handbook of human factors and ergonomics (salvendy/handbook of human factors 4e) || task analysis:...

PART 3DESIGN OF TASKS AND JOBS

Handbook of Human Factors and Ergonomics, Fourth Edition Gavriel SalvendyCopyright © 2012 John Wiley & Sons, Inc.

CHAPTER 13TASK ANALYSIS: WHY, WHAT, AND HOW

Erik HollnagelUniversity of Southern DenmarkOdense, Denmark

1 THE NEED TO KNOW 385

1.1 Role of Task Analysis in Human Factors 386

1.2 Artifacts and Tools 387

2 TASK TYPES AND TASK BREAKDOWN 388

2.1 Changing Views of Elementary Tasks 388

3 BRIEF HISTORY OF TASK ANALYSIS 389

3.1 Sequential Task Analysis 389

3.2 From Sequential to Hierarchical TaskOrganization 390

3.3 Functional Dependency and Goals–MeansTask Analysis 392

4 PRACTICE OF TASK ANALYSIS 393

4.1 Task Data Collection Techniques 394

4.2 Task Description Techniques 394

4.3 Task Simulation Methods 394

4.4 Task Behavior Assessment Methods 394

4.5 Future of Task Analysis 394

REFERENCES 395

1 THE NEED TO KNOW

The purpose of task analysis, as it is commonlypracticed, is to describe tasks and more particularly toidentify and characterize the fundamental characteristicsof a specific activity or set of activities. According tothe Shorter Oxford Dictionary , a task is “any piece ofwork that has to be done,” which is generally takento mean one or more functions or activities that mustbe carried out to achieve a specific goal. Task analysiscan therefore be defined as the study of what people,individually and collectively, are required to do in orderto achieve a given goal or objective.

Since a task by definition is a directed activitybecause it has a purpose or an objective, there is little orno methodological merit in speaking simply of activitiesand tasks without taking into account both their goalsand the context in which they occur. Task analysiscan therefore basically be defined as the study of whatpeople, individually and collectively, are required to doto achieve a specific goal, or, simply put, as who doeswhat and why .

Who Who refers to the people who carry out a task.In the case where this is a single person, task analy-sis is a description of individual work. It is, however,far more common that people have to work together, inpairs, as a team, or in an organization, in which casetask analysis is a description of the collective effort ofwhat the team does. Whereas a description of individualwork can focus on the activities, a description of col-lective work must also include how the collaboration isaccomplished, that is, the organization and coordinationof what the individuals do. Moving from the realm ofhuman work to artifacts or agents (such as robots), task

analysis becomes the analysis of functions (e.g., move-ments) that an artifact must carry out to achieve a goal.

In industrialized societies, tasks are in most casesaccomplished by people using some kind of technologi-cal artifact or system: in other words, a human–machinesystem. Task analysis is therefore often focused on thewhat of the human–machine system as such shoulddo: for instance, as task analysis for human–computerinteraction (e.g., Diaper and Stanton, 2003). Moregenerally, humans and machines working togethercan be described as cognitive systems or joint cog-nitive systems (Hollnagel and Woods, 2005). Indeed,at the time of writing (2010) the main issues are nolonger human work with technology, or human int-eraction with technology, but the coagency of mul-tiple functions, providers, stakeholders, and so on.Human work with technology is no longer a questionof human–technology interaction but rather a questionof how complex sociotechnical systems function. Thehuman–machine dyad and the focus on human–machineinteraction are relics from the early days of human fac-tors and are no longer adequate—if they have not alreadybecome irrelevant.

The built-in assumptions about the nature of whocarries out the task have important consequences for taskanalysis, as will be clear from the following. The useof the pronoun who should not be taken to mean thattask analysis is only about what humans do, althoughthat was the original objective. In contemporary termsit would probably be more appropriate to refer to thesystem that carries out the task.

What What refers to the contents of the task and isusually described in terms of the activities that constitutethe task. Task analysis started by focusing on physical

385Handbook of Human Factors and Ergonomics, Fourth Edition Gavriel SalvendyCopyright © 2012 John Wiley & Sons, Inc.

386 DESIGN OF TASKS AND JOBS

tasks (i.e., manifest work) but has since the 1970senlarged its scope to include cognitive or mental tasks.The content of the task thus comprises a systematicdescription of the activities or functions that make upthe task, either in terms of observable actions (e.g., gras-ping, holding, moving, assembling) or in terms of theusually unobservable functions that may lie behind theseactions, commonly referred to as cognition or cognitivefunctions .

Why Finally, why refers to the purpose or goal of thetask: for instance, the specific system state or conditionthat is to be achieved. A goal may be something thatis objective and physically measurable (a product)but also something that is subjective: for instance,a psychological state or objective, such as “havingdone a good job.” The task analysis literature hasusually eschewed the subjective and affective aspectsof tasks and goals, although they clearly are essentialfor understanding human performance as well as fordesigning artifacts and work environments.

Task analysis is supposed to provide concreteanswers to the practical questions of how things shouldbe done or are done. When dealing with work, andmore generally with how people use sociotechnicalartifacts to do their work, it is necessary to know bothwhat activities (functions) are required to accomplisha specified objective and how people habitually goabout doing them, particularly since the latter is usuallydifferent—and sometimes significantly different—fromthe former. Such knowledge is necessary to design,implement, and manage sociotechnical systems, andtask analysis looks specifically at how work takesplace and how it can be facilitated. Task analysistherefore has applications that go well beyond interfaceand interaction design and may be used to addressissues such as training, performance assessment, eventreporting and analysis, function allocation and automa-tion, procedure writing, maintenance planning, riskassessment, staffing and job organization, personnelselection, and work management.

The term task analysis is commonly used as a genericlabel. A survey of task analysis methods shows thatthey represent many different meanings of the term(Kirwan and Ainsworth, 1992). A little closer inspec-tion, however, reveals that they fall into a few maincategories:

• The analysis and description of tasks or workingsituations that do not yet exist or are based onhypothetical events

• The description and analysis of observations ofhow work is carried out or of event reports (e.g.,accident investigations)

• The representation of either of the above, inthe sense of the notation used to capture theresults (of interest due to the increasing use ofcomputers to support task analysis)

• The various ways of further analysis or refine-ment of data about tasks (from either of theforegoing sources)

• The modes of presentation of results and thevarious ways of documenting the outcomes

Methods of task analysis should in principle bedistinguished from methods of task description. A taskdescription produces a generalized account or summaryof activities as they have been carried out. It is basedon empirical data or observations rather than on designdata and specifications. A classical example is linkanalysis or even hierarchical task analysis (Annettet al., 1971). Properly speaking, task description orperformance analysis deals with actions rather than withtasks. This distinction, by the way, is comparable to theFrench ergonomic tradition where the described task isseen as different from the effective task. The describedtask (t ache prevue or simply t ache) is the intendedtask or what the organization assigns to the person,what the person should do. The effective task (t acheeffective or activite) is the actual task or the person’sresponse to the prescribed task, what the personactually does (Daniellou, 2005). Understanding thetask accordingly requires an answer to the question ofwhat the person does, while understanding the activityrequires an answer regarding how the person performsthe task. (In practice, it is also necessary to considerwhen and where the task is carried out.) An importantdifference between tasks and activities is that the latterare dynamic and may change depending on the circum-stances, such as fluctuations in demands and resources,changing physical working conditions, the occurrence ofunexpected events, and so on. The distinction betweenthe task described and the effective task can be appliedto both individual and collective tasks (Leplat, 1991).

1.1 Role of Task Analysis in Human FactorsTask analysis has over the years developed into astable set of methods that constitute an essential partof human factors and ergonomics as applied disciplines.The focus of human factors engineering or ergonomicsis humans at work, more particularly the human use oftechnology in work, although it sometimes may lookmore like technology’s use of humans. The aim ofhuman factors (which in the following is used as acommon denominator for human factors engineeringand ergonomics) is to apply knowledge about humanbehavior, abilities, and limitations to design tools,machines, tasks, and work environments to be as produc-tive, safe, healthy, and effective as possible. Fromthe beginning, ergonomics was defined broadly asthe science of work (Jastrzebowski, 1857). At thattime work was predominantly manual work, and toolswere relatively few and simple. Human factors, whichoriginally was called human factors engineering, cameinto existence around the mid-1940s as a way of solvingproblems brought on by emerging technologies suchas computerization and automation. Ergonomics andhuman factors thus started from different perspectivesbut are now practically synonymous. At the presenttime, literally every type of work involves the use oftechnology, and the difference between ergonomics andhuman factors engineering is rather nominal.

Throughout most of history, people have dependedon tools or artifacts to do their work, such as the

TASK ANALYSIS: WHY, WHAT, AND HOW 387

painter’s brush or the blacksmith’s hammer. As long asusers were artisans rather than workers and work wasan individual rather than a collective endeavor, the needof prior planning, and therefore of anything resemblingtask analysis, was limited or even nonexistent. Thedemand for task analysis arose when the use oftechnology became more widespread, especially whentools changed from being simple to being complex.More generally, the need of a formal task analysis ariseswhen one or more of the following three conditionsare met:

1. The accomplishment of a goal requires moreeffort than one person can provide or depends ona combination of skills that goes beyond what asingle individual can be expected to master. Insuch cases task analysis serves to break down acomplex and collective activity to descriptionsof a number of simpler and more elementaryactivities. For example, building a ship, incontrast to building a dinghy or a simple raft,requires the collaboration of many individualsand the coordination of many different types ofwork. In such cases people have to collaborateand must therefore adjust their own work tomatch the progress and demands of others. Taskanalysis is needed to identify the task com-ponents that correspond to what a person canachieve or provide over a reasonable period oftime as well as to propose a way to combine andschedule the components to an overall whole.

2. Tasks become so complex that one person canno longer control or comprehend them. Thismay happen when the task becomes so large ortakes so long that a single person is unable tocomplete it (i.e., the transition from individualto collective tasks). It may also happen whenthe execution of the task depends on the useof technological artifacts and where the use ofthe artifact becomes a task in its own right(cf. below). This is the case, for instance, whenthe artifacts can function in an independent orsemiautonomous way (i.e., they begin partiallyto regulate themselves rather than passivelycarry out an explicit function under the user’scontrol).

3. A similar argument goes when technologyitself—machines—becomes so complex thatthe situation changes from simply being one ofusing the technology to one learning how tounderstand , master , or control the technology.In other words, being in control of the tech-nology becomes a goal in itself, as a means toachieve the original goal. Examples are drivinga car in contrast to riding an ordinary bicy-cle, using a food processor instead of a knife,using a computer (as in writing this chapter)rather than paper and pencil, and so on. Inthese cases, and of course also in cases of farmore complex work, use of the technology isno longer straightforward but requires prepa-ration and prior thought either by the person

who does the task or work or by those whoprepare tasks or tools for others. Task analysiscan in these cases be used to describe situationswhere the task itself is very complex because itinvolves interaction and dependencies with otherpeople. It can similarly be used to describe situ-ations where use of the technology is no longerstraightforward but requires mastery of the sys-tem to such a degree that not everyone can applyit directly as intended and designed.

In summary, task analysis became necessary whenwork changed from something that could be done byan unaided individual to something requiring the collec-tive efforts of either people or joint cognitive systems.Although collective work has existed since the begin-ning of history, its presence became more conspicuousafter the Industrial Revolution about 250 years ago. Inthis new situation the work of the individual becamea mere part of the work of the collective, and individ-ual control of work was consequently lost. The workerbecame part of a larger context, a cog in complex socialmachinery that defined the demands and constraints towork. One important effect of that was that people nolonger could work at a pace suitable for them and pausewhenever needed but instead had to comply with thepace set by others—and increasingly the pace set bymachines.

1.2 Artifacts and ToolsIt is common to talk about humans and machines, orhumans and technology, and to use expressions suchas human–machine systems or, even better, human–technology systems . In the context of tasks and taskanalysis, the term technological artifact , or simply arti-fact , will be used to denote that which is being applied toachieve a goal. Although it is common to treat comput-ers and information technology as primary constituentsof the work environment, it should be remembered thatnot all tools are or include computers, and task analysisis therefore far more than human–computer interaction.That something is an artifact means that it has beenconstructed or designed by someone, hence that itexpresses or embodies a specific intention or purpose.In contrast to that, a natural object does not have anintended use, but is the outcome of evolution—orhappenstance—rather than design. Examples of naturalobjects are stones used to hammer or break somethingand sticks used to poke for something.

A natural object may be seen as being instrumentalto achieve something, hence used for that purpose. Inthe terminology of Gibson (1979), the natural objectis perceived as having an actionable property (anaffordance), which means that it is seen as being usefulfor a specific purpose. An artifact is designed with aspecific purpose (or set of uses) in mind and shouldideally offer a similar perceived affordance. To theextent that this is the case, the design has been success-ful. Task analysis (i.e., describing and understandingin advance the uses of artifacts) is obviously one of theways in which that can be achieved. Although thereis no shortage of examples of failure to achieve this


noble goal, it is of course in the best interest of thedesigner—and the producer—to keep trying.

When a person designs or constructs something forhimself or herself an artifact or a composite activity,there is no need to ask what the person is capable of,what the artifact should be used for, or how it shouldbe used. But the need is there in the case of a singlebut complex artifact where the use requires a series ofcoordinated or ordered actions. It is also there in thecase of more complex, organized work processes wherethe activities or tasks of an individual must fit into alarger whole. Indeed, just as the designer of a complexartifact considers its components and how they mustwork together for the artifact to be able to provide itsfunction, so must the work process designer consider thecharacteristics of people and how they must collaborateto deliver the desired end product or result. It is, indeed,no coincidence that the first task analyses were made fororganized work processes rather than for the single usersworking with artifacts or machines.

From an analytical perspective, the person’s knowl-edge of what he or she can do can be seen as corre-sponding to the designer’s assumptions about the user,while the person’s knowledge of how the artifact shouldbe used can be seen as corresponding to the user’sassumptions about the artifact, including the designer’sintentions. As long as the artifact or the work processesare built around the person, there is little need to makeany of these assumptions explicit or indeed to producea formal description of them: The user and the designerare effectively the same person. There is also little needof prior thought or prior analysis since the developmentis an integral part of work rather than an activity thatis separated in time and space. But when the artifact isdesigned by one person to be used by someone else, thedesigner needs to be very careful and explicit in makingassumptions and to consider carefully what the futureuser may be able to do and will do. In other words, it isnecessary in these cases to analyze how the artifact willbe used or to perform a task analysis.

2 TASK TYPES AND TASK BREAKDOWN

Task analysis is in the main a collection of methodswhich describe (or prescribe) how the analysis will beperformed, preferably by describing each step of theanalysis as well as how they are organized. Each methodshould also describe the stop rule or criterion (i.e., definethe principles needed to determine when the analysis hascome to an end, for instance, that the level of elementarytasks has been reached).

An important part of the method is to name andidentify the main constituents of a task and how theyare organized. As described later in this chapter, taskanalysis has through its development embraced severaldifferent principles of task organization, of which themain ones are the sequential principle, the hierarchicalprinciple, and the functional dependency principle. Todo so, the method must obviously refer to a classifica-tion scheme or set of categories that can be used todescribe and represent the essential aspects of a task.

The hallmark of a good method is that the classificationscheme is applied consistently and uniformly, therebylimiting the opportunities for subjective interpretationsand variations. Task analysis should depend not onpersonal experience and skills but on generalized publicknowledge and common sense. A method is alsoimportant as a way of documenting how the analysishas been done and of describing the knowledge thatwas used to achieve the results. It helps to ensure thatthe analysis is carried out in a systematic fashion sothat it, if needed, can be repeated, hopefully leading tothe same results. This reduces the variability betweenanalysts and hence improves the reliability.

The outcome of the task analysis accounts for theorganization or structure of constituent tasks. A criticalissue is the identification or determination of elementaryactivities or task components. The task analysis servesamong other things to explain how something should bedone for a user who does not know what to do or whomay be unable to remember it (in the situation). Thereis therefore no need to describe things or tasks that theuser definitely knows. The problem is, however, howthat can be determined.

Task analysis has from the very start tried todemarcate the basic components. It would clearly beuseful if it was possible to find a set of basic tasks—oractivity atoms—that could be applied in all contexts.This is akin to finding a set of elementary processes orfunctions from which a complex behavior can be built.Such endeavors are widespread in the behavioral andcognitive sciences, although the success rate usually isquite limited. The main reason is that the level of anelementary task depends on the person as well as on thedomain. Even if a common denominator could be found,it would probably be at a level of detail as to have littlepractical value (e.g., for training or scheduling).

2.1 Changing Views of Elementary Tasks

Although the search for all-purpose task components isbound to fail, it is nevertheless instructive to take a brieflook at three different attempts to do so. Probably thefirst—and probably also the most ambitious—attemptwas made by Frank Bunker Gilbreth, one of the pioneersof task analysis. The categorization, first reported about1919, evolved from the observation by trained motion-and-time specialists of human movement, specificallyof the fundamental motions of the hands of a worker.Gilbreth found that it was possible to distinguishamong the following 17 types of motion: search, select,grasp, reach, move, hold, release, position, pre-position,inspect, assemble, disassemble, use, unavoidable delay,wait (avoidable delay), plan, and rest (to overcomefatigue). (The basic motions are known as therbligs ,using an anagram of the developer’s name.)

A more contemporary version is a list of typi-cal process control tasks suggested by Rouse (1981).This comprises 11 functions, which are in alphabeti-cal order: communicating, coordinating tasks, executingprocedures, maintaining, planning, problem solving, rec-ognizing, recording, regulating, scanning, and steering.In contrast to the therbligs , it is possible to organizethese functions in several ways: for instance, in relation


to an input–output model of information processing, inrelation to a control model, and in relation to a decision-making model. The functions proposed by Rouse arecharacteristically on a higher level of abstraction thanthe therbligs and refer to cognitive functions, or cogni-tive tasks, rather than to physical movements.

A final example is the GOMS model proposed byCard et al. (1983). The purpose of GOMS, which isan acronym that stands for “goals, operators, methods,and selection rules,” was to provide a system formodeling and describing human task performance.Operators, one of the four components of GOMS,denote the set of atomic-level operations from which auser can compose a solution to a goal, while methodsrepresent sequences of operators grouped together toaccomplish a single goal. For example, the manualoperators of GOMS are: Keystroke key_name, Type_instring_of_characters, Click mouse_button, Double_clickmouse_button, Hold_down mouse button, Releasemouse_button, Point_to target_object, and Home_todestination. These operators refer not to physical tasks,such as the therbligs , but rather to mediating activitiesfor mental or cognitive tasks.

The definition of elementary tasks in scientificmanagement could comfortably refer to what peopledid, hence to what could be reported by independentobservers. The problem with defining elementary cog-nitive or mental tasks is that no such independent veri-fication is possible. Although GOMS was successful indefining elementary tasks on the keystroke level, it wasmore difficult to do the same for the cognitive or men-tal aspects (i.e., the methods and selection rules). Thephysical reality of elementary tasks such as grasp, reach,move, and hold has no parallel when it comes to cogni-tive functions. The problems in identifying elementarymental tasks are not due to a lack of trying. This hasindeed been a favorite topic of psychology from Donders(1969; orig. 1868) to Simon (1972). The problems comeabout because the “smallest” unit is defined by the the-ory being used rather than by intersubjective reality.In practice, this means that elementary tasks must bedefined relative to the domain at the time of the analy-sis (i.e., in terms of the context rather than as absolutesor context-free components).

3 BRIEF HISTORY OF TASK ANALYSIS

Task analysis has a relatively short history start-ing around the beginning of the twentieth century.The first major publications were Gilbreth (1911) andTaylor (1911), which introduced the principles of sci-entific management. The developments that followedreflected both the changing view of human nature,for instance, in McGregor’s (1960) theory X and the-ory Y, and the changes in psychological schools,specifically the models of the human mind. Of thethree examples of a classification system mentionedabove, Gilbreth (1911) represents the scientific man-agement view, Rouse (1981) represents the supervi-sory control view (human–machine interaction), andCard et al. (1983) represents the information-processing

view (human–computer interaction). These views canbe seen as alternative ways of describing the same real-ity: namely, human work and human activities. Onestandpoint is that human nature has not changed signifi-cantly for thousands of years and that different descrip-tions of the human mind and of work therefore onlyrepresent changes in the available models and concepts.Although this undoubtedly is true, it is also a fact thatthe nature of work has changed due to developments intechnology. Gilbreth’s description in terms of physicalmovements would therefore be as inapplicable to today’swork as a description of cognitive functions would havebeen in 1911.

3.1 Sequential Task AnalysisThe dawn of task analysis is usually linked to theproposal of a system of scientific management (Taylor,1911). This approach was based on the notion that tasksshould be specified and designed in minute detail andthat workers should receive precise instructions abouthow their tasks should be carried out. To do so, it wasnecessary that tasks could be analyzed unequivocally or“scientifically,” if possible in quantitative terms, so thatit could be determined how each task step should bedone in the most efficient way and how the task stepsshould be distributed among the people involved.

One of the classical studies is Taylor’s (1911)analysis of the handling of pig iron, where the work wasdone by men with no “tools” other than their hands.A pig-iron handler would stoop down, pick up a pigweighing about 92 pounds, walk up an inclined plank,and drop it on the end of a railroad car. Taylor and hisassociates found that a gang of pig-iron handlers wasloading on the average about 121/2 long tons per man perday. The aim of the study was to find ways in which toraise this output to 47 tons a day, not by making the menwork harder but by reducing the number of unnecessarymovements. This was achieved both by careful motion-and-time studies and by a system of incentives thatwould benefit workers as well as management.

Scientific management was based on four elementsor principles, which were used in studies of work.

1. The development of the science of work withrigid rules for each motion of every personand the perfection and standardization of allimplements and working conditions

2. The careful selection and subsequent trainingof workers into first-class people and theelimination of all people who refuse to or areunable to adopt the best methods

3. Bringing the first-class workers and the scienceof working together through the constant helpand watchfulness of management and throughpaying each person a large daily bonus forworking fast and doing what he or she is toldto do

4. An almost equal division of the work andresponsibility between workers and management

Of these four elements, the first (the developmentof the science of work) is the most interesting and


spectacular. It was essentially an analysis of a task intoits components, using, for example, the list of therbligsmentioned above. In the case of manual work this wasentirely feasible, since the task could be described as asingle sequence of more detailed actions or motions. Themotion-and-time study method was, however, unableto cope with the growing complexity of tasks thatfollowed developments in electronics, control theory,and computing during the 1940s and 1950s. Due tothe increasing capabilities of machines, people wereasked—and tasked—to engage in multiple activities atthe same time, either because individual tasks becamemore complex or because simpler tasks were combinedinto larger units. An important consequence of this wasthat tasks changed from being a sequence of activitiesreferring to a single goal to an organized set of activitiesreferring to a hierarchy of goals. The use of machinesand technology also became more prevalent, so thatsimple manual work such as pig-iron handling wastaken over by machines, which in turn were operatedor controlled by workers.

Since the use of technology has made work envi-ronments more complex, relatively few tasks today aresequential tasks. Examples of sequential tasks are there-fore most easily found in the world of cooking. Recipesare typically short and describe the steps as a simplesequence of actions, although novice cooks sometimesfind that recipes are underspecified. As an example of asequential task analysis, Figure 1 shows the process forbaking Madeleines.

3.2 From Sequential to Hierarchical TaskOrganization

The technological development meant that the natureof work changed from being predominantly manualand became more dependent on mental capabilities(comprehension, monitoring, planning). After a while,human factors engineering, or classical ergonomics,

Butter 18madeleine molds

Sift flour, semolina,and cornstarch

into a bowl

Whisk butter andicing sugar

together until paleand fluffy

Stir in the flourmixture using a

fork to form a softpaste

Press a little of themixture into eachmold and smooth

off the top

Bake in oven at180°C for 15–20

min

Leave to cool alittle in tins beforegently easing out

Begin

Figure 1 Sequential task description.

recognized that traditional methods of breaking the taskdown into small pieces, where each could be performedby a person, were no longer adequate. Since the natureof work had changed, the human capacity for processinginformation became decisive for the capacity of thehuman–machine system. This capacity could not beextended beyond its “natural” upper limit, and it soonbecame clear that the human capacity for learningand adaptation was insufficient to meet technologicaldemands.

To capture the more complex task organization,Miller (1953) developed a method for human–machinetask analysis in which main task functions could bedecomposed into subtasks. Each subtask could thenbe described in detail, for instance, by focusing oninformation display requirements and control actions.This led to the following relatively simple and informalprocedure for task analysis:

1. Specify the human–machine system criterionoutput.

2. Determine the system functions.3. Trace each system function to the machine

input or control established for the operator toactivate.

4. For each respective function, determine whatinformation is displayed by the machine tothe operator whereby he or she is directed toappropriate control activation (or monitoring)for that function.

5. Determine what indications of response ade-quacy in the control of each function will befed back to the operator.

6. Determine what information will be avail-able and necessary to the operator from thehuman–machine “environment.”

7. Determine what functions of the system must bemodulated by the operator at or about the sametime, or in close sequence, or in cycles.

8. In reviewing the analysis, be sure that eachstimulus is linked to a response and that eachresponse is linked to a stimulus.

The tasks were behavior groups associated with com-binations of functions that the operator should carry out.These were labeled according to the subpurpose they ful-filled within the system. Point 8 reflects the then-currentpsychological thinking, which was that of stimulus-and-response couplings. The operator was, in other words,seen as a transducer or a machine that was coupled tothe “real” machine. For the human–machine systemto work, it was necessary that the operator interpretthe machine’s output in the proper way and that he orshe respond with the correct input. The purpose of taskanalysis was to determine what the operator had to do toenable the machine to function as efficiently as possible.

3.2.1 Task–Subtask RelationThe task–subtask decomposition was a significantchange from sequential task analysis and was neces-sitated by the growing complexity of work. The


development was undoubtedly influenced by the emerg-ing practice—and later science—of computer pro-gramming, where one of the major innovations wasthe subroutine. Arguably the most famous example ofa task–subtask relation is the TOTE (test–operate–test–exit), which was proposed as a building block ofhuman behavior (Miller et al., 1960). This introducedinto the psychological vocabulary the concept of a plan,which is logically necessary to organize combinations oftasks and subtasks. Whereas a subroutine can be com-posed of motions and physical actions, and hence inprinciple can be found even in scientific management, aplan is obviously a cognitive or mental component. Thevery introduction of the task–subtask relation, and ofplans, therefore changed task analysis from describingonly what happened in the physical world to describingwhat happened in the minds of the people who carriedout the work.

Miller’s task–subtask analysis method clearlyimplied the existence of a hierarchy of tasks and sub-tasks, although this was never a prominent feature of themethod. As the technological environments developedfurther, the organization of tasks and subtasks becameincreasingly important for task analysis, culminatingwith the development of hierarchical task analysis(HTA) (Annett and Duncan, 1967; Annett et al., 1971).Since its introduction HTA has become the standardmethod for task analysis and task description and iswidely used in a variety of contexts, including interfacedesign.

The process of HTA is to decompose tasks intosubtasks and to repeat this process until a level ofelementary tasks has been reached. Each subtask oroperation is specified by its goal, the conditions underwhich the goal becomes relevant or “active,” the actionsrequired to attain the goal, and the criteria that mark theattainment of the goal. The relationship between a setof subtasks and the superordinate task is governed byplans expressed as, for instance, procedures, selectionrules, or time-sharing principles. A simple example ofHTA is a description of how to get money from abank account using an ATM (see Figure 2). In thisdescription, there is an upper level of tasks (marked, 1,2, 3), which describe the order of the main segments, and

a lower level of subtasks (marked 1.1, 1.2, etc.), whichprovide the details. It is clearly possible to break downeach of the subtasks into further detail, for instance, bydescribing the steps comprised by 1.2 Enter PIN code.This raises the nontrivial question of when the HTAshould stop (i.e., what the elementary subtasks or taskcomponents are; cf. below).

The overall aim of HTA is to describe a task insufficient detail, where the required level of resolutiondepends on the specific purposes (e.g., interactiondesign, training requirements, interface design, riskanalysis). HTA can be seen as a systematic searchstrategy adaptable for use in a variety of differentcontexts and purposes within the field of human factors(Shepherd, 1998). In practice, performing HTA com-prises the following steps. (Note, by the way, that this isa sequential description of hierarchical task analysis!)

1. Decide the purpose of the analysis.2. Get agreement between stakeholders on the

definition of task goals and criterion measures.3. Identify sources of task information and select

means of data acquisition.4. Acquire data and draft a decomposition table or

diagram.5. Recheck the validity of the decomposition with

the stakeholders.6. Identify significant operations in light of the

purpose of the analysis.7. Generate and, if possible, test hypotheses con-

cerning factors affecting learning and perfor-mance.

Whereas Miller’s description of human–machinetask analysis concentrated on how to analyze therequired interactions between humans and machines,HTA extended the scope to consider the context of theanalysis, in particular which purpose it served. Althoughthis was a welcome and weighty development, it leftthe actual HTA somewhat underspecified. Indeed, in thedescription above it is only the fourth step that is theactual task analysis.

Begin

1. Preparetransaction

1.1 Insertcredit card

1.2 EnterPIN code

1.3 Selecttype of

transaction

3.1 Removecredit card

3.2 Removemoney

2. Enteramount

3. Completetransaction

Figure 2 Hierarchical task description.


3.2.2 Tasks and Cognitive Tasks

In addition to the change that led from sequentialmotion-and-time descriptions to hierarchical task orga-nization, a further change occurred in the late 1980s toemphasize the cognitive nature of tasks. The need toconsider the organization of tasks was partly a conse-quence of changing from a sequential to a hierarchi-cal description, as argued above. The changes in thenature of work also meant that “thinking” tasks becamemore important than “doing.” The need to understandthe cognitive activities of the human–machine system,first identified by Hollnagel and Woods (1983), soondeveloped a widespread interest in cognitive task analy-sis, defined as the extension of traditional task analysistechniques to yield information about the knowledge,thought processes, and goal structures that underlieobservable task performance (e.g., Schraagen et al.,2000). As such, it represents a change in emphasis fromovert to covert activities. An example is the task anal-ysis principles described by Miller et al. (1960), whichrefer to mental actions as much as to motor behavior.Since many tasks require a considerable amount of men-tal functions and effort, in particular in retrieving andunderstanding the information available and in planningand preparing what to do (including monitoring of whathappens), much of what is essential for successful per-formance is covert. Whereas classical task analysis reliesvery much on observable actions or activities, the needto find out what goes on in other peoples’ minds requiresother approaches.

One consequence of the necessary extension of taskanalysis from physical to cognitive tasks was the realiza-tion that both the physical and the cognitive tasks wereaffected by the way the work situation was designed.Every artifact we design has consequences for howit is used. This goes for technological artifacts (gad-gets, devices, machines, interfaces, complex processes)as well as social artifacts (rules, rituals, procedures,social structures and organizations). The consequencescan be seen in the direct and concrete (physical) inter-action with the artifact (predominantly manual work) aswell as in how the use of the artifact is planned andorganized (predominantly cognitive work). Introducinga new “tool” therefore affects not only how work isdone but also how it is conceived of and organized. Yetinterface design and instruction manuals and procedurestypically describe how an artifact should be used butnot how we should plan or organize the use of it eventhough the latter may be affected as much—or evenmore—than the former. The extension of task analysisto cognitive task analysis should therefore be matchedby a corresponding extension of task design to cognitivetask design (Hollnagel, 2003).

3.2.3 Elementary Task

All task analysis methods require an answer to whatthe elementary task is. As long as task analysis wasoccupied mainly with physical work, the question couldbe resolved in a pragmatic manner. But when taskanalysis changed to include the cognitive aspects ofwork, the answer became more contentious. This is

obvious from the simple example of a HTA shownin Figure 2. For a person living in a developed orindustrialized society, entering a PIN code can beassumed to be an elementary task. It can neverthelessbe broken down into further detail by, for example,a motion-and-time study or a GOMS-type interactionanalysis. The determination of what an elementary taskis clearly cannot be done separately from assumptionsabout who the users are, what the conditions of use (orwork) are, and what the purpose of the task analysisis. If the purpose is to develop a procedure or a setof instructions such as the instructions that appear onthe screen of an ATM, there may be no need to gofurther than “enter PIN code” or possibly “enter PINcode and press ACCEPT.” Given the population ofusers, it is reasonable for the system designer to take forgranted that they will know how to do this. If, however,the purpose is to design the physical interface itselfor to perform a risk analysis, it will be necessary tocontinue the analysis at least one more step. GOMS is agood example of this, as would be the development ofinstructions for a robot to use an ATM.

In the contexts of work, assumptions about elemen-tary tasks can be satisfied by ensuring that users havethe requisite skills (e.g., through training and instruc-tion). A task analysis may indeed be performed withthe explicit purpose of defining training requirements.Designers can therefore, in a sense, afford themselvesthe luxury of dictating what an elementary task is as longas the requirements can be fulfilled by training. In thecontext of artifacts with a more widespread use, typ-ically in the public service domain, greater care mustbe taken in making assumptions about an elementarytask, since users in these situations often are “acciden-tal” (Marsden and Hollnagel, 1996).

3.3 Functional Dependency and Goals–MeansTask Analysis

Both sequential and hierarchical task analyses arestructural in the sense that they describe the order inwhich the prescribed activities are to be carried out.A hierarchy is by definition the description of howsomething is ordered, and the very representation ofa hierarchy (as in Figure 2) emphasizes the structure.As an alternative, it is possible to analyze and describetasks from a functional point of view (i.e., in termsof how tasks relate to or depend on each other). Thischanges the emphasis from how tasks and activities areordered to what the tasks and activities are supposed toachieve.

Whereas task analysis in practice stems from thebeginning of the twentieth century, the principle offunctional decomposition can be traced back at least toAristotle (Book III of the Nicomachean Ethics). Thisis not really surprising, since the focus of a functionaltask analysis is the reasoning about tasks rather thanthe way in which they are carried out (i.e., the physicalperformance). Whereas the physical nature of tasks haschanged throughout history, and especially after thebeginning of the Industrial Revolution, thinking abouthow to do things is largely independent of how thingsare actually done.


In relation to task analysis, functional dependencymeans thinking about tasks in terms of goals and means.The strength of a goals–means, or means–ends, decom-position principle is that it is ubiquitous, important, andpowerful (Miller et al., 1960, p. 189). It has thereforebeen used widely, most famously as the basis for theGeneral Problem Solver (Newell and Simon, 1961).

The starting point of a functional task analysis isa goal or an end, defined as a specified conditionor state of the system. A description of the goalusually includes or implies the criteria of achievementor acceptability (i.e., the conditions that determine whenthe goal has been reached). To achieve the goal, certainmeans are required. These are typically one or moreactivities that need to be carried out (i.e., a task). Yetmost tasks are possible only if specific conditions arefulfilled. For instance, you can work on your laptoponly if you have access to an external power sourceor if the batteries are charged sufficiently. When theseconditions are met, the task can be carried out. If not,bringing about these preconditions becomes a new goal,denoted a subgoal. In this way goals are decomposedrecursively, thereby defining a set of goal–subgoaldependencies that also serves to structure or organizethe associated tasks.

An illustration of the functions or tasks neededto start up an industrial boiler is shown in Figure 3

(see Lind and Larsen, 1995). The diagram illustrateshow the top goal, “St1 established,” requires that anumber of conditions have been established, whereeach of these in turn can be described as subgoals.Although the overall structure is a hierarchical orderingof goals and means, it differs from a HTA because thecomponents of the diagram are goals rather than tasks.The goals–means decomposition can be used as a basisfor identifying the tasks that are necessary to start theboiler, but this may not necessarily fit into the samerepresentation.

4 PRACTICE OF TASK ANALYSIS

As already mentioned, task analysis can be used fora variety of purposes. Although the direct interactionbetween humans and computers got the lion’s share ofattention in the 1990s, task analysis is necessary forpractically any aspect of a human–machine system’sfunctioning. Task analysis textbooks, such as Kirwanand Ainsworth (1992), provide detailed information andexcellent descriptions of the many varieties of taskanalysis. More recent works, such as Hollnagel (2003),extend the scope from task analysis to task design,emphasizing the constructive use of task knowledge.Regardless of which method an investigator decides to

St1established

St1 enabled

G4 achieved

St2established

St2 enabled FW present

MM2

Heat fromburners Pressure ok

Tr16established

CM1

Boiler drainsclosed

MM1

Level ofwater ok

Always

CM2

FW systemstarted

FW valveopened

Tr16enabled

Figure 3 Goals–means task description.


use, there are a number of general aspects that deserveconsideration.

4.1 Task Data Collection Techniques

The first challenge in task analysis is to know whererelevant data can be found and to collect them. Thebehavioral sciences have developed many ways ofdoing this, such as activity sampling, critical incidenttechnique, field observations, questionnaire, structuredinterview, and verbal protocols. In many cases, datacollection can be supported by various technologies,such as audio and video recording, measurements ofmovements, and so on, although the ease of mechanicaldata collection often is offset by the efforts needed toanalyze the data.

As task analysis extended its scope from physicalwork to include cognitive functions, methods were need-ed to get data about the unobservable parts of a task.The main techniques used to overcome this were “think-aloud” protocols and introspection (i.e., extrapolatingfrom one’s own experience to what others may do).The issue of thinking aloud has been hotly debated,as has the issue of introspection (Nisbett and Wilson,1977). Other structured techniques rely on controlledtasks, questionnaires, and so on. Yet in the end theproblem is that of making inferences from some set ofobservable data to what goes on behind, in the senseof what is sufficient to explain the observations. Thisraises interesting issues of methods for data collection tosupport task analysis and leads to an increasing relianceon models of the tasks. As long as task analysis is basedon observation of actions or performance, it is possibleto establish some kind of objectivity or intersubjectiveagreement or verification. As more and more of thedata refer to the unobservable, the dependence oninterpretations, and hence on models, increases.

4.2 Task Description Techniques

When the data have been collected, the next challengeis to represent them in a suitable fashion. It is importantthat a task analysis represent the information about thetask in a manner that can easily be comprehended.For some purposes, the outcome of a task analysismay simply be rendered as a written description of thetasks and how they are organized. In most cases this issupplemented by some kind of graphical representationor diagram, since this makes it considerably easier tograsp the overall relations. Examples are the diagramsshown in Figures 1–3. Other staple solutions are chart-ing and networking techniques, decomposition methods,HTA, link analysis, operational sequence diagrams(OSDs), and timeline analyses.

4.3 Task Simulation Methods

For a number of other purposes, such as those thathave to do with design, it is useful if the task canbe represented in other ways, specifically as some kindof description or model that can be manipulated. Thebenefit is clearly that putative changes to the task can beimplemented in the model and the consequences can beexplored. This has led to the development of a range of

methods that rely on some kind of symbolic model of thetask or activity, going from the production rule systemsto task networks (e.g., Petri nets). This developmentoften goes hand in hand with user models (i.e., symbolicrepresentation of users that can be used to simulateresponses to what happens in the work environment).In principle, such models can carry out the task asspecified by the task description, but the strength of theresults depends critically on the validity of the modelassumptions. Other solutions, which do not require theuse of computers, are mock-ups, walk-throughs, andtalk-throughs.

4.4 Task Behavior Assessment Methods

Task analyses are in many cases used as a starting pointto look at a specific aspect of the task execution, usuallyrisk or consequences for system safety. One specific typeof assessment looks at the possibility for humans to carryout a task incorrectly (i.e., the issue of “human error”and human reliability). Approaches to human reliabilityanalysis that are based on structural task descriptionsare generally oversimplified not only because humansare not machines but also because there is an essentialdifference between described and effective tasks orbetween “work as imagined” and “work as done.” Taskdescriptions in the form of event trees or as proceduralprototype models represent an idealized sequence orhierarchy of steps. Tasks as they are carried out or asthey are perceived by the person are more often seriesof activities whose scope and sequence are adjusted tomeet the demands—perceived or real—of the currentsituation (Hollnagel, 2010a). It can be argued that taskdescriptions used for risk and reliability analyses on thewhole are inadequate and unable to capture the realnature of human work. The decomposition principle hasencouraged—or even enforced—a specific form of taskdescription (the event tree), and this formalism has beenself-sustaining. It has, however, led human reliabilityanalysis into a cul-de-sac.

4.5 Future of Task Analysis

We started this chapter by pointing out that task analysisis the study of who does what and why , where thewho should be broadened to include individual work,collective work, and joint cognitive systems. The futureof task analysis is bright in the sense that there willalways be a practical need to know how things shouldbe done. The question is whether task analysis as it iscurrently practiced is capable of meeting this need inthe long run. There are several reasons why the replyneed not be unequivocally positive:

1. Task analysis has from the beginning been con-cerned mostly with individuals, whether as sin-gle workers or single users, despite the factthat most work involves multiple users (col-laboration, distributed work) in complex sys-tems (Hutchins, 1995). Although the importanceof distributed cognition and collective workis generally acknowledged, only few methodsare capable of analyzing that, over and above


Table 1 Tractable and Intractable Systems

Tractable System Intractable System

Number of details Descriptions are simple with few details Descriptions are elaborate with many detailsComprehensibility Principles of functioning are known Principles of functioning are partly unknownStability System does not change while being

describedSystem changes before description is completed

Relationship to othersystems

Independence Interdependence

Organization of tasks andactivities

Stable, work is highly regular and taskscan be prescribed

Unstable, work must be adjusted to match theconditions, tasks cannot be prescribed

representing explicit interactions such as in linkanalysis and OSDs.

2. Many task analysis methods are adequate fordescribing single lines of activity. Unfortu-nately, most work involves multiple threads andtimelines. Although HTA represents a hierar-chy of tasks, each subtask or activity is carriedout on its own. There is little possibility ofdescribing two or more simultaneous tasks, eventhough that is often what people have to copewith in reality. Another shortcoming is the dif-ficulty of representing temporal relations otherthan simple durations of activities.

3. There is a significant difference between des-cribed and effective tasks. Work in practice ischaracterized by ongoing adaptations and impro-visations rather than the straightforward carryingout of a procedure or an instruction. The reasonsfor this are that demands and resources rarelycorrespond to what was anticipated when thetask was developed and the actual situation maydiffer considerably from that which is assumedby the task description, thereby rendering thelatter unworkable.

The problem in a nutshell is that task analysis wasdeveloped to deal with linear work environments, whereeffects were proportional to causes and where order-liness and regularity on the whole could be assured.Sociotechnical systems have, however, since the 1980sbecome steadily more complex due to rampant techno-logical and societal developments. The scope of taskanalysis must therefore be extended in several direc-tions. A “vertical” extension is needed to cover the entiresystem, from technology to organization. A “horizontal”extension is needed to increase the scope to include bothdesign and maintenance. A second horizontal extensionis needed to include both upstream and downstream pro-cesses. The latter in particular means that previouslyseparate functions no longer can be treated as separate.There are important dependencies to what went before(upstream) and what comes after (downstream).

Today’s task analysis must therefore address systemsthat are larger and more complex than the systems ofyesteryear. Because there are many more details to con-sider, some modes of operation may be incompletelyknown, there are tight couplings among functions, andsystems may change faster than they can be described,

the net result is that many systems today are underspec-ified or intractable. For these systems it is clearly notpossible to prescribe tasks and actions in every detail.This means that performance must be variable or flex-ible rather than rigid. In fact, the less completely thesystem is described, the more performance variability isneeded.

It is useful to make a distinction between tractableand intractable systems (Hollnagel, 2010b). Tractablesystems can be completely described or specified, whileintractable systems cannot. The differences between thetwo types of systems are summarized in Table 1.

Most established safety methods have been devel-oped on the assumption that systems are tractable. Asthis assumption is no longer universally valid, it isnecessary to develop methods to deal with intractablesystems and irregular work environments. One way ofdoing that is to focus on which functions are requiredto achieve a goal and how they are organized rela-tive to the current situation (e.g., existing resourcesand demands). This can be seen as a natural con-tinuation of the development that has taken us fromsequential task analysis via hierarchical task analy-sis to functional dependency and goals–means anal-ysis. Doing so will have a major impact not onlyon how work situations are studied and analyzed butalso on how the efficiency and safety of work can beensured.

REFERENCES

Annett, J., and Duncan, K. D. (1967), “Task Analysis andTraining Design,” Occupational Psychology , Vol. 41,pp. 211–221.

Annett, J., Duncan, K. D., Stammers, R. B., and Gray, M. J.(1971), “Task Analysis,” Training Information Paper 6,Her Majesty’s Stationery Office, London.

Card, S., Moran, T., and Newell, A. (1983), The Psychologyof Human–Computer Interaction , Lawrence ErlbaumAssociates, Mahwah, NJ.

Daniellou, F. (2005), “The French-Speaking Ergonomists’ Ap-proach to Work Activity: Cross-Influences of FieldIntervention and Conceptual Models,” Theoretical Issuesin Ergonomics Science, Vol. 6, No. 5, pp. 405–427.

Diaper, D., and Stanton, N., Eds. (2003), The Handbook of TaskAnalysis for Human–Computer Interaction , LawrenceErlbaum Associates, Mahwah, NJ.


Donders, F. C. (1969), “On the Speed of Mental Processes,”Acta Psychologica , Vol. 30, pp. 412–431; translated from“Over de snelheid van psychische processen: Onder-zoekingen gedaan in het Physiologisch Laboratorium derUtrechtsche Hoogeschool, 1868–1869,” Tweede Reeks ,Vol. II, pp. 92–120.

Gibson, J. J. (1979), The Ecological Approach to Visual Per-ception , Lawrence Erlbaum Associates, Mahwah, NJ.

Gilbreth, F. B. (1911), Motion Study , Van Nostrand, Princeton,NJ.

Hollnagel, E., Ed. (2003), Handbook of Cognitive Task Design ,Lawrence Erlbaum Associates, Mahwah, NJ.

Hollnagel, E. (2010a). The ETTO Principle: Efficiency-Thor-oughness Trade-off. Why Things That Go Right SometimesGo Wrong , Ashgate, Farnham, UK.

Hollnagel, E., Ed. (2010b), Safer Complex Industrial Environ-ments , CRC Press, Boca Raton, FL.

Hollnagel, E., and Woods, D. D. (1983), “Cognitive SystemsEngineering: New Wine in New Bottles,” InternationalJournal of Man–Machine Studies , Vol. 18, pp. 583–600.

Hollnagel, E., and Woods, D. D. (2005), Joint CognitiveSystems: Foundations of Cognitive Systems Engineering ,CRC Press, Boca Raton, FL.

Hutchins, E. (1995), Cognition in the Wild , MIT Press,Cambridge, MA.

Jastrzebowski, W. (1857), “Rys ergonomiji czyli Nauki o Pracy,opartej naprawdach poczerpnietych z Nauki Przyrody”[An Outline of Ergonomics or the Science of WorkBased on the Truths Drawn from the Science of Nature],Przyoda i Przemysl , Vol. 29, pp. 227–231.

Kirwan, B., and Ainsworth, L. K., Eds. (1992), A Guide to TaskAnalysis , Taylor & Francis, London.

Leplat, J. (1991), “Organization of Activity in CollectiveTasks,” in Distributed Decision Making: Cognitive Mod-els for Cooperative Work , J. Rasmussen, B. Brehmer, andJ. Leplat, Eds., Wiley, Chichester, West Sussex, England.

Lind, M., and Larsen, M. N. (1995), “Planning Supportand the Intentionality of Dynamic Environments,” inExpertise and Technology: Cognition and Human–Computer Interaction , J. M. Hoc, P. C. Cacciabue,and E. Hollnagel, Eds., Lawrence Erlbaum Associates,Mahwah, NJ.

Marsden, P., and Hollnagel, E. (1996), “Human Interaction withTechnology: The Accidental User,” Acta Psychologica ,Vol. 91, pp. 345–358.

McGregor, D. (1960), The Human Side of Enterprise,McGraw-Hill, New York.

Miller, G. A., Galanter, E., and Pribram, K. H. (1960), Plansand the Structure of Behavior , Holt, Rinehart & Winston,New York.

Miller, R. B. (1953), “A Method for Man–Machine TaskAnalysis,” Technical Report 53–137, Wright Air ForceDevelopment Center, Dayton, OH.

Newell, A., and Simon, H. A. (1961), “GPS: A ProgramThat Simulates Human Problem-Solving,” in Proceedingsof a Conference on Learning Automata , TechnischeHochschule, Karlsruhe, Germany, April 11–14.

Nisbett, R. E., and Wilson, T. D. (1977), “Telling MoreThan We Can Know: Verbal Reports on Processes,”Psychological Review , Vol. 74, pp. 231–259.

Rouse, W. B. (1981), “Human–Computer Interaction in theControl of Dynamic Systems,” ACM Computing Survey ,Vol. 13, No. 1, pp. 71–99.

Schraagen, J. M., Chipman, S. F., and Shalin, V. L., Eds.(2000), Cognitive Task Analysis , Lawrence ErlbaumAssociates, Mahwah, NJ.

Shepherd, A. (1998), “HTA as a Framework for Task Analysis,”Ergonomics , Vol. 41, No. 11, pp. 1537–1552.

Simon, H. A. (1972), The Sciences of the Artificial , MIT Press,Cambridge, MA.

Taylor, F. W. (1911), The Principles of Scientific Management ,Harper, New York.

handbook of human factors and ergonomics (salvendy/handbook of human factors 4e) || task analysis:...

Documents