PLEASE SCROLL DOWN FOR ARTICLE

This article was downloaded by: [Walker, Guy Harrison]On: 23 January 2010Access details: Access Details: [subscription number 918818663]Publisher Taylor & FrancisInforma Ltd Registered in England and Wales Registered Number: 1072954 Registered office: Mortimer House, 37-41 Mortimer Street, London W1T 3JH, UK

ErgonomicsPublication details, including instructions for authors and subscription information:http://www.informaworld.com/smpp/title~content=t713701117

From ethnography to the EAST method: A tractable approach forrepresenting distributed cognition in Air Traffic ControlGuy H. Walker a; Neville A. Stanton b; Chris Baber c; Linda Wells d; Huw Gibson c; Paul Salmon e;Daniel Jenkins f

a School of the Built Environment, Defence Technology Centre for Human Factors Integration (DTCHFI), Heriot-Watt University, Edinburgh, UK b School of Civil Engineering and the Environment,University of Southampton, UK c School of Electronic and Electrical Engineering, University ofBirmingham, UK d Aerosystems International, Yeovil, UK e MUARC, Monash University, Melbourne,Australia f Sociotechnic Solutions, St Albans, UK

Online publication date: 22 January 2010

To cite this Article Walker, Guy H., Stanton, Neville A., Baber, Chris, Wells, Linda, Gibson, Huw, Salmon, Paul andJenkins, Daniel(2010) 'From ethnography to the EAST method: A tractable approach for representing distributedcognition in Air Traffic Control', Ergonomics, 53: 2, 184 — 197To link to this Article: DOI: 10.1080/00140130903171672URL: http://dx.doi.org/10.1080/00140130903171672

Full terms and conditions of use: http://www.informaworld.com/terms-and-conditions-of-access.pdf

This article may be used for research, teaching and private study purposes. Any substantial orsystematic reproduction, re-distribution, re-selling, loan or sub-licensing, systematic supply ordistribution in any form to anyone is expressly forbidden.

The publisher does not give any warranty express or implied or make any representation that the contentswill be complete or accurate or up to date. The accuracy of any instructions, formulae and drug dosesshould be independently verified with primary sources. The publisher shall not be liable for any loss,actions, claims, proceedings, demand or costs or damages whatsoever or howsoever caused arising directlyor indirectly in connection with or arising out of the use of this material.

From ethnography to the EAST method: A tractable approach for representing distributed

cognition in Air Traffic Control

Guy H. Walkera*, Neville A. Stantonb, Chris Baberc, Linda Wellsd, Huw Gibsonc, Paul Salmone andDaniel Jenkinsf

aSchool of the Built Environment, Defence Technology Centre for Human Factors Integration (DTC HFI), Heriot-Watt University,Edinburgh, UK; bSchool of Civil Engineering and the Environment, University of Southampton, UK; cSchool of Electronic andElectrical Engineering, University of Birmingham, UK; dAerosystems International, Yeovil, UK; eMUARC, Monash University,

Melbourne, Australia; fSociotechnic Solutions, St Albans, UK

(Received 19 September 2008; final version received 7 July 2009)

Command and control is a generic activity involving the exercise of authority over assigned resources,combined with planning, coordinating and controlling how those resources are used. The challenge forunderstanding this type of activity is that it is not often amenable to the conventional experimental/methodologicalapproach. Command and control tends to be multi-faceted (so requires more than one method), is made up ofinteracting socio and technical elements (so requires a systemic approach) and exhibits aggregate behaviours thatemerge from these interactions (so requires methods that go beyond reductionism). In these circumstances adistributed cognition approach is highly appropriate yet the existing ethnographic methods make it difficult to applyand, for non-specialist audiences, sometimes difficult to meaningfully interpret. The Event Analysis for SystemicTeamwork method is put forward as a means of working from a distributed cognition perspective but in a way thatgoes beyond ethnography. A worked example from Air Traffic Control is used to illustrate how the language ofsocial science can be translated into the language of systems analysis.

Statement of Relevance: Distributed cognition provides a highly appropriate conceptual response to complex worksettings such as Air Traffic Control. This paper deals with how to realise those benefits in practice without recourseto problematic ethnographic techniques.

Keywords: distributed cognition; air traffic control; command and control; situational awareness

Introduction

Command and control

In command and control scenarios there is a commongoal (comprising interacting sub-goals), there aremultiple individuals who need to communicate andcoordinate with each other in order to attain these goalsand, increasingly, there are ever more complex ways offacilitating this process with technology. Commandand control, at the most generic level, can be viewed asa form of management infrastructure for planning andorganisation (Harris and White 1987). It involves theexercise of authority and direction by properly desig-nated individual(s) over assigned resources, as well asplanning, directing, coordinating and controlling howthose resources are deployed (Builder et al. 1999).‘Command’ can be viewed as the definition of overallsystem objectives or goals, whereas ‘control’ is themanagement of process and activities that lead to theachievement of these objectives (or to compensate forchanges in the environment that hinder their achieve-ment). Many contemporary socio-technical systems

involve authority, planning, directing and coordinatingand can be considered as forms of command andcontrol. Air Traffic Control (ATC) is one example.

Distributed cognition

From a distributed cognition perspective (e.g.Hutchins 1995a), the task of ATC can be viewed as aform of ‘‘‘computation’’ to maintain separationbetween aircraft in a region of airspace’ (Fields, et al.1998, p. 86). It is further argued that the computationsdo not reside solely in the heads of individualcontrollers, instead they are distributed across theentire ATC system, comprising numerous controllers,teams and technical artefacts. The essence ofdistributed cognition is on ‘how [these computations]transcend the boundaries of the individual actor’(Rogers 1997, p. 1, Hutchins 1995a, Hollan et al.2000).

The language of representational states is used todescribe the visible and external manifestations of

*Corresponding author. Email: g.h.walker@soton.ac.uk

Ergonomics

Vol. 53, No. 2, February 2010, 184–197

ISSN 0014-0139 print/ISSN 1366-5847 online

� 2010 Taylor & Francis

DOI: 10.1080/00140130903171672

http://www.informaworld.com

Downloaded By: [Walker, Guy Harrison] At: 23:50 23 January 2010

various ‘environmental contributions’ to the totalsystem (Rogers and Ellis 1994, Fields et al. 1998).Representational states subsume the full range ofobservable interactions between people and artefacts,as well as the resulting states (and state changes)that arise from the various ‘computations to main-tain separation between aircraft’. For example, anobservable interaction might be a controller issuing aninstruction to an aircraft. The resulting state might bea corresponding change in the aircraft’s representationon the radar display.

In command and control situations these computa-tions and representational states interact. A change inrepresentational state leads to further computations,further representational states and further computa-tions. But whilst these simple low level mechanismscan be multiplied in simple ways to form the totalATC system, the high level function, the system’saggregate behaviour, can be highly complex andadaptive (Chalmers 1990). Phenomenon ‘whereincomplex, interesting high-level function is producedas a result of combining simple low-level mechanismsin simple ways’ is referred to as ‘emergence’ (Chalmers1990, p. 2). A key emergent property of ATC is the so-called ‘picture’ or, in ergonomics parlance, situationalawareness (SA).

Distributed situation awareness

The ability to sense changes in representational states,understand them, then perform some kind of compu-tation based on them, not only describes the essence ofdistributed cognition but also that of situation aware-ness. At an individual level SA is about the psycho-logical processes for, and information in workingmemory required, in order to develop ‘the picture’ (e.g.Endsley 1995, Bell and Lyon 2000). The controller’spicture is not the same as the radar display, it does notarise solely from any one physical or human compo-nent, it arises out of the interaction of many suchcomponents. It arises both from the parts of the ATCtask that have been overtly designed (i.e. radar displaysand other prescribed forms of communication) andthose parts of the task that have not been overtlydesigned (i.e. the clicking of new flight data stripsinforms controllers that a new aircraft is about to entertheir sector). Because the picture, or SA, arises fromthese myriad individual components yet its totalitycannot be predicted solely from any one of themindividually, it can be referred to as emergent.

Individualistic approaches to SA dominate ergo-nomics but whilst they may be appropriate for tasksthat are performed by individuals in isolation, fewcomplex tasks are performed entirely independentlyof others (Perry 2003). In systems terms, SA is what

helps entire socio-technical systems such as ATC tobe orientated towards and ‘tightly coupled to thedynamics of the environment’ (Moray 2004, p. 4). Adistributed cognition perspective applied to commandand control scenarios requires a shift from traditionalnotions of SA that focus on the individual (e.g.Endsley 1995) to those that focus on the system (e.g.Sandom 2001, Gorman et al. 2006, Salmon et al. 2008).At face value the response to this might be the conceptof team SA (e.g. Salas et al. 1995, Perla et al. 2000) buteven here there are problems. ‘The degree to whichevery team member possesses the SA required for hisor her responsibilities’ (Endsley 1995, p. 39) runs intodifficulties when confronted with the twin concepts of‘overlapping SA’ (i.e. portions of SA that areidentically shared between people, normallyrepresented as a Venn diagram) and ‘compatible SA’(i.e. that which is ‘not’ overlapping between two ormore people, but which fits together like a jigsaw;Salmon et al. 2008). The fundamental problem stemsfrom a tacit assumption that the ‘situation’ can bedefined as a single, objective, external reality and thatthe goal of the people operating within the situation isto respond to all features appropriately. This isproblematic on three counts:

(1) There are many aspects of command andcontrol scenarios that require the individual tomake judgements and interpretations (so theassumption of the ‘objective reality’ of asituation is not always valid).

(2) There are multiple sub-goals and multiple viewsof the situation (so the idea of a single reality isalso not valid).

(3) Different agents within the system use differentrepresentational states to inform and supporttheir work, so the notion that there can be asingle view of the situation (as opposed toseveral interlocking views) is also not easilysupported.

One way to resolve the mismatch betweenmainstream thinking in SA and distributed cognitionof the sort encountered in complex systems, such asATC, is to consider one of the relatively invariantproperties of it: information. According to Bell andLyon (2000): ‘all aspects of momentary SA areeventually reducible to some form of [. . .] informationin working memory’ (p. 42). Information, in the SAsense, refers to what, in distributed cognition language,are called representational states. The question to askis whether ‘working memory’ is the only place wheresuch states can be represented. Distributed cognitionwould suggest not. It suggests that non-humanartefacts can create, manage and share such states (to

Ergonomics 185

Downloaded By: [Walker, Guy Harrison] At: 23:50 23 January 2010

some extent at least), meaning that the technical aspectsof a socio-technical system will be contributing to theexchange of representational states too. The totality ofthis will be a form of systems-level awareness that is nottraceable to any one individual and consistent with adistributed cognition view of the world, nor does it resideexclusively in the minds of humans. Thus, not only is theindividual-level ATC ‘picture’ emergent, so to is thesystems-level ‘picture’.

Another consideration is that representationalstates can be promulgated around the system withvery little in the way of overt communication. One ofthe great strengths of expert operators is their ability tochunk information, to abstract and pattern match, todevelop a high level of awareness from relatively littleinformation in the world (e.g. Chase and Simon 1973).The update of a representational state for one agentmight lead to partial updating of that used by another.For example, as an aeroplane moves across a sector itsroute is plotted on the ATC displays and its positionupdated dynamically; if there is no definable risk, thenthe updating happens automatically and without theneed for intervention.

Beyond ethnography

Whilst there is much to say in favour of blending someof the ideas connected with SA and distributedcognition, there are several problematic issues foractually applying either approach. It should be clear bynow that purely individualistic approaches to analysismay not capture all the required information aboutcomplex command and control scenarios. It is also thecase that the traditional experimental approach maylack predictive efficiency in the face of significantnumbers of emergent behaviours. Distributed cogni-tion, focusing as it does ‘on the material and socialmeans of the construction of action and meaning’(Hollan et al. 2000, p. 178) employs ethnography tounderstand how information is used to supportdecision making, how it is represented and how it ismanifest in the physical and social world (e.g. Rogersand Ellis 1994, Hutchins and Klausen 1996, Hutchinsand Holder 2000, Hutchins and Holder 2001). Ethno-graphy is a form of naturalistic, qualitative descriptionbased on observation. True ethnography places arequirement to live as a member of the ‘tribe’ for anextended period (possibly weeks or months) and toundertake the tasks and rituals of the tribe. Theunderstanding gained from this participatory observa-tion normally forms the basis of a report written from afirst person perspective as a participant observer. In thefield of distributed cognition, ethnographic methods aredeployed. The problems with ethnography in terms ofapplying it in practice are as follows:

(1) The outputs of ethnographic approaches (asthey are used currently) remain couched at aqualitative, often highly discursive level ofanalysis (e.g. Hutchins 1995b, Hutchins andKlausen 1996, Hutchins and Palen 1997). Suchoutputs may not be easily reconciled with apredominantly engineering, or at least non-social science audience (McMaster and Baber2005).

(2) Depending on the level of analysis, having thenecessary requirement for an observer to be‘imbued’ in the culture of the scenario can, insome cases, be incompatible with objectivityand validity in measurement (e.g. Hutchins1995b).

(3) Ethnographic analyses are not alwayseasily amenable to generalisation, concerned asthey are with specific and localised scenariosand phenomena. It is certainly difficult toimagine some analyses being amenable torepetition.

These issues limit the practical value of distributedcognition in command and control settings andserve as a barrier to employing what is potentially auseful and enlightened approach to the analysis ofcomplex systems. A response to this problem,however, appears to lie in the fundamentals of thedistributed cognition approach itself. Bycharacterising complex systems using the languageand metaphors of cognitive science, many of thephenomena of interest, such as communications,shared awareness and other determinates of decisionsuperiority, are rendered physically and observablymanifest. That is to say, they no longer reside just inthe heads of individual actors, they becomemanifest in the way that information is represented,modified, communicated and shared. The purpose ofthis paper is to deploy the Event Analysis forSystemic Teamwork (EAST) method as a way ofcapturing these phenomena, to show that there areways to leverage the favourable theoreticalperspective of distributed cognition but in ways thatare appropriate and relevant to systems designers.

The Air Traffic Control work setting

A case study from the ATC work domain is used toillustrate the complementary approach to ‘doing’distributed cognition. The primary strategic objectiveof ATC is safety. This is achieved by providinginstructions on height, speed and route to aircraftpilots so that individual aircraft maintain legallymandated physical separation criteria (typically 3 or 5miles horizontally and 1000 feet vertically; Civil

186 G.H. Walker et al.

Downloaded By: [Walker, Guy Harrison] At: 23:50 23 January 2010

Aviation Authority 2004). A secondary tactical objec-tive is to optimise the routing of aircraft so that theytake off and land at prescribed times and follow themost expeditious routes to destinations. These strategicgoals translate into the following operational levelactivities:

. Keeping in radio and/or radar contact withaircraft.

. Instructing aircraft in relation to speed, altitudeand direction.

. Providing information to aircraft about weatherconditions.

. Ensuring that minimum distances are maintainedbetween aircraft.

. Handling unexpected events, emergencies andunscheduled traffic.

Underlying these activities are several key non-human artefacts. Like their human counterparts theycontain, represent and modify information and arepart of a system that a distributed cognition perspec-tive is able to model. Fields et al. (1998) provide acogent summary of the key informational artefactscontained within the ATC work domain. These are aslisted below.

Charts and standard routes

These can be seen as a prescription of how certainmanoeuvres should be made. The information isshared between controllers and pilots as well as beingstored on aircraft flight management systems (Fieldset al. 1998).

Flight data strip

These can be seen as a representation of the projectedstate of aircraft. It takes the form of a paper stripthat contains information on aircraft height, speed,heading and call sign. This information is modifiedby the controller by hand when instructions are issuedand confirmation of receipt is received (Fields et al.1998).

Flight data strip bay

Whereas the individual flight data strip is a projectedstate of an individual aircraft, the totality of data stripsthat refer to a region of airspace provides a ‘schematicmodel’ of air traffic progress. (Fields et al. 1998, p. 87).

Radar display

‘The radar screen provides controller’s with a snapshotof the current horizontal [. . .] locations of aircraft’

annotated with individual aircraft call signs and height(Fields et al. 1998, p.88).

Aircraft call signs

These are the unique alpha-numeric codes given toindividual aircraft. Aircraft call signs serve ‘as animportant (indeed the only) means of coordinatingthe information represented in [. . .] various media’,such as the flight strip and radar display (Fields et al.1998, p. 88).

Communications

These represent, in a practical sense, the ‘mediation ofcontrol’. Fields et al. (1998) argue thatcommunications can also be seen as a ‘system ofrepresentations’ (p. 4). Of critical importance is thatcommunications ‘are also situated in the network ofartefacts and information, and are [only] madecomprehensible by reference to a larger context ofshared representations’ (p. 88).

Distributed cognition methodology

The importance of methods

Explicit methods lie at the heart of ergonomics as adiscipline, enabling the practitioner to vary theirapproach between scientist (i.e. testing and developingtheories of human performance using rigorous datacollection and analysis techniques) and practitioner(evaluating the effects of change, developing best-practice and, fundamentally, addressing real-worldproblems). Ergonomics methods are useful in thescientist–practitioner model because of the structureand potential for repeatability that they offer over andabove ethnography alone.

Descriptive vs. formative methods

This study uses the EAST method in an attempt toreconcile distributed cognition with themethodological traditions of ergonomics. EAST isbased on the integration of seven individualmethods, which in turn is a reflection of the multi-faceted nature of the command and control. In otherwords, no one method can adequately describe all ofthe degrees of freedom inherent in such a complexsocio-technical system. That said, no such claim ismade for the EAST method, but it can be argued thatat least some of the major human dimensions of theproblem space can be explored by taking a multi-method approach.

EAST is a descriptive method. It does not specify aformal architecture and what ‘should’ happen. Even

Ergonomics 187

Downloaded By: [Walker, Guy Harrison] At: 23:50 23 January 2010

though it uses normative methods such as task analysisand process modelling, these are populated with dataon what is actually observed. Neither does EAST focuson constraints, boundaries and a problem space defi-ned formatively by the scope of what ‘could happen’,such as cognitive work analysis. EAST focuses on what‘did’ happen. CWA admits the possibility of non-linearand emergent behaviour, whereas EAST is designed toidentify specific instances of it.

Despite their descriptive vs. formative differences,EAST and CWA are both representative of a shiftin methodological thinking. They share two keyaspects: both acknowledge that complex socio-techni-cal systems require more than one approach (EASTcomprises seven individual methodologies, CWAcomprises five ‘phases’) and both acknowledge thatthese perspectives are as interlinked as the complexsocio-technical phenomenon under analysis (i.e. theyare both systemic in nature). This is a core principle ofsocio-technical design (Clegg 2000).

Method integration

The following formal methodologies combine to formEAST: hierarchical task analysis (HTA; Annett);coordination demand analysis; communications usagediagram; social network analysis (SNA; Driskell andMullen 2005); propositional networks; an enhancedform of operation sequence diagram. A multiplemethod approach has a number of compellingadvantages. Not only does the integration of existingmethods bring reassurance in terms of a validationhistory but it also enables the same data to beanalysed from multiple perspectives. Also, with over200 existing methodologies to choose from (Stantonet al. 2005) there seemed little pragmatic need todevelop yet more. Of course, multiple interconnectedmethods require greater effort to analyse but acompanion to EAST, called workload, error, situationawareness, tasks and time, is designed to help. This is asoftware tool that greatly streamlines and simplifiesthe application of the method and it was used in thecurrent analysis (Houghton et al. 2006).

EAST is structured as follows. The HTA providesinput into the analysis of team working (CDA),communications usage (CUD) and the linkage (viacommunications) between agents (SNA). Data forthe HTA are gathered from live observation of thescenario. The output of all these methods (HTA, CDA,CUD and SNA) is given a summary visual form byusing an enhanced OSD. Interview data, in the form ofthe critical decision method (CDM; Klein and Arm-strong 2005), are used to create a network of linked‘information objects’ or representational states. Thisrepresentation is called a propositional network

and is rather similar in concept to semanticnetworks. It is important to note at this point that thepurpose of this paper is not to introduce themethodological intricacies of the EAST method (thereader is referred to Stanton et al. 2005 for furtherdetail) but to show the effect of method integrationin terms of enabling a distributed cognitionperspective upon ATC.

Air Traffic Control scenarios

Data were collected from the ATC work domainbetween 28 and 30 June 2004 at a major UK terminaland area control centre. Separate analyses wereperformed on four discrete scenarios that wereobserved to take place repeatedly. By dividing up thecontroller’s task in this way it was possible to capturesomething of the dynamics of the system. In practice,multiple aircraft are presenting themselves to thecontroller(s), prompting them to engageinterchangeably in one of the following fourbehaviours:

Scenario 1: To bring aircraft inbound from amajor air route into a holding stack and then passthem onto an aerodrome controller (the holdingscenario).

Scenario 2: To deal with aircraft that have left theholding stack and are en route to airfield(s), butprior to the final approach phase (the approachscenario).

Scenario 3: To deal with over-flying aircraft insuch a way as to avoid conflict with the holdingstack and other en-route aircraft (the overflightscenario).

Scenario 4: To deal with departing aircraft in sucha way as to avoid conflict with the holding stack(the departure scenario).

Applying the method

The first stage of the EAST method involves a two-step process of observation and interview (Annett2005). This is similar in some respects to theethnographically based approaches currently used indistributed cognition research but here there is not aparticular requirement for the observer to be directly‘imbued’ in the scenario. The focus instead is onunobtrusive observation. The bulk of the analysisderives from live audio feeds, which detail who iscommunicating to whom and about what. Thesedata are supplemented by the observers’ notes andby in-depth technical critique and insight providedby subject matter experts.

188 G.H. Walker et al.

Downloaded By: [Walker, Guy Harrison] At: 23:50 23 January 2010

Whilst observational techniques provide informa-tion on the observable artefacts of interaction, theyproduce limited data on the representational statesinternal to individuals. It is to the CDM (Klein andArmstrong 2005) that relevant insights can be pro-vided. The CDM is a semi-structured interview tech-nique that uses cognitive probes in order to elicitinformation on expert decision making. This wasadministered to each participant in relation to eachscenario.

Representing distributed cognition

The data collection methods provide informationon the activities performed within ATC, how thoseactivities are facilitated and how non-human artefactsparticipate in the joint cognitive system by containing,representing and transforming information. Usingmore conventional ethnographic techniques, thesefacets of a scenario would be described, exhaustively,in a first person written account. While this cancapture the nuances of the situation as it was observed,and is often illustrated with diagrams and photographsof the work setting, it is not easy to summarise theaccount in a manner that brings out common features.In this paper the EAST method is used to presentmultiple views of the situation in such a way as tocomplement an ethnographic account, if such an

analysis were to be performed (which in a lot ofergonomics studies would be unlikely). The EASTmethod maps the tasks being performed (in a tasknetwork), the communications between agents (ina social network) and the representational statesbeing sensed (in a propositional network).Figure 1 illustrates the relationships between thesethree views.

Task networks

HTA is a means of describing a system in terms of astructured hierarchy of goals and sub-goals withfeedback loops (Annett 2005). Its appropriateness inthis instance can be put down to two key points.First, it is inherently flexible and the approach canbe used to describe any system. Second, it can be usedfor many ends, from person specification to trainingrequirements, to error prediction, to teamperformance assessment and system design. Key to itspresent application is its ability to model the temporalrelations between tasks and the external conditionsupon which task activity is cued. The task network,specifically, is a way of representing the detailedtask analysis in terms of the interaction of higher levelgoals. This technique allows complex task analysesto be easily compared according to differencesin overall task structure and type. The task

Figure 1. The three network-based approaches provide a detailed characterisation of a complex socio-technical system.SA ¼ situational awareness.

Ergonomics 189

Downloaded By: [Walker, Guy Harrison] At: 23:50 23 January 2010

analysis/network forms the foundation for developinginsightful social and propositional networks.

Social networks

SNA is particularly compatible with the distributedcognition perspective because it ‘[. . .] focuses on therelationships among actors embedded in their socialcontext’ (Driskell and Mullen 2005, p. 58–1). Further-more, it can represent the technological mediation ofcommunication and networks in which some of thenodes are non-human. A social network is a way torepresent individuals or teams who are linked bycommunications to each other and to subject that

network to mathematical analysis using Graph Theory(Driskell and Mullen 2005). Two mathematicalindices are used in the present analysis, namely‘centrality’ (i.e. a numeric ranking allowing key agentsin the network to be identified) and ‘density’ (theinterconnectivity of the network as a whole). Themetrics reveal certain important characteristics of thenetworks to be revealed, in turn allowing comparisonsbetween them.

Propositional networks

Propositional networks offer a novel and effectivemeans of visualising representational states as held/

Figure 2. Social network diagram illustrating agents in the air traffic control (ATC) work domain linked by communication(emphasis denotes non-human agent).

190 G.H. Walker et al.

Downloaded By: [Walker, Guy Harrison] At: 23:50 23 January 2010

experienced by the individuals at work within thesystem. The data used to construct them are based onthe outputs of the CDM, in which a content analysisof the interview transcripts permits representationalstates to be extracted and causal links between themdefined. Representational states take the form ofspecific knowledge objects within these networks,which, in turn, are analogous to propositions, i.e.entity or phenomena about which an individualrequires information in order to act effectively. Thepropositional network offers four perspectives.

(1) They do not differentiate between differenttypes of representational state (e.g. informationrelated to states, people or ideas); therefore,from a design perspective they do not constrainassessments to consideration of existing config-urations of people and states, rather to therequired representational states associated with

a scenario (Stanton et al. 2005, Walker et al.2006).

(2) The network shows the totality of informationused in the scenario (within the constraints ofthe data collection techniques used) regardlessof whether agents in the scenario are human ortechnical.

(3) Shared SA can be accessed from the CDM, inwhich multiple agents can be attributed tocommon knowledge objects/states within thenetwork.

(4) The dynamic aspects of SA can also becaptured by animating the propositionalnetwork. This is achieved by highlighting activeand non-active knowledge objects/statesoccurring in different task scenarios.

These individual network-based outputs not onlymeet the need to go beyond reductionism (and focuson interconnections as well as parts) but they can alsobe linked to provide several different perspectives onthe scenario.

Application to Air Traffic Control

Analysis of agents in the distributed cognition system

The social network representation provides aparticularly powerful example of distributed cognition;in particular, the idea of a joint cognitive system ofcollaborating human and non-human agents. TheSNA of all the actors participating in the ATC taskshows that there are 21 agents in the network, 14human and seven non-human (shown as shaded inFigure 2) joined by 22 communication links. Note thatthis network has been simplified somewhat. Thenumber of aircraft that an individual controller wouldbe expected to handle could be significantly larger thanthe four shown for illustration in Figure 2.

The metrics ‘status’ and ‘centrality’ are used toidentify key agents in the scenario and these are shownin Table 1. Between them they indicate the degree ofconnectedness an individual agent has and,consequently, the amount of influence that agent hason the performance of the network as a whole. Keyagents can be defined as having high status and highconnectedness. The analysis indicates that thecontroller (as one might expect) has high status andcentrality in the network, but the metric also revealsthat the most significant agent is in fact non-human;the ATC technical system (which is the technicalinfrastructure and assemblage of ground aids,communications systems, radar displays, etc). A greatdeal of information is received, stored, displayed orotherwise transformed by this (non-human) agent.This is a similar finding to that of Houghton et al.’s

Table 1. Results of social network centrality for individualagents (in descending order).

Agent Centrality

ATC System 16.8Controller 14.0Heathrow Director 12.3Desk Coordinator 12.3Welin Controller 11.6Adjacent Controller 11.3Flight Strip Computer (Heathrow) 10.5Senior Watch Assistant 10.2FMS 1 10.0FMS 2 10.0FMS 3 10.0FMS 4 10.0Flight Strip Computer (West Drayton) 10.0Group Supervisor 10.0Deputy Watch Manager 10.0Watch Manager 10.0Flight Strip Assistant 9.5Pilot 2 9.3Pilot 3 9.3Pilot 4 9.3Pilot 1 9.1

ATC ¼ Air Traffic Control; FMS ¼ Flight Management System.

Note: The most central node is the assemblage of displays,equipment and ground aids.

Table 2. Number of communications links that existbetween geographically disperse locations.

From/To Number of Links

Aircraft to Airfield 3Aircraft to TCC 6TCC to Airfield 3

TCC ¼ Terminal Control Centre.

Ergonomics 191

Downloaded By: [Walker, Guy Harrison] At: 23:50 23 January 2010

(2006) analysis of police control rooms (in which thecentral logging computer was also the most centralagent).

In addition to the distribution between collaborat-ing human and non-human agents, cognition is alsodistributed in a geographical sense, between, forexample, aircraft in flight within the sector, control-ler(s) based at the control centre and controllers basedat airfield(s). Table 2 presents the number of timescommunication links cross these geographical bound-aries. Geographical dispersion is further illustrated ingeographical regions. The type of communicationsmedia is physically constrained in cases where linksbetween nodes cross from one shaded region to theother. For example, verbal communications are notphysically possible between controllers on the groundand pilots in flight without some sort of facilitation by

technical means (such as radio). There remain a host ofother local conditions that may also require some formof technological mediation. For example, individualcontrollers have to remain at their workstation duringtheir shift in order to communicate with controllerswho are not immediately adjacent, the telephone isused and/or the agent acting in the role of deskcoordinator may have to facilitate.

Facilitating technology

Figure 4 shows how ATC operations are facilitated byseven types of communication media.

Face-to-face communication comprises verbalcommunications. Face-to-face communication alsocontains a visual component. Controllers will pointand demonstrate visually to aid in understanding

Figure 3. Geographical dispersion of human and non-human agents in the air traffic control (ATC) work domain.

192 G.H. Walker et al.

Downloaded By: [Walker, Guy Harrison] At: 23:50 23 January 2010

instructions. Telephone and radio technology facil-itates voice communication. Radio also facilitates thedissemination of data, as do data network facilities.Written communication is dominated by the flightdata-strip aspect of the task, although machine inter-faces (e.g. the radar display) also contain writteninformation as well as visual representations. Table 3provides a summary by crossing communicationsmodality with communications media to provide atechnology/modality matrix.

ATC is a highly evolved and proceduralised workdomain and the prominent role of implicitcommunications is also noted. The CDM interviewgives access to some of these unobservable artefacts ofinteraction, one of which is a system of passing aircraftbetween sectors that does not rely on explicit verbalexchanges. The system works by putting aircraft into aparticular position in the new sector at a pre-agreedheight, speed and heading. As soon as the controllersees an aircraft in this position, he/she knows it is beingpassed onto them and they can take control of it. Awider awareness is gained via open channel radiocommunications and instructions to aircraftoverheard from other controllers. Several contextualfactors, such as the high tempo of operations andthe proceduralised nature of the task, enable thislevel of shared understanding and this form ofcommunication to take place. From a distributedcognition perspective it can be seen how acomparatively lengthy verbal exchange is oftenreplaced with a far simpler visio-spatial task(facilitated by external technological artefacts such as

Figure 4. Social network overlain with the media that facilitates communication.

Table 3. Communications modality/technology matrix.

Technology

Modality Phone Radio Network Strips MMI In-Person

VerbalWrittenDataVisual/Other

MMI ¼ Human Machine Interaction.

Ergonomics 193

Downloaded By: [Walker, Guy Harrison] At: 23:50 23 January 2010

Figure 5. Knowledge network representing the entire knowledge base for the air traffic control work domain. Knowledge that isactive during task enactment represents situational awareness at the systems level.

Table 4. Summary table of key knowledge objects (KO)that are active during different phases of air traffic controlprocess.

KnowledgeObject Departure Over-Flight Holding Approach

PressuresRunwayRadarStackAircraftFlight LevelStripPositionRead-backInstructionAcknowledgeCall SignFlight PlanPlanOptionsSector

TOTALKO (16)

10 10 7 8

Note: The table demonstrates changes in systems level state ofsituational awareness.

the radar display and flight data strips; Hutchins andKlausen 1996). These are captured in the socialnetworks. In other words, the presence of a link doesnot necessarily have to connote an overt form ofcommunication but also implicit ones.

Control architecture

Although Figures 2, 3 and 4 are visually complex, it ispossible to discern certain features of the socialnetwork. It can be noted that there is a diverginghierarchy from the Watch Manager level downwards.This hierarchy splits into the Group Supervisor level(and downwards towards individual controllers) andthe Senior Watch Assistant level (and downwardstowards Flight Strip Assistants). According to Dekker(2002) this so-called ‘split architecture’ ‘provides someof the benefits of centralised planning with tacticaladjustments to new information from subordinateunits’ (p. 5). At this level of ‘air traffic management’,the inherent complexity appears to justify a degreeof hierarchical sub-division so that aspects of com-plexity can be spread across agents. The possible

194 G.H. Walker et al.

Downloaded By: [Walker, Guy Harrison] At: 23:50 23 January 2010

trade-off in situations where the state of the world ischanging rapidly (a fast tempo), and where decisionsneed to be enacted quickly, is that any delay in thedissemination of information through a hierarchybecomes critical (Dekker 2002).

At the tactical or ‘air traffic control’ level, in caseswhere one agent requires some form of assistance, orto avoid the degradation of aircraft separation, thecontroller can interact sideways through the structureto an adjacent controller so that issues can be resolvedand task load shared or re-distributed quickly. Thispattern of communications is known as ‘peer to peer’or a ‘negotiation architecture’. In these instances, eachagent is more or less independently responsible for adefined area (Dekker 2002) and is able to reactpromptly to rapid changes in the environment.

A further characteristic of the scenario (from astructural perspective) is the amount of informationsharing. This relates back to the ATC system andits displays (which subsumes the entire radar,aircraft identification, warning, communication andground systems in general) being the highest rankingagent in terms of ‘centrality’. Overlain across thesetwo interleaved command architectures (split andnegotiated) is the provision of high quality informationto facilitate decision making within them. This overallconfiguration, known variously as a ‘negotiationarchitecture with information sharing’ is an emergingparadigm in several alternate domains, where it isreferred to as ‘network centric’. The benefits of thisarchitecture are agility, fast response and the ability toquickly organise and re-organise. In the theoreticalwork by Dekker (2002), the negotiation architecturewith information sharing proved to be the mosteffective in high tempo tasks.

Systemic situational awareness

Systemic SA is modelled using propositional networks(Figure 5). From these networks it is possible todivine certain structural characteristics of the knowl-edge base that underpins effective SA for the totalATC task.

Central nodes in the network are identified using acentrality metric based simply on divining nodes withfive or greater links. Table 5 presents an analysis ofcore knowledge objects for each individual scenariobased on this criterion. By representing and simplifyingSA in this manner, the table shows that depending onthe task (which occurs at different points in time) notonly does the type of information change, but byimplication, so does the structure of it. This informa-tion can be used to explore and understandthe information needs of controllers and, moreover,how one might go about supporting them.

Temporal aspects of command and control in Air TrafficControl

The controller will typically switch backwards andforwards among the four scenarios detailed above. Asthey do so, not only does the type and structure ofinformation change (as shown in Table 4) but so do thecharacteristics of the task and social networks. Tovisualise the temporal aspects of the interrelationsbetween the networks requires an alternate form ofrepresentation. Some form of animation seems a likelycandidate for bringing this facet to life but this, andfurther refinements to the method, though eminentlyfeasible, are within the purview of future work.

Conclusions

The purpose of this paper has been to show how theEAST method puts ergonomic analyses in touch withthe distributed cognition perspective, rendering theoutput much more tractable than comparableethnographic techniques. Although this paper isnecessarily couched at a summary level of analysis,the following key characteristics of the ATC workdomain are identified in Table 5. Within the table is achecklist used to sign-post where the system designerwill find detailed insights into the type, nature andstructure of these distributed cognition artefacts.

This paper has focused on a systems leveldescription of the ATC scenario and serves to illustratethat this description can be achieved with the EASTmethod in live settings. The strength of this descriptivelevel of analysis can be summarised as follows:

. The methods avoid bias by focusing on objectiveand manifest phenomena.

. The methods are applicable to any domain andthe results gained are comparable across andwithin domains.

. The results are graphical and easily interpreted,yet amenable to further summarisation usingtables and numerical indices.

. The summary level is underpinned byconsiderable detail that can be explored furtherin the context of system design.

. It is consistent with existing narrative approachesto distributed cognition analysis.

As well as providing a descriptive level ofanalysis, the real potential of the method lies in itsability to offer predictive insights, that is to ‘model’complex socio-technical systems. In theory, itshould be possible to subject the networks to knownchanges (in task, social or information structure)and to derive outputs as to the effect of these undervarious performance contexts. For example, an

Ergonomics 195

Downloaded By: [Walker, Guy Harrison] At: 23:50 23 January 2010

alteration in the command structure may influence thetype of communications available to an agent, there-fore affecting the type of information that is able tobe communicated to other agents, leading to adeleterious affect on SA (or indeed vice versa). Themodelling aspect of this work is at a nascent stagebut shows promise for further development. Whatcan be communicated about the current applicationof the method to live data is that the highly relevanttheoretical perspective of distributed cognition iswithin the reach of systems designers and ergonomists.

Acknowledgements

This work from the Human Factors Integration DefenceTechnology Centre (HFI DTC) was part-funded by theHuman Sciences Domain of the UK Ministry of DefenceScientific Research Programme.

References

Annett, J., 2005. Hierarchical task analysis. In: N.A. Stanton,A. Hedge, K. Brookhuis, and E. Salas, eds. Handbook ofhuman factors and ergonomics methods. London: CRC,33.1–33.7.

Bell, H.H. and Lyon, D.R., 2000. Using observer ratings toassess situation awareness. In: M.R. Endsley, ed. Situa-tion awareness analysis and measurement. Mahwah, NJ:Lea, 129–146.

Builder, C.H., Bankes, S.C., and Nordin, R., 1999. Commandconcepts: A theory derived from the practice of commandand control. Santa Monica, CA: Rand.

Chalmers, D.J., 1990. Thoughts on emergence [online].Available from: http://consc.net/notes/emergence.html[Accessed April 2005].

Chase, W.G. and Simon, H.A., 1973. Perception in chess.Cognitive Psychology, 4, 55–81.

Civil Aviation Authority, 2004. CAP 413: RadiotelephonyManual, edition 15. Gatwick, UK: Safety RegulationGroup, Civil Aviation Authority.

Clegg, C.W., 2000. Sociotechnical principles for systemdesign. Applied Ergonomics, 31, 463–477.

Dekker, A.H., 2002. C4ISR architectures, social networkanalysis and the FINC methodology: An experiment inmilitary organisational structure. Edinburgh, SouthAustralia: DSTO Electronics and Surveillance ResearchLaboratory.

Driskell, J.E. and Mullen, B., 2005. Social networkanalysis. In: N.A. Stanton, A. Hedge, K. Brookhuis,and E. Salas, eds. Handbook of human factors andergonomics methods. London: CRC, 58.1–58.6.

Endsley, M.R., 1995. Toward a theory of situationawareness in dynamic systems. Human Factors, 37 (1),32–64.

Fields, R., et al., 1998. Air traffic control as a distributedcognitive system: a case study of external representations.In: T.R.G. Green, L. Bannon, C.P. Warren, andJ. Buckley, eds. ECCE9: Proceedings of the ninthEuropean conference on cognitive ergonomics. Limerick,Republic of Ireland: EACE, 85–90.

Gorman, J.C., Cooke, N.J., and Winner, J.L., 2006.Measuring team situation awareness in decentralizedcommand and control environments. Ergonomics, 49(12–13), 1312–1325.

Harris, C.J. and White, I., 1987. Advances incommand, control & communication systems. London:Peregrinus.

Hollan, J., Hutchins, E., and Kirsh, D., 2000.Distributed cognition: toward a new foundation forhuman-computer interaction research. ACMTransactions on Computer Human Interaction, 7,174–196.

Table 5. Key characteristics of the air traffic control work domain and the network approach in which detailed systems levelinsights reside.

Specified in detail by:

Key Characteristic Task Network Social Network Knowledge Network

The coordination of individuals and teams toachieve a common goal (comprising separatethough interacting sub-goals, knowledge andsituational awareness (SA))

The technologically mediated geographicaldispersion of ‘agents’ (from controllers locatedin area control centres, or actual airfields, tospecific aircraft in controlled airspace).

A substantial technology infrastructure (providinghigh quality, accurate, real-time information viaseven different communications modalities),which supports the dispersion and representationof knowledge and a systems level state of SA.

A high degree of information sharing (where mostagents have access to parts of the technologyinfrastructure)

A highly evolved and proceduralised mode ofoperation, (involving implicit communicationbetween and among separate actors in thescenario)

196 G.H. Walker et al.

Downloaded By: [Walker, Guy Harrison] At: 23:50 23 January 2010

Houghton, R.J., et al., 2006. Command and control inemergency services operations: A social network analysis.Ergonomics, 49 (12–13), 1204–1225.

Hutchins, E., 1995a. How a cockpit remembers its speeds.Cognitive Science, 19, 265–288.

Hutchins, E., 1995b. Cognition in the wild. Cambridge, MA:MIT Press.

Hutchins, E. and Holder, B., 2000. Conceptual models forunderstanding an encounter with a mountain wave.Proceedings of HCI-Aero 2000, Toulouse, France,September.

Hutchins, E. and Holder, B., 2001. What pilots learn aboutautoflight while flying on the line. 11th internationalsymposium on aviation psychology, 6 March.

Hutchins, E. and Klausen, T., 1996. Distributed cognition inan airline cockpit. In: Y. Engestrom and D. Middleton,eds. Cognition and communication at work. Cambridge,UK: Cambridge University Press, 15–34.

Hutchins, E.L. and Palen, L., 1997. Constructing meaningfrom space, gesture, and speech. In: L.B. Resneck, R.Saljo, C. Pontecorvo, and B. Burge, eds. Discourse, toolsand reasoning: Essays in situated cognition. Heidelberg,Germany: Springer-Verlag, 23–40.

Klein, G. and Armstrong, A.A., 2005. Critical decisionmethod. In: N.A. Stanton, A. Hedge, K. Brookhuis,and E. Salas, eds. Handbook of Human factors andergonomics methods. London: CRC, 35.1–35.8.

McMaster, R. and Baber, C., 2005. Integrating humanfactors into systems engineering through a distributedcognition notation. The IEE and MoD HFI Symposiumon People and Systems. London: IEE, 77–83.

Moray, N., 2004. Ou’ sont les neiges d’antan? In: D.A.Vincenzi, M. Mouloua, and P.A. Hancock, eds. Humanperformance, situation awareness and automation: Currentresearch and trends. Mahwah, NJ: LEA, 1–31.

Perla, P.P., et al., 2000. Gaming and shared situationawareness [online]. Available from: http://www.thoughtlink.com/publications/GamingSSA00Abstract.htm [Accessed April 2005].

Perry, M., 2003. Distributed cognition. In: J.M. Carroll, ed.HCI models, theories and frameworks. San Francisco,CA: Morgan-Kaufmann, 93–224.

Rogers, Y., 1997. A brief introduction to distributed cognition.Brighton, UK: Interact Lab, University of Sussex.

Rogers, Y. and Ellis, J., 1994. Distributed cognition: analternative framework for analysing and explainingcollaborative working. Journal of InformationTechnology, 9 (2), 119–128.

Salas, E., Prince, C., and Baker, D.P., 1995. Situationawareness in team performance – implications formeasurement and training. Human Factors, 37, 123–136.

Salmon, P.M., et al., 2008. What really is going on? Reviewof situation awareness models for individuals and teams.Theoretical Issues in Ergonomics Science, 9 (4), 297–323.

Salmon, P.M., et al., 2008. What really is going on? Reviewof situation awareness models for individuals and teams.Theoretical Issues in Ergonomics Science, 9 (4), 297–323.

Sandom, C., 2001. Situation awareness. In: J. Noyes andM. Bransby, eds. People in control: Human factorsin control room design. London: Institution of ElectricalEngineers, 51–68.

Stanton, N.A., et al., 2005. Human factors methods: Apractical guide for engineering and design. Aldershot:Ashgate.

Walker, G.H., et al., 2006. Analysing the role ofcommunications technology in C4i scenarios: Adistributed cognition approach. Journal of IntelligentSystems Special Issue: Information and CommunicationTechnology, 15 (1–4), 299–328.

Ergonomics 197

Downloaded By: [Walker, Guy Harrison] At: 23:50 23 January 2010