rep. No. 769). Farnborough, UK:Royal Air Force Institute for Aviation Medicine.

Cullen, L. (1999). Validation of a methodology for predicting perfor-mance and workload (European Experimental Centre Note No.7/99). Brussels, Belgium: EUROCONTROL.

Dailey, J. T. (1984). Characteristics of the air traffic controller. In S. B.Sells, J. T. Dailey, & E. W. Pickerel (Eds.), Selection of air trafficcontrollers (No. FAA-AM-84-2; pp. 128–141). Washington, DC:Federal Aviation Administration Office of Aviation Medicine.

Davis, C. G., Danaher, J. W., & Fischl, M. A. (1963). The influence ofselected sector characteristics upon ARTCC controller activities(Contract No. FAA/BRD-301). Arlington, VA: Matrix.

Dixon, S. R., Wickens, C. D., & Chang, D. (2005). Mission control ofmultiple unmanned aerial vehicles: A workload analysis. HumanFactors, 47, 479–488.

Durso, F. T., Hackworth, C. A., Truitt, T. R., Crutchfield, J. M., Nikolic,D., & Manning, C. A. (1998). Situation awareness as a predictor ofperformance for en route air traffic controllers. Air Traffic ControlQuarterly, 6, 1–20.

Endsley, M. R. (1995). Toward a theory of situational awareness indynamic systems. Human Factors, 37, 32–64.

Endsley, M. R., & Smolensky, M. W. (1998). Situational awareness in airtraffic control: The picture. In M. W. Smolensky & E. S. Stein (Eds.),Human factors in air traffic control (pp. 115–154). San Diego, CA:Academic Press.

European Organisation for the Safety of Air Navigation (EUROCON-TROL). (1999). Operational concept document of the strategy2000+. Brussels, Belgium: Author.

Farmer, E., & Brownson, Q. (2003). Review of workload measurement,analysis and interpretation methods (European Air Traffic Man-agement Programme Rep. No. CARE-Integra-TRS-130-02-WP2).Brussels, Belgium: EUROCONTROL.

Federal Aviation Administration. (2005). Flight plan 2005–2009.Washington, DC: Author.

Galster, S. M., Duley, J. A., Masalonis, A. J., & Parasuraman, R. (2001).Air traffic controller performance and workload under mature freeflight: Conflict detection and resolution of aircraft self-separation.International Journal of Aviation Psychology, 11, 71–93.

Gopher, D., & Donchin, E. (1986). Workload: An examination of theconcept. In K. R. Boff, L. Kaufman, & J. P. Thomas (Eds.), Hand-book of perception and performance: Volume 2. Cognitive process-es and performance (pp. 41–49). New York: Wiley.

Griffin, M. A., Neal, A., & Neale, M. (2000). The contribution of taskperformance and contextual performance to effectiveness in air traf-fic control: Investigating the role of situational constraints. AppliedPsychology: An International Review, 49, 517–533.

Gronlund, S. D., Ohrt, D. D., Dougherty, M. R. P., Perry, J. L., &Manning, C. A. (1998). Role of memory in air traffic control. Journalof Experimental Psychology: Applied, 4, 263–280.

Grossberg, M. (1989). Relation of sector complexity to operational errors(Quarterly Rep. of the FAA Office of Air Traffic Evaluations andAnalysis). Washington, DC: Federal Aviation Administration.

Hart, S. G., & Staveland, L. E. (1988). Development of NASA-TLX(Task Load Index): Results of empirical and theoretical research. InP. A. Hancock & N. Meshkati (Eds.), Human mental workload: Ad-vances in psychology (Vol. 52, pp. 139–183). Oxford, UK: NorthHolland.

Hart, S. G., & Wickens, C. D. (1990). Workload assessment and predic-tion. In H. R. Booher (Ed.), MANPRINT: An emerging technology.Advanced concepts for integrating people, machine, and organiza-tions (pp. 257–296). New York: Van Nostrand Reinhold.

Hendy, K. C., Liao, J., & Milgram, P. (1997). Combining time and inten-sity effects in assessing operator information-processing load.Human Factors, 39, 30–47.

Hilburn, B. (2004). Cognitive complexity in air traffic control: A litera-ture review (EUROCONTROL Experimental Centre Note No.04/04). Brussels, Belgium: EUROCONTROL.

Hilburn, B., & Jorna, P. G. A. M. (2001). Workload and air traffic con-trol. In P. A. Hancock & P. A. Desmond (Eds.), Stress, workload, andfatigue: Human factors in transportation (pp. 384–394). Mahwah,NJ: Erlbaum.

Histon, J. M., & Hansman, R. J. (2002). The impact of structure on cog-nitive complexity in air traffic control (Rep. No. ICAT-2002-4).Cambridge, MA: MIT International Center for Air Transportation.

Hollnagel, E. (2002). Time and time again. Theoretical Issues inErgonomic Science, 3, 143–158.

Hollnagel, E., & Woods, D. D. (2005). Joint cognitive systems: Founda-tions of cognitive systems engineering. Boca Raton, FL: CRC Press.

Hopkin, V. D. (1995). Human factors in air traffic control. London:Taylor & Francis.

Hurst, M. W., & Rose, R. M. (1978). Objective job difficulty, behavioralresponse, and sector characteristics in air route traffic control centers.Ergonomics, 21, 697–708.

International Civil Aviation Organization. (2004). Outlook for airtransport to the year 2015 (Circular 304). Montreal, Canada:Author.

Jones, D. G., & Endsley, M. R. (1996). Sources of situational aware-ness errors in aviation. Aviation, Space, and EnvironmentalMedicine, 67, 507–512.

Kallus, K. W., Van Damme, D., & Barbarino, M. (1999). Model of thecognitive aspects of air traffic control (European Air TrafficManagement Programme Rep. No. HUM.ET1.ST01.1000-REP-04). Brussels, Belgium: EUROCONTROL.

Kallus, K. W., Van Damme, D., & Dittman, A. (1999). Integrated joband task analysis of air traffic controllers: Phase 2. Task analysis


of en-route controllers (European Air Traffic ManagementProgramme Rep. No. HUM.ET1.ST01.1000-REP-04). Brussels,Belgium: EUROCONTROL.

Kantowitz, B. H. (1994). Pilot workload and flight deck automation. InM. Mouloua & R. Parasuraman (Eds.), Human performance inautomated systems: Current research and trends (pp. 212–223).Hillsdale, NJ: Erlbaum.

Kirwan, B., & Flynn, M. (2002). Towards a controller-based conflictresolution tool – A literature review (European Air Traffic Manage-ment Programme Rep. No. ASA.01.CORA.2.DEL04-A.LIT).Brussels, Belgium: EUROCONTROL Experimental Centre.

Kirwan, B., Scaife, R., & Kennedy, R. (2001). Investigating complexityfactors in UK air traffic management. Human Factors and AerospaceSafety, 1, 125–144.

Kopardekar, P., & Magyarits, S. (2003, June). Measurement and pre-diction of dynamic density. Paper presented at the 5th USA/EuropeATM R & D Seminar, Budapest, Hungary.

Lamoureux, T. (1999). The influence of aircraft proximity data on thesubjective mental workload of controllers on the air traffic controltask. Ergonomics, 42, 1482–1491.

Laudeman, I., Shelden, S., Branstrom, R., & Brasil, C. (1998). Dynamicdensity: An air traffic management metric (NASA-TM-1988-11226). Moffet Field, CA: NASAAmes Research Center.

Law, D. J., Pellegrino, J. W., Mitchell, S. R., Fischer, S. C., McDonald,T. P., & Hunt, E. B. (1993). Perceptual and cognitive factors gov-erning performance in comparative arrival-time judgments. Journalof Experimental Psychology: Human Perception and Performance,19, 1183–1199.

Leedal, J. M., & Smith, A. F. (2005). Methodological approaches toanesthetist’s workload in the operating theatre. British Journal ofAnaesthesia, 94, 702–709.

Leiden, K. J., Kopardekar, P., & Green, S. (2003, November). Controllerworkload analysis methodology to predict increases in airspacecapacity. Paper presented at the AIAA’s 3rd Annual Aviation Tech-nology, Integration, and Operations (ATIO) Forum, Denver, CO.

Leplat, J., & Bisseret, A. (1966). Analysis of the processes involved inthe treatment of information by the air traffic controller. Controller,1, 13–22.

Loft, S., Humphreys, M., & Neal, A. (2003). Prospective memory inair traffic control. In G. Edkins & P. Pfister (Eds.), Innovation andconsolidation in aviation (pp. 287–293). Aldershot, UK: Ashgate.

Majumdar, A., & Ochieng, W. (2002). Factors affecting air traffic con-troller workload: A multivariate analysis based upon simulationmodeling of controller workload. Transportation Research Record:Journal of the Transportation Research Board, 1788, 58–69.

Majumdar, A., Ochieng, W., McAuley, G., Lenzi, J. M., & Lepadatu, C.(2004). The factors affecting airspace capacity in Europe: A crosssectional time-series analysis using simulated controller workloaddata. Journal of Navigation, 57, 385–405.

Manning, C., Mills, S., Fox, C., & Pfleiderer, E. (2001). Investigatingthe validity of performance and objective workload evaluationresearch (POWER) (Rep. No. DOT/FAA/AM-01/10). Washington,DC: Federal Aviation Administration Office of Aviation Medicine.

Manning, C., Mills, S., Fox, C., Pfleiderer, E., & Mogilka, H. (2002).Using air traffic control taskload measures and communicationevents to predict subjective workload (No. DOT/FAA/AM-02/4).Washington DC: Federal Aviation Administration Office of AviationMedicine.

Masalonis, A. J., Callaham, M. B., & Wanke, C. R. (2003). Dynamicdensity and complexity metrics for realtime traffic flow manage-ment. McLean, VA: MITRE.

Means, B., Mumaw, R. J., Roth, C., Schlager, M. S., McWilliams, E.,Gagné, E., et al. (1988). ATC training analysis study: Design of thenext generation ATC training system (Rep. No. FAA/OPM 342-036). Washington, DC: U.S. Department of Transportation/FederalAviation Administration.

Metzger, U., & Parasuraman, R. (2001). The role of the air traffic con-troller in future air traffic management: An empirical study of activecontrol versus passive monitoring. Human Factors, 43, 519–528.

Mogford, R. H., Guttman, J., Morrow, S. L., & Kopardekar, P. (1995). Thecomplexity construct in air traffic control: A review and synthesisof the literature (No. DOT/FAA/CT-TN95/22). Atlantic City, NJ:Federal Aviation Administration, William Hughes Technical Center.

Mogford, R. H., Murphy, E. D., Yastrop, G., Guttman, J. A., & Roske-

Hofstrand, R. J. (1993). The application of research techniques fordocumenting cognitive processes in air traffic control (Rep. No.DOT/FAA/CT-TN93/39). Atlantic City, NJ: Federal AviationAdministration.

Moray, N. (1979). Mental workload: Its theory and measurement. NewYork: Plenum Press.

Moray, N. (1997). Human factors in process control. In G. Salvendy(Ed.), Handbook of human factors and ergonomics (pp. 1944–1971).New York: Wiley.

Neal, A., Griffin, M., Neale, M., Bamford, E., & Boag, C. (1998). Humanfactors issues in the transition to a CNS/ATM environment: Finalreport (Report prepared for AirServices Australia). Brisbane,Australia: University of Queensland.

Niessen, C., Eyferth, K., & Bierwagen, T. (1999). Modeling cognitiveprocesses of experienced air traffic controllers. Ergonomics, 42,1507–1520.

Nunes, A., & Scholl, B. J. (2004). Interactions between convergenceangle, traffic load, and altitude distribution in air traffic control. InD. A. Vincenzi, M. Mouloua, & P. A. Hancock (Eds.), Human per-formance, situation awareness and automation technology: Currentresearch and trends (pp. 180–185). Mahwah, NJ: Erlbaum.

Pawlak, W. S., Brinton, C. R., Crouch, K., & Lancaster, K. M. (1996,July). A framework for the evaluation of air traffic control complex-ity. Presented at the AIAA Guidance, Navigation and Control Con-ference, San Diego, CA.

Pickup, L., Wilson, J. R., Sharpies, S., Norris, B., Clarke, T., & Young,M. S. (2005). Fundamental examination of mental workload in therail industry. Theoretical Issues in Ergonomics Science, 6, 463–482.

Rantanen, E., & Levinthal, B. (2005). Time-based modeling of humanperformance. In Proceedings of the Human Factors and Ergonom-ics Society 49th Annual Meeting (pp. 1200–1204). Santa Monica,CA: Human Factors and Ergonomics Society.

Rantanen, E., & Nunes, A. (2005). Hierarchical conflict detection in airtraffic control. International Journal of Aviation Psychology, 15,339–362.

Recarte, M. A., & Nunes, L. M. (2003). Mental workload while driving:Effects on visual search, discrimination and decision-making.Journal of Experimental Psychology: Applied, 9, 119–137.

Redding, R. E., Ryder, J. M., Seamster, T. L., Purcell, J. A., & Cannon,J. R. (1991). Cognitive task analysis of en route air traffic control:Model extension and validation (Report to the Federal AviationAdministration). McLean, VA: Human Technology. (ERIC Docu-ment Reproduction Service No. ED 340 848)

Remington, R. W., Johnston, J. C., Ruthruff, E., Gold, M., & Romera,M. (2000). Visual search in complex displays: Factors affectingconflict detection by air traffic controllers. Human Factors, 42,349–366.

Rodgers, M. D., & Drechsler, G. K. (1993). Conversion of the CTA,Inc., en route operations concept database into a formal sentenceoutline job task taxonomy (Rep. No. DOT/FAA/AM-93/1). Wash-ington, DC: Federal Aviation Administration Office of AviationMedicine.

Roske-Hofstrand, R., & Murphy, E. D. (1998). Human information pro-cessing in air traffic control. In M. W. Smolensky & E. S. Stein(Eds.), Human factors in air traffic control (pp. 65–114). San Diego,CA: Academic Press.

Rouse, W. B., Edwards, S. L., & Hammer, J. M. (1993). Modeling thedynamics of mental workload and human performance in complexsystems. IEEE Transactions on Systems, Man, and Cybernetics,23, 1662–1670.

Schaefer, D., Meckiff, C., Magill, A., Pirard, B., & Aligne, F. (2001,October). Air traffic complexity as a key concept for multi-sectorplanning. Presented at the Digital Avionics Systems Conference,Daytona Beach, FL.

Schmidt, D. K. (1976). On modeling ATC workload and sector capac-ity. Journal of Aircraft, 13, 531–537.

Schmidt, D. K. (1978). A queuing analysis of the air traffic controller’swork load. IEEE Transactions on Systems, Man, and Cybernetics,SMC-8, 492–498.

Seamster, T. L., Redding, R. E., Cannon, J. R., Ryder, J. M., & Purcell,J. A. (1993). Cognitive task analysis of expertise in air traffic con-trol. International Journal of Aviation Psychology, 3, 257–283.

Sheridan, T. B. (2002). Humans and automation: System design andresearch issues. Santa Monica, CA: Wiley/Human Factors andErgonomics Society.


Sperandio, J. C. (1971). Variation of operator’s strategies and regulat-ing effects on workload. Ergonomics, 14, 571–577.

Sperandio, J. C. (1978). The regulation of working methods as a func-tion of workload among air traffic controllers. Ergonomics, 21,195–202.

Stamp, R. G. (1992). The DORATASK method of assessing ATC sectorcomplexity — An overview (DORA Communication 8934, Issue2). London: Civil Aviation Authority.

Stein, E. S. (1985). Air traffic controller workload: An examination ofworkload probe (No. DOT/FAA/CT-TN82/24). Atlantic CityInternational Airport, NJ: Federal Aviation Administration Tech-nical Center.

Weitzman, D. O. (1993). Identifying information processing require-ments for air traffic control problem solving: Designing for diver-sity. In Proceedings of the Human Factors and Ergonomics Society37th Annual Meeting (pp. 103–107). Santa Monica, CA: HumanFactors and Ergonomics Society.

Wickens, C. D. (1984). Processing resources in attention. In R.Parasuraman & D. R. Davies (Eds.), Varieties of attention (pp.63–102). New York: Academic Press.

Wickens, C. D. (1992). Engineering psychology and human perfor-mance. New York: Harper Collins.

Wickens, C. D. (2002). Situation awareness and workload in aviation.Current Directions in Psychological Science 11, 128–133.

Wickens, C. D., Mavor, A. S., & McGee, J. P. (1997). Flight to the future:Human factors in air traffic control. Washington, DC: NationalAcademy Press.

Willems, B., Allen, R. C., & Stein, E. S. (1999). Air traffic control spe-cialist visual scanning: 2. Task load, visual noise, and intrusionsinto controller airspace (DOT/FAA/CT-TN99/23). Atlantic City,NJ: Federal Aviation Administration.

Wilson, G. F. (2002). An analysis of mental workload in pilots duringflight using multiple physiological measures. International Journalof Aviation Psychology, 12, 3–18.

Wyndemere, Inc. (1996, October). An evaluation of air traffic controlcomplexity (Final Rep.; Contract No. NAS2-14284). Moffett Field,CA: NASAAmes Research Center.

Shayne Loft is a postdoctoral research fellow in the KeyCentre for Human Factors and the School of Psycholo-gy at The University of Queensland. He received hisPh.D. in psychology in 2004 from the University ofQueensland.

Penelope Sanderson is a professor of cognitive engi-neering and human factors at The University of Queens-land, where she directs the Australian Research CouncilKey Centre for Human Factors. She received her Ph.D.in psychology in 1985 from the University of Toronto.

Andrew Neal is a senior lecturer in the Key Centre forHuman Factors and the School of Psychology at TheUniversity of Queensland. He received his Ph.D. in cog-nitive and organizational psychology in 1996 from theUniversity of New South Wales.

Martijn Mooij is a research assistant at the School ofPsychology, Key Centre for Human Factors, Universityof Queensland. He received an M.Sc. in human-machine interaction in 2001 from the Technical Uni-versity of Eindhoven, Netherlands.

Date received: November 2, 2005Date accepted: April 5, 2006

modeling and predicting mental workload in en route air ... · placed on the human operator and the...

Documents