Borrower: USD Call#: BF311 .H3345 1986

Lending String: NOS,*GZV,WIS,MZF,EM2 Location: Oversize (3rd Floor)

Patron: Schieber. Frank iiiiiiii !!!!!!! iiiiiiii !!!!!!! Journal Title: Handbook of perception and human ODYSSEY ENABLED performance I ~ ~ - Charge iiiiiiii Volume: 2 Issue: Maxcost: 0.00 -(!) === Month/Year: 1986Pages: 1-49 +titlepage/verso - Shipping Address: - !!!!!!! -> -iiiiiiii Article Author: ·-(!) UNIVERSITY OF SOUTH DAKOTA ...... - USD LIBRARIES ILL ~ Article Title: Gopher, D. & Donchin, E; Chapter -iiiiiiii 414 E. CLARK STREET 0-. !!!!!!! 41. Workload-An examination of the concept. VERMILLION, SD 57069-2390 -I ...... ~ N co Fax: :::J N Imprint: New York ; Wiley, c1986. Ariel: 192.236.50.50 Mco Email: [email protected] ~ ILL Number: 85467209 z 1- llllllllllllllllllllllllllllllllllllllllllllllllll '0cu :J ....I

--~ --- --·

NOTICE: WARNING I CONCERNING COPYRiGHT RESTRiCTIONS

• The copyright law of the United States (Title 17. C nited States Cod~) govems the making ofphotocopiL's or other reproductions of copyrighted matenaL

• Under certain conditions speci lied in the law, libraries and archivt:s are authorized to fum ish a pholm.:opy or other reproductiorl. One of these speci fie "fair use" condition~ is that the photocopy or reproduction is nol to be "u:scJ for any purpose other than private study, scholarship, or research." [f u user makes a request tor, or tater uses, a photocopy or reproduction for purposes in excess of "fair use," that user may be liable for copyright infringement.

I HANDBOOK OF PERCEPTION AND HUMAN PERFORMANCE

VOLUME II Cognitive Processes and Performance

Editors: N/\51\ KENNETH R. BOFF Armstrong Aerospace Medical Research Laboratory LLOYD KAUFMAN New York University

JAMES P. THOMAS University of California at Los Angeles jl

.. A Wiley-lnterscience Publication ~ JOHN WILEY AND SONS New York • Chichester • Brisbane • Toronto • Singapore CHAPTER 41

WORKLOAD-AN EXAMINATION OF THE CONCEPT

DANIEL GOPHER Faculty of Industrial Engineering and Management, Israel Institute of Technology, The Technion, Haifa, Israel

EMANUEL DONCHIN Department of Psychology, University of Illinois, Champaign, Illinois

CONTENTS 2.4.3. The Breakdown ofthe Single Channel, 41-12 2.5. The Architecture of the Central Processor, 41-13 1. Introduction 41-2 2.6. Energy Constraints on Processing Capabilities, I 1.1. Task Difficulty and Workload, 41-2 41-14 1.2. Workload and the Limitations on Performance, 2.6.1. The Single-Resource Model, 41-14 41-3 2.6,.2. Multiple-Resource Models, 41-16 1.3. The Logical Status of Workload, 41-3 2.7. Controlled and Automatic Processes, 41-18 1.3.1. A Definition of Workload, 41-3 2.8. The Nature of Capacity Limitations, 41-19 1.3.2. Workload as an Intervening Variable, 41-4 2.8.1. Dimensions of a Load Profile, 41-19 1.3.3. Workload as a Hypothetical Construct, 41-4 2.8.2. Conscious Control and Allocation Policy, 1.4. Workload and Attention, 41-4 41-20 2. The Limitations of the Central Processor: Historical Background 41-5 3. Techniques for the Measurement of Workload 41-21 2.1. Origins of the Notion of a Central Limited 3.1. Criteria for the Evaluation of Workload Processor, 41-5 Measures, 41-21 2.1.1. The Human as an Information Channel, 3.2. Workload as a Property ofthe~perator-Task 41-5 Loop, 41-21 2.1.2. The Putative Advantages of Information 3.2.1. Normative and Descriptive Approaches, Theory, 41-6 41-21 2.2. Channel Capacity as a Measure of Limits on the 3.2.2. Overview of Section, 41-22 Central Processor, 41-6 3.3. Subjective Measures, 41-22 2.2.1. Absolute Judgment, 41-6 3.3.1. Consistency of Subjective Estimates, 41-23 2.2.2. The Span of Short-Term Memory, 41•7 3.3.2. Theoretical·Considerations in the 2.3. Summary, 41-8 Interpretation and Use of Subjective 2.4. Bottleneck Models of Human Limitations, 41-8 Measures of Workload, 41-23 2.4.1. Single-Bottleneck Models, 41-9 3.3.3. Methodological Considerations, 41-24 2.4.2. Welford's Single-Channel Model, 41-10 3.4. Performance Measures, Primary Task, 41-24 41-1 41-2 HUMAN PERFORMANCE

3.5. Arousal Measures, 41-25 difficulty is, ofcourse, obvious. It is easier to read Agatha Christie 3.6. Specific Measures, 41-25 than James Joyce; it is more difficult to fly an airplane than to 3.6.1. Selection of a Secondary Task, 41-26 operate a washing machine. It is also clear that the same task 3.6.2. Types of Interference and Lack of will prove more difficult to some individuals than to others. Interference, 41-26 Furthermore, the ease with which the same individual performs 3.6.3. The Problems of Concurrency, 41-27 a given task on different occasions may vary. Yet, despite the 3.6.4. Allocation Policy, 41-28 apparent simplicity of the concept and the fairly easy judgments 3.7. Performance Operating Characteristics (POC), observers can make regarding the difficulty of tasks, the mea­ 41-28 surement of task difficulty is in itself a rather difficult task. 3.8. Basic Assumptions of the POC Methodology, 41-31 The problem exists because the difficulty of any task can 3.9. Psychophysiological Measures of Workload, 41-32 be inferred not directly from its physical (or "structural") de­ 3.9.1. Introductory Comments on the P300 scription, but rather from the interaction between task and Component, 41-32 operator. Hence, system designers need to be able to discriminate 3.9.2. The Latency of the P300 Component, 41-33 among alternate design options in favor of those that will ease 3.9.3. P300 and Perceptual/Central Processing the operator's task. On occasion, system managers must be able Resources, 41-35 to monitor variations in the difficulty a task presents to an 3.9.4. P300 and Resource Reciprocity, 41-40 operator actively using a system. It is on such occasions that 3.9.5. Summary and Conclusions, 41-41 the need to measure ''task difficulty" arises. And it is on such 4. Epilogue 41-42 occasions that the apparent simplicity of the measurement task is revealed as horrendously complex. Acknowledgment 41-43 Superficially the measurement task appears quite straight­ Reference Note 41-43 forward. One simply asks an operator whether the task is difficult and uses the subjective response as an index of difficulty. As References 41-44 discussed in Section 3.3 (and as O'Donnell and Eggemeier note in Chapter 42), such subjective reports have serious limitations. An operator is often an unreliable and invalid measuring in­ strument. Neither can it be assumed that the quality of per­ formance is a good measure of the difficulty of a task. People 1. INTRODUCTION often cope with an increase in task difficulty by increasing mental l and physical effort devoted to the task so that performance may ·~t 1.1. Task Difficulty and Workload remain stable despite a great increase in difficulty. The mea­ ·w"' surement of difficulty must capture the interaction between l The study and measurement of mental workload absorb sub­ these relevant variables. Workload is the label assigned to this stantial energy, resources, and effort. Major conferences devoted interactive feature. to the analysis and measurement of workload (Moray, 1979; An analogy might clarify the matter. Consider a simple Frazier & Crombie, 1982) follow a fairly uniform course. The resistive circuit in which a voltage source, say, a batter, is concept of workload is examined, attempts at a definition are imposing a voltage across some resistor. Both the voltage sup­ made, and the usual conclusion is that workload is a multidi­ plied by the battery and the resistance that characterizes the mensional, multifaceted concept that is difficult to define. It is resistor depend on the properties of these devices regardless, generally agreed that attempts to measure workload relying to a first approximation, ofthe circuit in which they are embedded on a single representative measure are unlikely to be of use. (though both voltage and resistance depend on such circum­ The conferees generally proceed to examine many different stances as the ambient temperature). However, specifying the procedures to measure and analyze workload. Practitioners tend voltage or the resistance alone will tell us very little about the to imply that the measure or class of measures they advocate circuit. There is, however, another variable that actually cap­ is uniquely preferable to many of the other proposed measures. tures the properties of the circuit, reflecting the interaction Persuasive arguments are presented for the unique suitability between the voltage and the resistance. The current flowing in of (1) subjective measures of workload (Sheridan, 1980), (b) the circuit is determined jointly by the properties of the battery secondary-task procedures (Jex, ~976; Pew, 1979), and (3) phys­ and those of the resistor and it is therefore a parameter of the iological measures (Donchin, 1979; Mulder, 1979). These writers circuit rather than of its individual elements. Variations in the are actually quite correct in holding to their particular views, current, given fixed voltage and resistance, can also be used as since the measures described often yield an interesting set of a means for assessing the degree to which the ambient conditions relations and in some circumstances seem to meet the practical affect the properties of circuit elements. needs of design engineers. We suggest that workload is analogous to current. Of course, O'Donnell and Eggemeier (Chapter 42) provide a survey, unlike the case with current, we cannot derive the workload written from the perspective of the design engineer, organizing associated with the interaction between a specific operator and the many techniques proposed for measuring workload. As they a specific task from an equation that combines measurable at­ indicate in the opening paragraphs of their chapter, the interest tributes of the operator with measurable attributes of the task in workload arises in evaluating "task difficulty." The assessment to yield workload. The intent of this chapter is to review the of workload is a direct assessment of a class of difficulties that search for relevant task and subject attributes as well as the operators confront when performing an lli!Signed task. The need combination rules that may be applied to these measurable for this concept, and for the vast literature devoted to its analysis variables so as to provide a measure of workload. However, it and measurement, is generated by the complexity of the de­ is important to emphasize at the outsetthat we view workload ceptively simple concept difficulty. That tasks vary in their as an attribute of the interaction between a person and a task.

~ WORKLOAD-AN EXAMINATION OF THE CONCEPT 41-3 The concept is introduced, we shall argue, as a hypothetical suggest that, for reasons still to be determined, the demands construct to summarize the difficulty that a task presents to an imposed on the operator by the task have become larger than operator. This review will therefore be concerned with two main they were when performance was at its best. It is at this the­ topics: (1) the manner in which the task~perator loop can be oretical juncture that the concept of workload is normally in­ modeled to derive useful explications of the concept of workload; troduced. That is, workload is invoked to account for those aspects and (2) the theoretical status of the different modes of mea­ of the interaction between a person and a task that cause task surement employed in the assessment of workload. demands to exceed the person's capacity to deliver. Note that the concept is needed only for those cases in which the required 1.2. Workload and the Limitations on Performance performance of a sp~cified quality is clearly within the per­ former's current repertoire. In this discussion we are ignoring, The term workload is used to describe aspects of the interaction in keeping with the tradition in discussions of workload, the between an operator and an assigned task. Tasks are specified effects ofleaming on an individual's repertoire. We are concerned in terms of their structural properties; a set of stimuli and with the deployment of the available responses at the time a responses are specified with a set of rules that map responses task was assigned without considering how the repertoire has to stimuli. There are, in addition, expectations regarding the been created. Subsequent training may well change the rep­ quality of the performance, which derive from knowledge of ertoire. Indeed, in Section 2. 7 we do di8CUSS the effects of practice the relation between the structure of the task and the nature on workload, but we restrict our interest to the improvement of human capacities and skills. Expectations may also be based in the performance of a previously acquired skill rather than on the individual operator's past performance or on knowledge the acquisition of new skills. of the way others perform similar tasks. Frequently, these ex­ Workload seems to refer to a cost the operator incurs as pectations are not met even though the individual is motivated tasks are performed. In principle, if the capacity to perform is to accept the assignment and intends to perform according to available, a failure to perform must imply limits to its use. In expectations. Often we ascribe such failures in performance to the same way that physical workload was used by ergonomists increased difficulty of the task. It is in the attempt to explain to specify limits on the muscles' ability to deliver, so is mental and cope with these interactions that the concept of workload workload invoked to explain the mind's inability to deliver. finds its primary use. Because muscular work is directly observable, its physiology Often a failure to perform is easy to explain: The task was and mechanisms are relatively well understood. But mental evidently beyond the capacities of the organism. Ifone required workload is clearly an attribute of the information processing a human to jump from one rim of the Grand Canyon to the and control systems that mediate between stimuli, rules, and other, it is unlikely that the concept of workload would be invoked responses. Mental workload is an attribute of the person-task to explain an inability to perform the task. One would likely loop, and the effects of workload on human performance can say that this task is beyond human ability. In the terminology therefore be examined only in relation to a model of human of the previous section, we do not attribute the failure to an information processing. The logical status of the concept depends interaction between the operator and the task. Similarly, it is on one's stance regarding cognition. The following section ex­ not particularly useful to ascribe to excessive workload the amines the mental workload as a concept within this framework. inability of a monolingual English-speaking student to read yesterday's conversation exercise in Latin. The 'teacher will, 1.3. The Logical Status of Workload no doubt, assume that the failure in performance indicates that the student has never learned, and therefore lacks the capacity, 1.3.1. A Definition of Workload. The current concept of to perform the task. In other words, there are many occasions workload implies that limitations exist in the information pro­ in which a failure to perform according to expectations is at­ cessing structures, making it difficult for a person to fully use tributed to very obvious limitations on the capacities, either the information processing apparatus in the service of the target innate or learned, of the organism rather than to the interaction task. The key assumption is that to perform a task the organism between the organism and the task. In such cases there is no uses effectors through which the overt responses are made as need to invoke workload as an explanatory concept. well as sensors through which information is gathered. The However, often the limitations are not obvious. Ifthe student path between sensors and effectors is an elaborate information has read the exercise many times in the past and is generally processing apparatus with structural properties and a limited known to be attentive in class and eager to please, a sudden capacity. These limitations on the capacity of the information failure to perform will force the teacher to assume that "other processing system must be measured and modeled if we are to factors" have intervened to limit performance. Let us assume, account for performance failures attributed to mental workload . .for simplicity, that motivation is never a problem in the context In other words, mental workload may be viewed as the of our discussion. The intention to perform is genuine, as is the difference between the capacities of the information processing deep desire to perform well. If this is clear, and if performance system that are required for task performance to satisfy per­ shows a decrement despite a demonstrated past ability to perform formance expectations and the capacity available at any given adequately, then we are inclined to interpret a failure to perform time. Task difficulty is thus manifested by a difference between according to expectations as evidence for deterioration in the the expected and the actual performance. It is necessary to specify performer's ability to execute the task that may be due to an who is doing the expecting ifthe term expected performance is increase in workload. to have precise meaning. The level of"expected" task perform­ The deterioration may be due to pathology. A stroke may ance in any constellation is established by the level of perform­ have incapacitated the subject. In such an event we shall again ance of the same task under the least demanding circumstances. ascribe the difficulty to the pathology without feeling bound to For example, if a person can listen to two conversations when invoke the concept of workload. On the other hand, if we know they are presented singly, failure to monitor both eoncurrently that there has been no overt pathology, then we may indeed · requires the invocation of workload. l 41-4 HUMAN PERFORMANCE This definition implies that workload is an intervening a precipice the walker is less likely to conduct a conversation variable rather than a hypothetical construct. MacCorquadale and seems oblivious of all surrounding activity. We may also and Meehl (1948) discussed the distinction between these two note that task performance is far more robust to changing con­ metatheoretical terms. An intervening variable is a theoretical ditions when the plank is at zero height than when it is up concept, such as electrical resistance, that is, higli. As we continue this description, we invoke observables ... simply a quantity obtained by a specified manipulation of that reflect differences in task performance. However, these the values of empirical variables; it will involve no hypothesis observables were not included in our initial definition of the as to the existence of unobserved entities or the occurrence of ta~k. Neither are we as sure when we invoke them that they unobserved processes; it will contain, in its complete statement will indeed catch the flavor of the difference we are sensing. for all purposes of theory and prediction no words which are not However, we allude precisely to the effect that raising the plank defined either explicitly or by reduction sentences in terms of has on task performance when we refer to mental workload. the empirical variables. (MacCorquadale & Meehl, 1948, p. 103) 1.3.3. Workload as a Hypothetical Construct. Workload 1.3.2. Workload as an Intervening Variable. If we could logically carries the· excess meaning that, according to elucidate functions that relate observed decrements in per­ MacCorquadale and Meehl (1948), characterizes hypothetical formance to the specified criteria} performance, these functions constructs~ This label is applieq to concepts that "involve terms would define workload in the same way that resistance is defined which are not wholly reducible to empirical tenns; they refer by Ohm's law. The system's inability to perform as desired to processes or entities that are not directly observable (although would be attributed to some internal property labeled workload they need not be in principle unobservable)" (p. 104). that resists the execution of the task as specified. Of course, in In this framework, workload is a concept like electron, rather the same way that current could be further analyzed and ex­ than like current; We imply that we are describing some entity plained in terms of atomic theory, so we would ultimately expect or some property of entities that is not given entirely by the to account for workload in cognitive and neurophysiological relationship between our empirical observations. At the same terms. But, even without this clarification, the concept can be time, we assume that. this excess meaning can be captured, employed usefully in theories of human performance and in studied, and measured in ways that would advance our under­ applications of these theories in engineering psychology. standing of the system and make it possible to use the concept However, the situation is a bit more complex. The fact is for practical activities. that on too many occasions the observations that compel us to invoke the concept of workload are inconsistent with its per­ 1.4. Workload and Attention ception as in intervening variable. Quite clearly it is often vir­ tually impossible to infer the degree to which a given task is By defining workload in terms of the limitations on the capacity beyond the capacity of a given individual on any occasion. Op­ of an information processing system, we are underlining the erators are seen to cope with extraordinary demands so that close affinity between the literature concerned with workload their observed performance displays no obvious variance. To and the literature that focuses on attention. In both bodies of define workload by direct observations on performance, we must literature, the prime concern ofinvestigators is to assess, and conclude that there are no changes in workload even though possibly explain, performance limitations that are manifested our common sense and subsequent analysis suggest otherwise. despite an apparent ability of the individual to perform a task. Consider, for example, a human presented with a plank Thus, the psychologist attending the proverbial cocktail party which is 1 foot wide and 20 feet long. It is solid and sturdy and (Cherry, 1957) and listening to at least two concurrent conver­ can, beyond doubt, carry the person's weight without breaking. sations is clearly capable of following either. Yet, one conver­ Assign the person the task of getting on the plank at its proximal sation may come to predominate at the expense of the other. end and walking to its distal end. Measuring task performance In fact, the content of one of the conversations is often ignored. in terms ofthe person's ability to walk the plank, score "1" for This failure to converse is remarkable, and it is related to our a successful traverse and "0" for a failure. Be even more careful discussion of workload because the person is clearly capable of and measure the time interval from the instant the plank is switching "attention" from one conversation to another de­ approached to the instant the task is completed. It is conceivable pending on the level of interest the ignored conversation prom­ that for an experienced person walking the plank will be achieved ises. with equal facility when the plank is laid on the floor, when it This ability to switch from task to task implies that the is suspended across two chairs so that it is 3 feet from the contending tasks are all within the person's repertoire when ground, and when it is suspended above the middle ring of a performed singly. The limitation, as in the case of workload, circus at 50 feet. Performance, as we defined it, may show ab­ appears to be in the system's capacity to deal with multiple solutely no variance. And yet, we know intuitively that the demands. As with workload, investigators have suggested that workload associated with each ofthe three tasks is quite different. the limit 9n attention reflects constraints inherent in the struc­ We bother to note that workloads are different because we ture and organization of a central processor. Our discussion of would like to know how much additional work we can impose workload begins by examining various attempts to use the con­ on the plank walker. We are persuaded that the person walking cept of a central processor and its limitations on the capacity the plank as it is laid on the floor can undertake additional of such a processor in modeling human performance in a number tasks, while the one hovering over the rink will resist under­ of domains. taking more. But what is it about walking a plank when it is It is useful to explicate the concept of a limited-capacity very high that is different from walking it when it is lying on processor at the beginning of this discussion since most of the the floor? Of course, in one case the consequences of failure are methods proposed, and employed, in the study of workload can more disastrous than in the other. The increased danger will be viewed as attempts to identify and quantify these limitations. "concentrate the mind," as Samuel Johnson said of the con­ Following the discussion of the limited-capacity processor we demned. We would not be surprised that when balancing above shall review the same classes of measures of workload discussed

'- WORKLOAD-AN EXAMINATION OF THE CONCEPT in detail by O'Donnell and Eggemeier (Chapter 42). We will chunks of information in working memory (Miller, 1956), Gestalt first consider the introspective, or "subjective," measures (Moray, and grouping principles in perception, integrality of dimenaiOill 1982), which assume that operators can perceive their own (Garner, 1974),andeffectofreorganizationandtraining(Neiuer, limitations and make truthful, or at least usable, reports. The 1976; Schneider & Shiffrin, 1977). Note that James and his second class of procedures assumes that the operators' intro­ contemporaries tended to identify attention with the contents spections need to be augmented by additional objective obser­ of consciousness. The limitations on the central processor are, vations. Secondary-task techniques (Ogden, Levine, & Eisner, in this framework, identical to the limitations on consciousness. 1979) assume that it is possible to assess the limitations on However, an exclusive identification of attention and of infor­ capacity by imposing yet another task on the subject. Failures mation processing with the operation of consciousness is not to perform this secondary task are taken as evidence that the tenable. It may do when one attempts, as James did, to develop ''primary" task is exceeding the limits on the system's capacity. a comprehensive catalog of mental activities. However, when Finally we review psychophysiological techniques that record the goal is a system to predict behavior, it cannot be ignored the activity of certain bodily systems and assume, as do the that the scope of information processing far exceeds the scope subjective measures, that an individual's body can reveal the of consciousness. Indeed, it is possible to assert that many of load on the system. They replace introspection by a reliance on the processes of key interest to the theorist ofhuman information the wisdom of the body (Cannon, 1932). processing are not, and cannot be, accessible to consciousness Each of these approaches is illustrated and the advantages (Broadbent, 1982; Kahneman & Treisman, 1983; Kaufman, 1979; and drawbacks within the framework of our concept of workload Posner & McCleod, 1983). are considered. The final section proposes an approach to the Therefore, the phenomena of attention and workload en­ analysis and measurement of workload that integrates these compass a broad spectrum of selection, transformation, and considerations. processing activities only part of which is accessible to con­ sciousness. Thus, with mounting evidence on the structure of the information processing system, and with increasingly de­ 2. THE LIMITATIONS OF THE CENTRAL tailed functional descriptions of the systems, it becomes evident PROCESSOR: HISTORICAL BACKGROUND that theories of attention and workload account for a vast array ofprocesses about which no subjective information is available. 2.1. Origins of the Notion of a Central Limited The limitations imposed on performance may derive from Processor mechanisms that are opaque to subjective assessment, such as The notion of a central limited processor can be traced to the memory search, the construction of grammatical sentences, and concept of attention discussed by such pioneers of scientific psy­ allocation of attention to one ofseveral channels. It is unfortunate chology as William James (1890) and Titchner (1908). Another and counterintuitive that consciousness reveals only a smat­ important influence has come from communication engineering tering of the workings of the system of interest. Yet, if we are in the years following World War II (Broadbent, 1958; Miller, to model and measure limits on the information processing 1956). William James and his contemporaries equated selective mechanisms, we must go beyond William James and quantify processing and attention with the mechanism that regulates limitations outside of consciousness. It will be easier to review consciousness. James's discussion of the limits of consciousness ' the nature of such limitations after we introduce the concept implied, in modern terms, that the mechanisms operate as if of information channel that has played an important role in they were a central limited processor. It is so structured that the analysis of attention and workload. it can attend to one single event at any one time. In James's 2.1.1. The Human as an Information Channel. The effects words: of communication engineering on psychological theory can be seen by examining the effect of information theory on experi­ Every one knows what attention is. It is the taking possession by the mind, in a clear and vivid form of one out of what seem mental psydtology during the 1950s and 1960s. The analysis several simultaneously possible objects or trains of thought. Fo­ is of particular interest here because the very notion ofa "limited­ calization, concentration, of consciousness are of its essence. It capacity" that is central to current discussions of workload de­ implies withdrawal from some things in order to deal effectively rives from c'ommunication engineering and in particular from with others, and is a condition which has a real opposite in the the theory of information developed by Shannon and Weaver confused, dazed, scatter-brained state which in French is called (1949). The majot contribution of information theory to psy­ DISTRACTION, and ZERSTREUTHEIT in German. (pp. 403- chology has been the acceptance of"information" as a commodity 404) that can be manipulated, transmitted, and transformed. More­ When making these claims, James was careful to specify his over, discussions of information could proceed without any ref­ notion of the possible nature of such "single" events that can erence to the physical implementations of the information pro­ capture the totality of consciousness at a given moment: cessing device. One can describe the properties of information The number of things we may attend to is altogether indefinite, flow, or of the transformation of information, by the same formal depending on the power of the individual intellect, on the form system whether the system is implemented by vacuum tubes, of the apprehension, and on what the things are. When appre­ transistor&, or, it is hoped, neurons. Thus psychologists attempted hended conceptually as a connected system, their number may to explain human information processing in terms of the flow be large. But however numerous the things, they can only be of information within the organism. The human processing known in a single pulse of consciousness for which they form system was likened to a communication channel that processes one complex "object," so that properly speaking there is before mej!Sages and transmits information from a stimulus set to a the mind at no time a plurality of IDEAS, properly so called. response set (Attneave, 1959). (p. 405) In ordinary usage, the term information refers to the prop­ This definition of a processing event anticipates by many erties of events, stimuli, or messages that reduce one's uncer­ years the contemporary distinction between isolated items and tainty about the true state of affairs. In formal information 41-6 HUMAN PERFORMANCE

theory, information is "transmitted" to the extent that the ap­ 3. Since the information measure is dimensionless, it seems pearance t;Jf a message reduces the prior probability of a response likely that information theory will provide a general metric in the response set. The greater the change between the prior by which the processing demands of various tasks can be and the posterior probability of the response, as a result of the compared regardless of differences in input modalities, re­ presentation of a message, the greater the amount of information sponse modes, and experimental conditions. transmitted by the message. 4lhis enterprise, if successful, would provide both a theoretical. A key concept of information theory is that of the com­ framework and a metric for the analysis of those task attributes munication channel, which exists between any two communi­ that lead to invocation of the workload concept. Limitations on cating points. It is defined by its capacity to transmit information the operators are inherent iil the framework and a metric derives between sender and receiver, and it is characterized by a number directly from the Shannon-Weaver (1949) definition ofchannel of quantifiable parameters, the most crucial of which, for our capacity. It is instructive, therefore, to examine the efforts made discussion, is that of channel "capacity." One of Shannon and to implement this program and the reasons it has failed to Weaver's (1949) principal insights was that channels can vary provide a comprehensive solution to the definition and mea­ in their capacities and that these differences can be quantified surement of the limitations on human operators. within the framework of information theory. A channel displays its full capacity if it imposes no reduction in the transmission 2.2. Channel Capacity as a Measure of Limits on of information between sender and receiver. Degradations of the Central Processor channel capacity are measured as decrements in information transmission. If the uncertainty of a receiver is reduced at a A variety of experimental work. designed to test th'e applicability lower rate than would result from the informatipn available of information theory in the analysis of human performance from the sender, channel capacity is assumed to have been was undertaken. The common purpose of these studies was to degraded. The communication engineer equipped with detailed model a variety of tasks in terms of their information trans­ specifications of the communication system can, indeed, measure mission characteristics. The studies were designed explicitly channel capacity and can design systems that optimize channel · to assess the capacity limitations of the human information capacity. processing system. In the terms of this discussion, task difficulty The transition to psychology is intuitive but difficult. There is equated in these models with the channel's communication were many attempts to borrow the· concept of channel capacity load. The larger the required capacity, relative to available in its most precise guise for modeling data on human perform­ capacity, the greater the communication load. If the human ance. The appeal of this model as an attempt to develop an information processing system can be adequately described in "objective" approach to the study ofbehavior appeared to promise terms of classical information theory, especially if channel and a viable substitute to stimulus-response models that preserved channel capacity can account for the limitations on performance, much of their spirit, as has been noted by G. A. Miller (1956) the need to invoke mental workload as a concept is obviated. in his classic paper ''The Magical Number Seven, Plus or Minus Performance decrements would then be explained in terms of Two." specifiable mechanisms in the language of information theory. The amount of information is exactly the same concept that we It is important to examine the degree to which information have talked about for years under the name of variance. The theory has provided satisfactory descriptors ofhuman perform­ equations are different, but if we hold tight to the idea that ance. It would be beyond the scope of this chapter to review in anything that increases the variance also increases the amount detail the vast literature describing information theory in psy­ of information we cannot go far astray. (pp. 87 -97) chology. The reader is referred to such sources as Coombs, Dawes, and Tversky (1970), Fitts and Posner (1967), Keele 2.1.2. The Putative Advantages of Information Theory. For (1973), and Sheridan and Ferrel (1974). Here we shall briefly the stimulus-response (S-R) psychology of the time, the mission discuss studies of the absolute judgments task, choice reaction was to develop laws to predict responses that organisms make time, immediate memory, and the control of movements to il­ to stimuli. In this context, information theory appeared to provide lustrate the achievements and disappointments resulting from a useful language. It is possible to conceptualize the laws of this effort. behavior as descriptors of a communication channel between 2.2.1. Absolute Judgment. A paradigm in which infor­ stimuli and responses. If behavior is organized, consistent, and mation theory appeared to be particularly useful is the absolute "lawful,'' there is mapping from the set of stimuli to the set of judgment task. The subject is presented, on each trial, with one responses. The presentation of a stimulus will yield a specific of a set of stimuli and is asked to identify that stimulus. Stimuli response. In information theory terms we say that information may vary along a single dimension (e.g., loudness), or along has been transmitted between the stimulus and the response. several dimensions (e.g., loudness and pitch). The investigator The adoption of information theory 1tnd the communication is generally interested in determining the number of different channel metaphor have thus offered to the emerging science of stimuli that the subject can identify. For example, Pollack (1952) "objective" experimental psychology three main advantages: asked listeners to identify tones by assigning a different number 1. Information theory provides a formal quantitative approach to each tone. He varied the frequency of tones and covered the to the modeling of behavior. range 100-3000 Hz in equal logarithmic steps. When a tone 2. Information theory appears to enable S-R-based modeling, was sounded, the listener had to respond with the tone's assigned with few general, nonspecific assumptions regarding the number. characteristics of the central processor. Assuming that the The results of this experiment are plotted in Figure 41.1. central processor operates like a limited communication When the set comprised no more than two or three stimuli, channel, one could consider workload entirely in terms of subjects identified all stimuli correctly. Confusions were rare the formal properties of stimulus and response. with four tones, but as the number of tones in the set exceeded WORKLOAD-AN EXAMINATION OF THE CONCEPT 41-7

5 ~ I PITCHES 2.2.2. The Span of Short-Term Memory. Another difficulty e 100 -8000Hz with viewing the human information processing system as a z classic communication channel arises in the attempt to gen­ 0 4 / / eralize the conclusions of one application to another. Consider, ~ / :::::E / for example, studies of the span of short-term memory. In these 0:: / 0 / tasks, subjects are presented with a list ofitems in rapid succes­ .... 3 / ~ / sion and are asked to recall as many items as they can from 0 w this list. For example, Hayes (discussed in Miller, 1956) read 1- 1- 2 aloud to subjects five lists, at a rate of one item per sec. Varied :E in complexity, each list contained either binary digits, decimal VIz

...... _ WORKLOAD-AN EXAMINATION OF THE CONCEPT 41-9 may recognize in these questions issues that accompany any ceasing mechanisms, one of which is more constrained than the discussion of mental workload. We will return later to a detailed others. The processing capability of this mechanism then sets discussion of this complex topic. Note, however, that within the the limits for the entire system in compiling task demands. fr,amework of information theory-based analysis these issues The first and most influential single-bottleneck model was pro­ played a central role and were associated with a formal technical posed by Donald Broadbent in 1958 as a summary statement meaning. If capacity is limited to about 2.5-3 bits of information to his book Perception and Communication. Broadbent proposed per second, and the demands imposed by a certain task are 2 the information flow model summarized in Figure 41.4. bits/sec, the processor can be said to have 1 bit of spare capacity. In many ways, this is still, quite explicitly, a communication Although the calculus was too simple as a model of reality, the channel model. The central construct was the limited-capacity underlying concept is remarkably viable. channel (the P system) preceded by a selective filter and a short­ Attempts to develop models of the limited processor in terms term store. This processor leads into a long-term store and a of strict information theory were abandoned by the late fifties mechanism that selects and controls the system's responses. in favor of models postulating internal mechanisms that operate The filter restricts the entry of information into the processor on information and determine channel capacity. While it took so that only relevant inputs are fully analyzed. It also affects as its starting point a metric of capacity, the new models were response selection and transfer of information into long-term based on one or another mechanism that was responsible for memory. the observed limitations in performance. Investigators in this The filter was assumed by Broadbent to operate in a manner phase were concerned not with precise metrics for expressing that can be described by classical information theory. However, the limits, but rather with demonstrating empirically that their as the communication channel was supposed to model only one structural model of the mind was valid. The emphasis was on component of the system, Broadbent's model was better able postulating and demonstrating bottlenecks in information flow. to cope with the complexities of the entire information processing The workload construct played a minimal role in these efforts system. By postulating that long-term storage and the generation as they were primarily viewed by the practitioners as studies of responses are independent, Broadbent's formulation allows of "attention." Yet, because this work provided a framework . the triggering of responses without intervention by the central within which the workload construct developed, we will review processor. Moreover, long-term memory can influence the flow briefly a couple of prominent ''bottleneck" models and trace of information through the system. Thus, the metrics provided their influence on the development of workload as a hypothetical by information theory could be used to describe the limitations construct. In reviewing these models, we will distinguish between on performance imposed by the filter. Other more cognitive single- and multiple-bottleneck models. The former assume that factors could explain the control of the filter. The operation of the limitation on performance can be localized in one universal the filter, in Broadbent's formulation, is influenced by properties mechanism; the latter take a pluralistic view of human limi­ of the incoming information as well as by information of long­ tations, admitting that we can be imperfect in many different term store. ways. Selection by the filter is based, according to Broadbent, on 2.4.1. Single-Bottleneck Models. Single-bottleneck views physical features of the input. These are monitored in parallel, promote an information flow model that comprises several pro- but only events that satisfy certain physical criteria are admitted

... :------.------,

Store of conditional probabilities of past events

Limited capacity channel (P system)

Figure 41.4. A diagram of the flow of information within the nervous system. Information received by the senses is transmitted in parallel to the short-term buffer and arrives at the filter. The filter is tuned to pass only messages having relevant physicai properties, one message at a time, to the central processor. The central processor precedes the long-term store and the response mechanisms. The filter protects the central processor from overload. Screening of information is done at an early stage, before semantic analysis (the costly operation) has been performed. (from D. E. Broadbent, Perception and communication. Copyright 1958 by Pergamon Press. Reprinted with permission.) ·~

41-10 HUMAN PERFORMANCE to further processing by the central processor. Broadbent's filter inherent part of the study of workload by suggesting how the model thus makes strong claims regarding information flow in efficiency of selection might affect the whole system's workload the information processing system. Moreover, information theory level. asserts a differential cost to different classes of operations em­ 2.4.2. Welford's Single-Channel Model. We examine an­ ployed in processing the input. The filter is designed to optimize other single-bottleneck model to illustrate a class of models the information flow given the system's tasks. that focused on the structure of the central processor as a source Broadbent's bottleneck model is clearly one of workload as of the limitations on the system. The difference between Broad­ much as one of attention. The experimental paradigms within bent's model and Welford's (1952, 1959, 1967) model is instructive which ·the model was spawned focused on failures in perform­ because both illustrate the dependence of theories on the ex­ ance. The filter is, in effect, the source of the subjects' inability perimental paradigms within which they develop. While ·~ i to process multiple inputs when the information load is excessive. Broadbent's work was designed primarily to account for such If workload is viewed as the difference between actual and ex­ qualitative differences in performance as failures in recall, pected performance given structural definition of a task, then Welford had been mostly concerned with temporal aspects of the workload imposed on a subject by one or many tasks will performance. The slowing of responses due to the interaction be determined by the degree to which the information load between tasks serves as the primary source of data for Welford's imposed by the task exceeds the capacity of the filter. (Even model. Thus, even though Welford, like Broadbent, proposed a though Broadbent:s filter is presumably intended to protect a single-channel model that was influenced by classic information limited-capacity processor from excessive information load, the theory concepts, the two models are quite different. construct regarding which the theory speaks is actually the The key observation that Welford's theory is designed to filter itself rather than the limited-capacity processor. In the explain is that the response to a second stimulus is delayed if present context, the filter carries the theoretical load.)· the subject has not yet responded to a stimulus just presented. The filter can be thought of as an active, early, workload­ The shorter the interval between the first and second stimulus, establishing gatekeeper that limits the information load on the the longer the response to the second stimulus is delayed. Wel­ system. The reader will note that there are, in this view, im­ ford, like Broadbent, postulated a three-stage model of the in­ j plications for system design. If workload is established by the formation flow within the organism and located the bottleneck filter, tasks need to be designed so that they match the properties of processing in the limited capacity of the central processor. of the filter. For example, the nature of the displays associated Welford attributed the delay to the limitations on the operation with a task need to be considered so that requirements of the of a central decision mechanism that can process only one task task are minimal in terms of processing load. at a time. The time required by the mechanism to process one The filter model suggests that additional costs may be in­ task was labeled the psychological refractory period (PRP), as l curred when tasks require integration of separate physical at­ a countermatch to the term refractory phase used by Teleford tributes in a stimulus array. Consider, for example, the task (1931) to describe delays in response of a synapse or a nerve to described by Treisman and Gelade (1980) in which subjects successive stimulation. found it relatively difficult to count the number of blue triangles Welford's main thesis was that performance is limited by in an array of blue and white squares and triangles but relatively the operation of a single-channel decision mechanism. This easy to count all blue objects or all triangles in the same display. mechanism can deal with data of only one signal, or a group of That is, the identification of a conjunction of attributes, requiring signals, at a time, so that data from a signal arriving during parallel processing of the stimuli along several dimensions, the reaction time to a ptevious signal have to wait until the imposed a heavier workload on the system. Issues related to decision mechanism becomes free. The decision mechanism is this interesting finding are discussed at a more formal level in frequently occupied by feedback from execution of the move­ Sperling and Dosher, Chapter 2. ments or termination of the response; therefore additional delays Another example of a Broadbentian implication to task may occur even when a signal arrives shortly after the response design derived from the suggestion that tasks requiring semantic to a previous signal. analysis, such as word identification, categorization, or decision Welford (1967) reviewed-data from his work and the work making, cannot rely on the peripheral filter. The central pro­ of others to support his arguments and validate the general cessing system must be able to deal with these tasks without form of the relationship depicted in Figure 41.5. the presumed protection the central system receives, according These experiments also showed that the delays were not to Broadbent, from the filter. Such tasks would benefit from eliminated with training, and that they were localized at the ''tagging" all relevant input units with a common distinguishable central processor rather than at a peripheral sensory or motor physical characteristic since irrelevant units will be rejected site. This latter claim is based on the appearance of response from analysis at an early stage. delays in the case in which one signal was auditory and the In summary, Broadbent's model proposed several general other visual, even when the two responses were made by different principles that made it possible to describe the information flow hands, or when no overt response had to be made to the first in the human information processing system and that serve as stimulus (Davis, 1956, 1957, 1959; Fraise, 1957; Slater-Hammel, anchors for much of the current research in the field of workload 1958; Welford, 1952, 1959). assessment. He suggested that the path between stimulus and The model also recognized the possibility that a grouping response might be viewed as three successive stages and de­ of stimuli may occur if the interval between the stimuli is short. scribed the functional properties of each stage in terms that Welford estimated the time required for closure of the gate to could be applied to the analysis of the task components. He be about' 30 msec, during which information may enter and assigned a cost function to the performance of mental operations, grouping can occur. Ifthe rate of an arriving sequence of events identifying operations that are more or less costly to the system. is faster than the processing speed of the central mechanism, Finally, he emphasized the study of selective attention as an the response to each event will be proportionately lengthened,

""-- WORKLOAD-AN EXAMINATION OF THE CONCEPT 41-11 joint demands of tasks do not exceed the limit of the central processor. Additional tasks can enter the central processor at any time if the currently processed task does not exhaust the

RT 2 capacity of the processor. The limits discussed by Welford are more structural, in the sense that the mechanism is completely inaccessible to serve in the performance of other tasks when it is occupied by one task. Parallel processing depends on the time of arrival of tasks and the ability to group them. When grouping is possible, the tasks, in effect, become a single task that is likely to occupy the central processor for a longer duration. The implication of this type of limitation is that once a task or a task element enters the decision mechanism its pro­ cessing proceeds uninterrupted. Partial processing or errors RT2 by itself due to interactive influence of elements are assumed to be rare. The decrements in performance that are manifestations of ~r workload appear as prolonged response times or as omissions

RT 1 Feedback RT 1 of the response. In contrast, partial processing and interaction -Time errors are consistent with the model of performance limitations t------lnterstirrulus interval oso-----1 proposed by Broadbent. This difference between the models may Figure 41.5. Ideal plots of the latency of the second response (RT2l as a result from different experimental paradigms and the principal function of the interval between the first and the second stimuli (lSI), when dependent measures from which each of these models was de­ 52 comes during the processing and response of the first stimulus (RT1).line rived. (a) shows the results expected when RT 1 and RT 2 are exactly the same in all Even at this early stage of the analysis of limitations on trials. line (b) shows the expected smoothing due to intertrial variability of system performance, we encounter an interaction between the RTs (after Welford, 1967). It can be seen that the predicted delay in response measures of performance and the model. Reductions in the caused by the psychological refractory period is maximal when two stimuli "quality of performance" can be assessed in a variety of ways; are presented simultaneously but are not grouped. It diminishes in a 45° therefore, an operator performing a task may appear to be either slope as the overlap between the two stimuli decreases. line (c) depicts a fulfilling or failing to fulfill performance expectations depending possibility of an additional delay due to feedback processes from completion on the measure of performance selected by the observer. Much of the first response. of the controversy in this field may be traced to such different perspectives on performance. A celebrated controversy that confronted "early" selection with "late" selection in attention, perhaps enough that some stimuli will not be processed. A tem­ with Broadbent favoring early selection and Welford's model porary buffer was assumed to store at most two events, thereby lengthening the period of graceful degradation. Figure 41.6 illustrates the manner in which Welford's model was used to interpret the results of an experiment by Conrad 120 (1954) requiring subjects to respond by pressing a key or turning a knob each time one of a number of rotating pointers coincided with one of several irregularly spaced marks on the edge of 'f'100 dials. The number of dials (two, three, or four) and the speed ·.!! ~ of rotation of the pointers were used to vary the information .E on the display. When the level on each of these variables was ..§, 80 Vl increased the portion of events omitted was increased and re­ t..J sponses were delayed. Vl ~ 60 Q. The reaction time data were fitted with a model that as­ Vl t..J sumed a constant reaction time for a given number of dials. 11:: This number was converted to bits of information and related 40 to the Hick-Hyman formula for the determination of an appro­ priate reaction time' value. (See Keele, Chapter 30.) As can be seen, the fit of the equation to the actual data is quite convincing. o Lt--r------.--...--~~0 40 60 80 100 120 It is instructive to compare the notions of central limitation and workload that emerge from Welford's single-channel op­ SIGNALS (minuteJ eration with those of Broadbent discussed earlier. Both models rely heavily on the information theory paradigm, both propose Figure 41.6. Conrad's (1954) data fitted assuming that signals arrive at three main stages in the flow of information, and both locate random intervals and that dealing with each takes an equal timet. It is also assumed that, if the subject cannot deal with the signal at the instant it the bottleneck in the operation of the central mechanism. How­ arrives, it can wait, but that data from not more than two signals at a time ever, they differ substantially in their assumptions regarding can wait in this way. In other words, the latitude in responding is two signals. the operational rules governing this mechanism. Broadbent 1: two dials, t = 0.37 sec; II: three dials, t = 0.59 sec; Ill: four dials, t = emphasized a limit in terms of the total amount of work per 0.74 sec. (From A. T. Welford, A single channel operation in the brain, Acta unit time. His model allows parallel processing as long as the Psychologica, 1967,27. Reprinted with permission.) ·~

41-12 HUMAN PERFORMANCE implying late selection, may be due in large part to different mation (e.g., Deutsch & Deutsch, 1963; Keele, 1973; Norman, perspectives on performance from the different vantage points 1968). The main contention of this alternative was that encoding of dichotic listening and reaction time studies. We will carry and intake of information are not as demanding and can be this cautionary note into further discussions of workload in the handled in parallel. The system becomes single channeled and remainder of this chapter. Performance is in the eye of the limited once the process of evaluating information and selecting beholder, and workload measures are limited by the perspective an appropriate response has begun. Data in support of this of th.eir design. Performance quality varies with the type of argument were obtained in experiments in which probes are performance being studied, and there is a corresponding lim­ inserted during an interval (e.g., Posner & Boies, 1971). The itation to performance-related measures of workload. response time to a probe stimulus presented during the pro­ 2.4.3. The Breakdown of the Single Channel. ·Both single­ cessing of a primary stimulus turns out to be unaffected during channel models were provocative enough to stimulate extensive the first 300 msec after the presentation of the primary task, experimental work, but both lacked the power to account for but the response is delayed considerably if the probe is presented the growing body of data. The ultimate test for any model of later in the interval. It is as if the interference is most powerful information flow within the organism is its ability to relate the beyond the point at which the main processing has been com­ functional properties of the model to the characteristics ofhuman pleted, and the subject is involved in "thinking" or deciding tasks, thereby improving the prediction of behavior. about the proper response. Welford's single-channel operation could not explain, for The inadequacies of the early~selection models were also example, the finding that varying the difficulty associated with implied by various demonstrations that the recall of information the response to the second stimulus affects the pattern of the (which implies that the information had been processed and responses to both stimuli (e.g., Karlin & Kastenbaum, 1968; represented in memory) is more demanding than the intake of Keele, 1973). lithe second stimulus is not examined before the multiple inputs (e.g., Martin, 1970; Trumbo & Milone, 1971). analysis of the first stimulus is completed, how can its difficulty Thus, the notion of a bottleneck early in the processing stream modify the interval between the first and the second response? gave way to models of processing that assumed a more central ' Another criticism of Welford's model was suggested by ("late" in the jargon ofthe times) locus for the limitations. We "~· see here the same process that affected the utility of information Kahneman (1973). He argued that, instead of examining the t• change in the speed with which subjects respond to the second theory models. The information processing system, upon close stimulus as a function of the interval between the first and the examination, is active and dynamic, and any attempt to model i it with a peripheral, data-driven model is unlikely to capture sec.ond stimuli (lSI), one should examine the interval between i.l the first and the second responses (IRI). When this measure is its full capabilities, even if-one allows the peripheral filter to be controlled by central factors. Adequate models of attention, employed, clear evidence emerges for processing in parallel of j the two stimuli. That is, the second stimulus does not wait Until and by implications of workload, require consideration of the analysis of and response to the first one are completed. internal structure and the operating modes of the information Similarly, experimental tests of predictions derived from processing system. Broadbent's model yielded evidence that is inconsistent with Even though an early-bottleneck model fails to provide a the early selection hypothesis and with the idea that the filter full account ofthe operation ofthe information processing system, serves to protect the central mechanism from possible "overload" its rejection as a general model should not imply a rejection of of high-level semantic processing demands. These assumptions the possibility that there are such early-selection processes. were challenged by two main groups of findings, the first of The observations on which these models were based cannot be which addressed the assumed costs of semantic processes. It ignored. Thus, for example, the dramatic loss of information showed that subjects were able to follow, with little impairment arriving in the unattended ear in shadowing tasks is real enough, in intelligibility, verbal messages presented at twice the normal as are the difficulties encountered when attending to and in­ rate (Fairbank, Guttman, & Miron, 1957). Also, if the infor­ tegrating simultaneous sources of information. Furthermore, mation content of a passage was doubled by using low-level rather dramatic direct evidence for the operation of an early ' approximations to English, shadowing performance did not de­ selection has been provided by Hillyard and his colleagues crease to 50% (Treisman, 1965). These findings conflict with (Hillyard, 1984). These investigators recorded event-related i Broadbent's interpretation of the inability to follow the verbal brain potentials (ERPs) from human subjects while these subjects content ofthe message on the irrelevant ear in a dichotic listening were engaged in Broadbentian multiple-channel monitoring. task as indicative of the high demands imposed by verbal pro­ (For a definition ofERPs and a review ofthe manner in which cessing on the central processor, precluding simultaneous lis­ they are recorded and interpreted, see Section 3.9.) The ERPs ., tening and processing of material from the two ears. recorded from the attended channel were distinctly different i Another series of studies examined the claim that filtering from those recorded from the unattended channel. Moreover, ~ by physical cues discards the irrelevant information from further these differences appeared as early as 100 msec after the eliciting analysis. It was found in "shadowing" experiments that, if the stimulus. Naatanen (1983) presented a theory of selective at­ text presented on the relevant ear was switched to the irrelevant tention that integrates these ERP data with the literature re­ ear, subjects followed the text and abandoned the information viewed in this section. Other attempts to defend the validity of presented on the relevant ear (Treisman, 1960). Other exper­ an early-selection process can be found in Broadbent (1982) and iments have shown that subjects could detect significant words in Kahneman and Treisman (1983). on the irrelevant ear (Moray, 1959, 1967) or monitor for target The current versions of this approach tend to resemble the words on the two ears. All the above could not have happened version of the filter model proposed by Treisman (1964, 1969). if Broadbent were correct in assuming that information on the Treisman's model differed from Broadbent's in two major aspects: irrelevant ear does not reach the level of semantic analysis. (1) The filter was assumed to attenuate rather than completely These findings led several researchers to propose a late­ block the information arriving in the unattended channel; and rather than an early-selection bottleneck in the flow of infor- (2) the filter was given a strategic flexibility, in the sense that

"'-- WORKLOAD: AN EXAMINATION OF THE CONCEPT 41-13 it could be deployed at various locations along the information it became necessary to develop a methodology that could be processing path, to increase the cost-effectiveness of the selective applied to the analysis of the system. A natural extension of process. In Treisman's words: the models reviewed is the identification of multiple components Four types of attention strategy are distinguished: the first re­ in the system, each of which is describable in simple terms and stricts the number of inputs analyzed; the second restricts the with simple rules for governing the interactions between them. dimensions analyzed; the third the items (defined by sets of critical The components tended to be viewed as functional entities, features) for which subject looks or listens; and the fourth selects each playing an identifiable role in the informational trans­ which results of perceptual analysis will control behavior and actions of the system. A simple rule for the interactions between be stored in memory. (Treisman, 1969, p. 282) components is that they are strung in sequence, each feeding Note that for Treisman the filter is an "attention strategy" its successors with information. Such a view of the system leads rather than a fixed structure, implying that filters can be ac­ naturally to the concept of a "stage,'' with each component serving tivated at different locations depending on the conditions at as a stage in a sequence of processing. According to this model, which a task is employed and the nature of the competing en­ the operation of a successive stage does not begin before preceding vironment. This later argument has testable implications for stages have completed operations (Sanders, 1980; Sternberg, the processing operations carried out between stimulus and 1969). Other approaches are possible. Thus McClelland (1979) response. Tests of relevance could now be performed at the proposed a cascade model and Eriksen and Schultz (1979) a input, the dimensions (analyzers), the memory and response, continuous flow model according to which all stages of processing or the item levels. The model represents a further step in the operate continuously, passing information from one stage to decomposition and elaboration of the original global notion of the next as the information becomes available. information. An implication of this approach to the study of The term processing stage thus refers to an aggregate of workload is that the assessment of the workload imposed by processing structures or computational processes that represent tasks must now be detailed further in terms of the components a common mental operation. Thus the preprocessing of a stim­ ofprocessing involved. This approach is particularly significant ulus, feature extraction, response choice, response programming, as it is related to the degree to which -parallel rather than and motor adjustment are all examples of stages (Sanders, 1980, sequential processing is the dominant mode. In a system com­ 1983). The number of stages is not fixed and depends on the prising many mechanisms and processes, elements of a task requirements of the specific task. For example, assume that the are likely to occupy different processing mechanisms simulta­ subject is required to determine whether two letters in a pair neously and to be at various processing states. In such a system, are "the same." If the decision is based on the character and on the original notion of a single-flow, sequential streaming of the font in which the pair is presented (so that Aa and AB may information loses much clarity. and power. This view of the be considered different while AA and aa call for the response limitations of the information processing system leads naturally "same"), then there is a need for a stage in which the physical to a view of workload as a multifaceted concept as there are features are compared and one in which the letters are identified. various reasons for a decrement in performance. In contrast, if the font alone is the criterion for responding (for It seems evident that even though Treisman's variant of example, AB and ab are the "same" and Aa and Bb are different) the filter model is a "single-channel" model it actually permits then only physical features need to be identified (Posner, 1978). a rather elaborate definition of the channel concept. In fact, The main experimental tool in the analysis of the string the very notion of a channel loses much of its value when it of stages for a given task is the prolongation of response time refers to an ensemble of processing entities that communicate as a result of manipulating task variables. In the majority of with each other under complex control schemes. Thus, the con­ studies, such an analysis is based upon the additive factors cept of what constitutes a channel has been considerably paradigm proposed by Sternberg (1969). Donders suggested in changed. 1868 (see reprint in Donders, 1969) that it is possible to ma­ Regardless of the remaining viability of the channel concept, nipulate task variables and then measure the consequent the decades following the predominance of models derived from changes in reaction time in order to establish the internal stages, communication theory were characterized by a shift to an ex­ unobservable by themselves. It was Sternberg (1969), however, amination of the internal structure of the processor with a specific who incorporated this logic in the general framework of infor­ interest in its architecture. We proceed to review this phase. It mation processing and developed a formal approach to the ex­ will be seen that a dominant concept in this paradigm was traction of stages. Sternberg reasoned that variables having "stages of processing." The relevance of this literature to the an additive effect on response time must operate at different analysis of workload may not be obvious. The concern with stages, while variables that interact influence at least one com­ architecture reduced the prominence of the processes that ac­ mon stage. For example, signal quality and stimulus-response count for limitations of processing. However, as it became clearer compatibility were shown to have additive effects on reaction that the architecture of the system is complex, it became evident time in a variety of situations (e.g., Sanders, 1979; Shwartz, that limitations in processing can appear in a variety of nodes Pomerantz, & Egeth, 1977; Sternberg, 1969). In contrast, stim­ in the system, and the cost of different mental operations emerged ulus-response compatibility was found to interact with time as a central issue in the attempt to link the limitations of the uncertainty (Broadbent & Gregory, 1965; Sternberg, 1965). central processor to the components of task demands. The fol­ Signal quality and compatibility were interpreted to affect the lowing section reviews studies of the architecture of the pro­ stages of encoding and response choice, respectively, while cessing system (e.g., Sanders, 1980; Sternberg, 1969). compatibility and uncertainty were both associated with the response selection stage. Table 41.1 summarizes variables found 2.5. The Architecture of the Central Processor to have additive or interactive effects on response time. The general structure of information flow and the logic As it became evident that the limitations on performance require underlying the additive factors methodology were generalized an examination of the structure of the central processing system, to other tasks and to different problem areas such as workload 41-14 HUMAN PERFORMANCE

Table 41.1. Summary of Experimental Variables That Were Found to The apparently complex architecture revealed by contem­ Have Additive and Interactive Effects on Choice Reaction Time porary analyses of the information processing system is costly. Additive effects: The complexity of the structure, multiplied by allowing con­ Signal quality x S-R compatibility: Sternberg (1969), Shwartz et al. currency and parallelism into the system, has made it difficult (1977), Frowein and Sanders (1978), Sanders (1979). to provide simple tools for workload measurement. It is possible, Signal contrast x S-R compatibility: Shwartz et al. (1977), Sanders however, to ignore the structural complexity of a system when (1977). trying to assess its limitations if the critical limitation can be Signal contrast x signal discriminability: Pachella and Fisher ascribed to a uniform source viewed independently of the system I (1969). structure. Such a solution is provided by adopting yet another j Shwartz et al. (1977). metaphor, the energy metaphor. In analyses of workload, the 1 Signal contrast x signal quality: Frowein (Note 2), Sanders and energy metaphor has played a dominant role, and it is reviewed Akerboom (Table 3). ~ in the following section. The energy metaphor, as well as related I Signal contrast x word frequency: Becker and Killion (1977). ~ Signal quality x word frequency: Stanners et al. (1975). metaphors such as resources, is consistent with the view of Signal discriminability x S-R compatibility: Fisher and Pachella workload presented in the introduction to this chapter. (1969), Shwartz et al. (1977) Ifworkload is a hypothetical construct designed to account S-R compatibility x instructed muscle tension: Sanders (1979). for the variations in the function of the operator-task loop, Signal quality x instructed muscle tension: Sanders (1979). then the test of the utility of a construct is determined by the S-R compatibility x response specificity: Sanders (1970). descriptive economy it allows rather than by its fidelity as a Interactive effect: structural model of the system under consideration. As the next Signal quality x movement frequency x movement predictallility: section will show, energy andresource models do provide rather Wertheim (1979). attractive descriptors of the system's functions, .. though here l Stimulus contrast x S-R compatibility: Stanovich ancl Pachella again the system will require substantial elaboration of these j (1977). models. Stimulus contrast x meaningfulness: Miller and Pachella (1976). I Priming x word frequency: Becker and Killion (1977). 2.6. Energy Constraints on Processing Capabilities Priming x signal quality: Meyer et al. (1975). Priming x signal contrast: Becker and Killion (1977). 2.6.1. The Single-Resource Model. The most comprehen­ sive attempt to employ the energy metaphor in the analysis of Based upon the rationale of the additive factor methodology, those variables workload, and of attention, can be found in Kahneman's book that interact influence to the same stage, while those that are additive affect I separate stages. (Adapted from Sanders, 1980.) Attention and Effort (1973), which attributes a system's failure to perform to a shortage in the supply of what he calls processing I resources. Resources is a label applied to a single undifferentiated j assessment and dual-task performance (e.g., Logan, 1978, 1979; pool of energizing forces necessary for task performance. In its Whitaker, 1979; Wickens, 1980) as well as task analysis (Mane, derivation, the concept of resources is closely related to the i Coles, Wickens, & Donchin, 1983). The paradigm and its as­ concept of arousal, which has an important role in some theories sumptions and findings played an important role in the for­ of attention. Consider for example Lindsley's activation theory i mulation of an elaborate model of the central processor whose (Lindsley, 1951.) This psychophysiological model assumes a t limitations are, or whose principal interest is in, the analysis structure of analyzers and processors that is fixed in its capacity. ~ ~ of workload. Processing is possible to the degree that any given processor is As Treisman's elaborated filter demands a multifaceted "activated." The activation, in Lindsley's model, depends largely view of workload, so does a view of the system that locates the on the influence of the reticular activating system that emerged effects of different task attributes in different processing stages. from the studies of Magoun (1949) and his collaborators im­ Neither of these models is consistent with a notion that workload mediately following World War II. There is a fairly direct route is due to a specific and unique deficit in the system. A task may from the ''physiological" activation that underlies the ability vary in ways that limit performance because one or several of the system to process information to Kahneman's resources. stages of processing are affected. The effect of variations on the It is critical to understand that neither activation nor re­ structure of a task and on the demands it imposes may be additive sources are directly observable entities. In both cases, these for some variables, interactive for other variables. It may also are hypothetical constructs introduced to organize observations happen that task demands will not be affected by a manipulation on performance. This point is sometimes neglected in discussions of task structure. Some variations in task demands may be of resource models. The literature tends to slide into a reification totally irrelevant to performance of one task yet be crucial to of resources in a manner that suggests that the concept refers t performance of another. One of the major advantages of the to actual observable entities; it is better to admit that the resource additive factors technique is that it provides a methodology for is a concept that does not relate to a specific structural model ~ operationalizing an analysis of a task's structure in terms of of the system:. In adopting this concept one asserts that organisms task variables. This view of the system is important in some of behave as if they are resource-dependent systems in the sense the common approaches to the measurement of workload. The that there are clear limits on their capacity to perform. As it notion that one must consider the system as having a structure is rather difficult to pin the limitations on any specific mech­ underlies the approach to workload that requires a formal task anism, it may be useful to pool the limitations and consider analysis in terms of its specific "structural" demands as a pre­ their appearance to reflect a decline in the level of a hypothetical condition to the analysis of task workload. This contribution is entity, the "resource." This usage of the resource concept is not not vitiated by the fact that it is obvious that the human in­ much different from the use one makes of arousal, which, after formation processing system shows extensive concurrency and all, is a hypothetical construct drafted to account for the evident interaction among its elements. differences in state that we display as we move on a continuum WORKLOAD-AN EXAMINATION OF THE CONCEPT 41-15

from slumber to activity. Something clearly changes when one a rise in these activities causes an increase in the level of arousal, wakens from sleep. The change is most properly described by effort, and attention. The general structure of the model is de­ reference to a host of specific changes in neural circuitry and picted in Figure 41.7. in the levels of a variety of neurochemical substances. Yet, for An important construct in Kahneman's model is the mech­ a variety of purposes, one may ignore this complexity and de­ anism responsible for the allocation policy. This mechanism scribe the organism as moving along a dimension of arousal. directs and supervises the allocation of resources and is influ­ This is a useful theoretical device as long as one does not infer enced by enduring dispositions, momentary intentions, and the from its success that inside the body is a specific system that feedback from o~going activities. embodies, or secretes, an entity corresponding to arousal. In a structural model, failures to perform occur when a A discussion of performance limitations organized in terms mechanism is required to carry out incompatible operations. of resources is not a structural model of the information pro­ For example, one's eyes cannot monitor simultaneously two cessing system. Rather, it is an assertion that it is possible to screens placed on opposite sides of a room. This limitation on treat the system, for the purpose of describing its limitations, performance is attributed to the specific structure ofthe human as if it depends on the availability of some hypothetical resource. visual system. In an energy-oriented model, such as Kahneman's In a sense, we are implying by use of this concept that it is not capacity model, decrements in performance are due to demands quite necessary to know why an operator fails to perform. As of two concurrent activities exceeding the available capacity. long as we can quantify the deviation between expectation and A structural model therefore implies that the interference be­ actual performance, we can treat all failures alike. This caveat tween two taSks is specific. In a capacity approach it is nonspecific is in place whether one adopts the uniform-resource model pre­ and depends only on the total demands of the two tasks. sented by Kahneman (1970) or some version of the multiple­ A concept of a central, limited source of processing energy resource models we will review in Section 2.6.2. provides a convenient solution to the conflicting data on the Kahneman viewed the amount of resources available at location of the bottleneck in the information flow within the any time as limited, but the limit varied with the level of arousal, processing system. Moreover, the concept is consistent with the according to the classical inverted-U function relating effec­ conclusion that there exists a fairly general source oflimitations tiveness of performance to arousal. Changes in the level of on the central processor. Such a general source of limitations arousal and consequent changes in capacity are assumed to be appears necessary since one often observes interference between controlled by feedback from the execution of ongoing activities; tasks that have very little in common. Why, for example, should

MISCELLANEOUS SOURCES OF AROUSAL:

anxiety, fear, anger, sexual excitement muscular strain, effects of drugs, intense stimulation, etc.

MISCELLANEOUS MANIFESTATIONS OF AROUSAL: pupillary dilation, I• increased skin conductance, ' fast pulse, etc. ~ ' Available capacity Available capacity and arousal increase to meet demands for processing capacity

~--'~" ' rrrr-~~ ___ ~~-----~o Possible activities

Responses t ~· Figure 41.7. Energetical model of capacity. Performance is constrained by the availability of mental energy from a single undifferentiated pool. Note the close linkage between capacity and the physiological determinants of arousal. Note also the introduction of an allocation policy mechanism and the closure of the loop from the present activity to the capacity pool, which suggests the idea of elasticity in the total amount of available resources. (From D. Kahneman, Attention and effort. Copyright 1973 by Prentice­ Hall, Inc. Reprinted with permission.)

I' Is 41-16 HUMAN PERFORMANCE

there be any interference between the ability to recall and the the shared task (e.g., Gopher & Navon, 1982; Gopher, Brickner, difficulty of a simultaneous manual tracking task (Johnson, & Navon, 1982; Wickens & Kessel, 1981). These results are Schulman, Greenberg, & Martin, 1970)? Or, why should the inconsistent with the notion of a single undifferentiated pool ability to attend to auditory messages be affected by the need ofprocessing energy. Rather, they suggest the existence ofseveral to monitor a visual display for a critical letter (Kahneman, more specific sources ofinterference and competition. Note that Beatty, & Pollack, 1967)? This interference may be explained we are again driven by the data to the conclusion we drew from by invoking the concept of a general energy source of a fixed the analysis of single-bottleneck models. Again, the concept of capacity made available to one or another task but not to both. a single general cause of performance limitations attributed to However, this source of energy is not observable. No direct workload must be rejected. observations on the level of available resources have been pro- 2.6.2. Multiple-Resource Models. The energy, or the re­ . posed either by Kahneman or by others. sources, metaphor is an attractive approach to the analysis of This aspect of the resource concept tended to lead its pro­ workload. It has an affinity to the origins of the concept in the ponents to rely on observations of physiological systems in an ergonomists' physical workload, and it is an intuitively appealing attempt to monitor the level of available resources, or the level way to describe what is lacking when performance falters for of demand on the resources. Such attempts were made by several "no apparent reason." The evident weakness of the single-re­ researchers prior to Kahneman (e.g., Berlyne, 1951, 1960, 1970; source model leads, therefore, to the development of multiple­ Easterbrook, 1959; Wachtel, 1967). However, none offered as resource models, according to which the human system is best detailed a model. Kahneman's measure of effort derived mostly modeled as possessing a number of processing mechanisms, from his own data on the effects of mental effort on pupil dilation. each requiring its own supply of "resources." The capacity of Several experiments conducted by Kahneman and his colleagues each of the structures, that depended on the level of arousal (Kahneman et al., 1967, 1968, 1969, 1971) showed changes in and its own specific dependence on this level, can be deployed pupillary response that paralleled variations in task demands. at any moment among a number of tasks. Thus, there is con­ For example, when subjects had to memorize and transform a tinuing competition for resources between tasks that overlap list of digits, the pupil was largest when the demands on memory in resource needs (Gopher & Sanders, 1984; Norman & Bobrow, were highest. Pupil dilation was greater when the transfor­ 1975; Navon & Gopher, 1979; Sanders, 1983; Wickens, 1980, mation the subjects were to perform on the digits was more 1983). difficult (Kahneman, Peavler, & Onuska, 1968). Within this framework the key questions are: What is the These data demonstrate the sensitivity of the pupillary nature of resources? How are they related to one another? How response to processing effort and serve to demonstrate the com­ can the demand composition of tasks be expressed in terms of petition for "energy" between tasks with little structural sim­ the underlying resources? Norman and Bobrow (1975), the first ilarity. Additional support for the relationship between the to employ the term resources, were not very specific. They used pupillary response and processing effort has come from recent a general analogy to a computer system, and resources were studies by Beatty and his associates, summarized in Beatty defined in general terms with reference to all processing facilities. (1982), who continued and elaborated Kahneman's research. Specific examples were ". . . such things as processing effort, With a notion of flexible limits on processing capacity that the various forms of memory capacity, and communication change with feedback from behavior (which in turn influences channels ... "(p. 45). Note that in Norman and Bobrow's writing, the level of arousal), an independent phy!'Jiological measure of the term resource lacks the energylike flavor of Kahneman's processing demands has a crucial role. Performance measures description of the system. Rather, the system appears to be by themselves are poor indicators ofresource limitations, because described by Norman and Bobrow on a structural level. Yet, performance is both the result of the limitation and a trigger even this seemingly structural approach is nonspecific in mod­ of a change in the limit via recruitment of additional resources. eling the nature of the deficits. Once performance limitations Only an independent measure can break this vicious circle. The are not attributable to any deficiencies in the supply of data, main test of the model is the validity of the claims that all they are attributed to an inadequacy of resources. tasks, regardless of structure, compete with each other, and Navon and Gopher (1979, 1980), who also elaborated the that an increase in the difficulty (demands) of one task will be multiple-resource framework, were no more specific in their reflected in the ability to perform another one simultaneously. initial theoretical analysis. They employed an economic met­ l By and large, it seems fair to say that the single-resource model aphor and drew an analogy between the problems facing a person did not fare well under experimental analysis. performing one or two tasks and a manufacturer of one or more Data from a variety of experimental conditions have shown products who has to optimize the use of resources such as labor, I that performance on some tasks interferes with one type of equipment, raw materials, and so on. In subsequent experimental task, but not with another, while the third group is equally work (Gopher & Brickner, 1980; Gopher et al., 1982), they con­ affected when paired with members of the first two groups. For clude that it is reasonable to assume the existence of at least example, mental arithmetic is little impaired when jointly per­ two relatively independent types of resources; one is related to formed with a pursuit-tracking task but severely degraded if perceptual and computational processes, and the other is linked paired with a choice reaction task to visually presented digits. with selection and generation of motor activity. In contrast, the choice reaction task is equally degraded when The nature of resources is a central issue of concern in the paired with mental arithmetic or pursuit tracking (see Navon work of Wickens (1980, 1981, 1983) and his colleagues (Wickens & Gopher, 1979; Ogden, Levine, & Eisner, 1979; Wickens, 1980, et al., 1981, 1982, 1983). Wickens (1980) proposes three plausible 1983, for this and additional examples). Other studies have candidates for the structural composition of resource reservoirs: shown that some manipulations of task variables within the stages of processing, cerebral hemispheres, and modalities of same pair of concurrently performed tasks affect both tasks, processing (both encoding and response). Classification by pro­ while others degrade performance on one task only and cannot cessing stages follows the architecture of the processing system be compensated by shifting resources from the performance of as it emerges from experiments based upon the additive factors

~- WORKLOAD-AN EXAMINATION OF THE CONCEPT 41-17 methodology (Sanders, 1979, 1980; Sternberg 1969). Hemi­ showed that the effects on response time of stressors such as spheres of processing are suggested from theoretical analysis sleep loss, time on task, or psychoactive drugs are selective. and experimental works that view the cerebral hemispheres as The stressors affect specific mechanisms but have no general acting partially as separate resource reservoirs by virtue of effect on performance. For example, amphetamines were found their functional and spatial separation (e.g., Friedman & Polson, to interact only with those variables associated with motor ac­ 1981; Friedman, Polson, Defoe, & Goskill, 1982; Kinsbourne, tivity, while barbiturates interacted only with variables that 1975; Kinsbourne & Hicks, 1978). The direct relevance of this influence processing activity as related to feature extraction. literature to task analysis bears more careful examination These findings were interpreted within the framework of (Donchin, McCarthy, & Kutas, 1977). the neurophysiological model of attention control proposed by Justification of modalities of processing and response as Pribram and McGuinness (1975), who identify three main en­ classification criteria is provided by those studies that compared ergetical generators of processing activity: arousal, activation, auditory and visual modes of presentation and verbal versus and effort. Integrated within the proposed structure of processing manual modes of response in dual-task paradigms (e.g., Gopher, stages, arousal is linked to input encoding activity. Activation Brickner, & Navon, 1982; McLeod, 1972, 1978; Wickens & Kes­ is argued to energize output processes, and effort is given the sel, 1981; Wickens & Sandry, 1982). Wickens (1981, 1983) pro­ role of activating the central decision-making and choice mech­ poses a three-dimensional descriptive framework that can serve anisms. In addition, effort is assumed to secure an optimal level to organize these categories (Figure 41.8) and also to include a of operation of the other two energy sources and to act as a distinc.tion between types of representation codes, verbal and coordinator between their activities. Whether the effort level spatial. by itself is in an optimal state depends on evaluation of the The framework proposed by Wickens summarizes factors state of adaptation of the organism to environmental demands. that demonstrably influence the pattern of interference between In other words, effort invested depends on motivation and on two concurrently performed tasks and are linked with a com­ an assessment of the situation. There is a resemblance between petition for access to a central mechanism. However, it says the role of the evaluating mechanism in the cognitive-ener­ very little on the way energetic and structural elements are getical stage model and those given to the allocation policy related to each other. This relationship between energy and construct in Kahneman's (1973) single-capacity model, although structure is the major concern in the cognitive-energetical stage in the present model the evaluation mechanism has an inde­ model proposed by Sanders (1983) and Gopher and Sanders pendent link with each of the three energy mechanisms and (1984) (Figure 41.9). hence may have a separate influence on different aspects of The cognitive-energetical stage model is an attempt to performance. integrate energetic concepts with a structural description of With an increase in overall complexity, and with less sim­ the system. Its development derived from data on the effects of plicity than was typified in the original single-channel and stressors on performance in choice reaction tasks (e.g., Frowein, central-capacity models, an integration of these complementary 1981; Sanders, Wijmen, & Van Arkel, 1982). This research frameworks may possess the degrees of freedom necessary to

STAGES

Central Encoding processing Responding

Spatial Manual «'.s-"' ""'~ 01< 15'~15'

C/l Visual w ;:: ' Verbal ~ ~cal ""' :J I I I < 0 0 l::; '"""" Spatial o0 "~.s­ ~Verbal

Figure 41.8. Conceptual framework of multiple resources. The overall level of competition and interference between concurrently performed tasks is determined by the degree of their overlap (sharing) on four dimensions: modality of inputs, type of coding operations, stages of processing, and the nature of responses. (From C. D. Wickens, Processing resources in attention. In R. Parasuraman, j. Beatty, & R. Davies (Eds.), Varieties of attention, Academic Press, 1984. Reprinted with permission.) 41-18 HUMAN PERFORMANCE

Evaluation Evaluation mechanism

Energetical mechanisms

Processing stages

Feedback Experimental S-R Time loops variables Stimuli intensity Signal quality compatibility uncertainty

Figure 41.9. Cognitive-energetical model of multiple resources. The model represents an integration of structural and energetical views. The sources of energy supplies are differentially linked to mental operations ,,r: organized in independent processing stages. Arousal and activation are the main energizers of automatic ~~ processing activity. Their operation is augmented and balanced by the effort resource. Effort represents ~~ "- the forces of voluntary attention. It is guided by the evaluation mechanism and is selectively energizing­ 1 choice ofresponse operations. (From A. F. Sanders, Stress and human performance. In G. Salrendy & E. Smith (Eds.), Machine pacing and occupational stress, Taylor & Francis, Ltd., 1981. Reprinted with permission.)

cover the processes limiting the work of the <;entral apparatus. 2.7. Controlled and Automatic Processes Which of the identified structures and energy sources will be a productive descriptor of an underlying processor is still un­ Reduction of workload with continued practice has been at­ known. The two frameworks were developed as a post hoc in­ tributed to development of automatic links between stimulus tepretation of different bodies of data. Preliminary research and response that can be operated with minimal interaction has supported a distinction between perceptual and motor re­ from the central processor. Automatic processing is defined as sources (Gopher & Navon, 1980; Gopher, Navon, & Brickner, a fast parallel process not limited by short-term memory, re­ 1982; Wickens & Kessel, 1982). Studies by Wickens and his quiring minimal processing effort, amenable to little direct colleagues and experiments from other laboratories report en­ control by the subject, and requiring an extensive and consistent, couraging results concerning the separation of tasks along mo­ training to develop. Walking, speech production, and driving dalities of processing and types of representation codes (Baddeley after years of experience are examples of highly automated & Liberman, 1980; Brooks, 1967; Moscovitz & Klein, 1980; behaviors. Reisberg et al., 1984; Vidulich & Wickens, 1981). Initial data Controlled processing is relatively slow, is mentally de- from measurement of the event-related electrical activity of . manding, requires considerable involvement of short-term the brain during task performance (Donchin, Kramer, & Wick­ memory, exhibits a large degree of voluntary control by the ens, 1983; Israel, 1980; Wickens, Kramer, & Donchin, 1983) subject, and requires little or no training to develop. In recent and experiments on the application ofstressors mentioned earlier years there has been considerable interest in the comparative in this section support arguments for the existence of several study of these two processes, accompanied by a detailed analysis sources of energetical activity. However, the general state of -of the ways in which automaticity is developed (e.g., Laberge, this knowledge is still preliminary. Even so, the formal properties 1975, 1981; Norman & Shallice, 1981; Schitfrin & Domais, 1981; of models akin to multiple-resource models are being clarified. Schneider, Dumais, & Shiffrin, 1983; Schneider & Fisk, 1983). (See Sperling & Dosher, Chapter 2.) The main requirement for automaticity is the existence of Our survey of approaches to the limitations on the infor­ consistent mapping (CM) between stimulus and response. For mation processing system concludes by considering an important example, in a task requiring the identification of target letters aspect of performance neglected by most of the models reviewed, in an array of distracting letters on a visual display, those the influence of practice on performance. It is well known that letters employed as targets should never become distractors, practice is the single most powerful factor improving the ability and vice versa. Under such conditions, a high level of parallel of the system to perform a task. Thus, nothing is as likely to processing is developed, and reaction time is reduced and appears reduce associated workload as is practice. Yet few of the models to be independent of the number. of distracting letters on the of attention and performance incorporated the effects of practice display (e.g., Schneider & Shiffrin, 1977). In contrast, if the on workload and its interaction with other factors that contribute mapping is frequently changed around, that is, targets become to workload. The framework in which this factor is beginning distractors and distractors targets (VM), response times are to receive its due respect is the analysis of the distinction between slow, the slope of the training curve is shallow, and there is a "automatic" _and "controlled" processing (Schneider & Shiffrin, linear increment in the time to identify a letter with an increase 1977). in the number of distractors on the display (see Figure 41.10.) WORKLOAD-AN EXAMINATION OF THE CONCEPT 41-19 The same phenomena have been observed when a task calls for pattern is similar. One begins with an attempt to develop a a distinction between words, or identification of noun member­ simple model, using a formal metric with an intent to use a ship in semantic categories (Fisk & Schneider, 1983). unidimensional descriptor of the load imposed on the central There are two main points at which the theory of automatic processor by a given task. The theoretical and empirical analysis processes affects the theory of workload. The first is the claim conducted within this framework inexorably reveals that the that different task components may have been automated to phenomena are much more complex and multifaceted than in­ different degrees at any given stage of prat:tice and therefore itially assumed. The load on the central processor can arise in may impose different demands on the central processor at dif­ a variety of ways, all describable to some extent within the ferent stages of practice. The second is the emphasis on a de­ framework. But i:n .each case, the unidimensional framework velopmental and flexible rather than an evolutionary and rigid gives way to a multidimensional version, one bottleneck becomes approach to the emergence of structures in the information several, one serial pathway gives way to cascading stages, and processing system. . a single unified reservoir ofresources is replaced by a multiplicity Evaluation of processing costs based upon an analysis of of such reservoirs. The corollary of this development is that the the degree oftask automaticity is clearly orthogonal to an anal­ use of a unique measure of task workload is replaced by a load ysis based upon stages of processing, types of codes, or modes profile rather than by a single measure or a quantity derived of input and output. It emphasizes the consequences ofthe nature from a unidimensional scale. In a load profile, the major com­ and length of training and attempts to explain the large vari­ ponents of a task are analyzed and evaluated in terms of the ations revealed in the processing demands of the same task main dimensions that have emerged in experimental research. between and within individuals based upon their experience 2.8.1. Dimensions of a Load Profile. To recapitulate, in (see Anderson, 1981; Laberge, 1975). A systematic consideration any attempt to define and measure workload, we must attend of this dimension incorporates the influence of practice in the to structural and energetic aspects. At the same time we cannot general model of the flow of information within the organism, ignore the consistency with which stimuli are mapped to re­ without contesting the relevance of other dimensions. The idea sponses and the level of practice in the performance of the task. of gradual development of automated structures that can be For example, consider the evaluation of an operator's ability to operated as a whole with little investment of processing effort control a device manually while searching memory for infor­ does present an alternative perspective of the structural or­ mation. Relevant dimensions for the analysis of this combination ganization of the processing system. The underlying metaphor of tasks may include modes of input and response (auditory or is one of self-organizing communication network that develops visual presentation, verbal or motor responses, etc.), type of to improve the transmission of information within the system. central codes (spatial or verbal), hemispheres' involvement, and the nature of the requirement of mental operations (e.g., feature 2.8. The Nature of Capacity Limitations extraction, short-term retention, categorization, etc.) as sug­ gested by Wickens (1983), Gopher, Brickner, and Navon (1982), This review of attempts to model the limitations of the processing Friedman et al. (1981), and Sanders (1983). In addition, one system makes it evident that in each class of models the historical should consider the portion of the task that can rely on existing

Memory Size 1 Memory Size 2 Memory Size .3

Pos Neg 1100 ..-..o Varied mappings '(i) l'---V Consistent § mappings Q) 900 g~ w :::!'t= 700 5 -----rr------____ :q t= ~ w~ 500 a: ... ~------~------. ·-----·------~ ~ 300~;-~~-----:_L~--.------~~------~J 2 4 2 4 2 4 I FRAME SIZE (number of letters) .~ Figure 41.10. An example of performance based upon consistent and varied mapping. Subjects were asked to detect the presence or absence of a letter from a predesignated set in visual frames including different number of letters. Under consistent conditions, the same set of seven letters always served as targets, and another set of eight letters were always used to select distractors. In varied mapping, the same letters switched roles, serving as targets in some blocks and distractors in others. The figure depicts mean reaction times for correct responses as a function of the number of digits on the display. Data for both negative (absence) and positive (presence) trials are presented. Compare the linear increase of response time with the increase in the number of elements in varied mapping, to the flat slopes under consistent mapping. (Adapted from Schneider & Shiffrin, 1977, experiment 2.) 41-20 HUMAN PERFORMANCE or developing automatic components (Schneider, 1983) and weigh that never reaches awareneBB (e.g., Dixon, 1981; Marcel, 1980; the degree of dependence on controlled processes. Underwood, 1979). The surprise expressed at these findings is Analysis of effort and energetical demand can follow the quite puzzling, since it is evident that much of the brain's in­ scheme processed by Pribram and McGuinness (1975) and formation processing is never available to conscious awareness. Sanders (1983). A relevant consideration is the ability of the The numerous vegetative functions requiring information pro­ information intake processes and response activation mecha­ OOBBing are clearly beyond the ken of awareness. The limitations nisms to function adequately with minimal involvement of the on the scope of consciousness are such that consciously manip­ central effort mechanism. This question is not unrelated to the ulating each of the muscles involved in the act of standing analysis of controlled and automatic processes. It is reasonable would present us with a workload exceeding the capacity of all that a development ofdirect links between encoding and response humans. It makes just as much sense that we cannot hope to activation and reduced dependence of these processes on the accomplish within the limited capacity of consciousneBB the coordination of voluntary effort can be achieved with increased incredibly complex task of searching through memory or de­ automaticity. The energetical pools of arousal and activation termining the grammaticality of our sentences. Indeed, when that regulate these proceBBes can thus obtain independence forced to analyze one sentence explicitly and consciously, we through the process of increased automaticity. This line of rea­ may find the task leaving no spare capacity at all. And yet we soning is an important bridge between energetical and structural all generate continually, and relatively correctly, an endless constructs in the study of workload. It can also capture the stream of sentences, not evaluated consciously for their gram­ dynamic properties ofthe workload phenomena in which a person matical structure. In short, information proceBBing is usually may act more like a single processor in the beginning of training unconscious despite the important role played by consciousness. and gradually develop into a multiple resource mode when pro­ The implications of this perception to the evaluation of workload cessing mechanisms and energetical pools gain sufficient in­ measures will become evident as we discuss "subjective" mea­ dependence. We can further speculate that the structural di­ sures of workload. mensions and processing stages emerging from eXperimental As a specific structure, consciousness does play a role in research as significant qualifiers of central processor work may most of the models discussed. Broadbent (1958, 1971) identified also represent the most natural organizing framework within conscious attention with the limited-capacity processor of his which automatic segments of behavior or action schema develop. model. Welford (1967) has associated it with the operation of 2.8.2. Conscious Control and Allocation Policy. We con­ the decision mechanisms. The elaboration of these single-channel clude this section with remarks concerning the role of con­ views, and the introduction of multiple-resource frameworks, sciousness in this analysis of the limitations on human per­ required further qualification of these claims. At present, the formance. For William James, as noted in Section 2.1, the executive and supervisory aspects of consciousness appear best limitations on attention were due to a competition for the pos­ to correspond in these models to the discussion of attention session of consciousness. As analysis of human performance strategies and allocation policy. The cost of conscious activity continued, this framework unifying attention and consciousneBB seems to be most related to the topics ofperformance organization disintegrated. The status of consciousness in current models and task automaticity. Accordingly, the natural candidates for and its relationship to the study of workload have become quite teaming up with this construct are the attention policy mech­ complex. In effect, while consciousness is clearly an important anism in Kahneman's (1973) capacity model, the effort lmd element in all tasks, current models of performance ascribe to evaluation mechanisms in the cognitive-energetical stage model it only a partial role. It is but one structure deployed in the (Gopher & Sanders, 1984; Pribram & McGuinness, 1975), and service of behavior. It would be as much an error to assume the processing activity labeled controlled processes in network that our interest is exclusively in consciousness as it would be models of the system (e.g., Norman & Shallice, 1981; Schneider, to ignore it altogether. Successful performance is as dependent 1983). on an operator's conscious application of procedural knowledge But, as we know, the conscious apparatus is limited and is as it is on his or her nonconscious implementation of automated easily consumed, as already argued by William James (1890). sequences developed over the period in which task expertise Indeed, there is an obvious relationship between the processes was acquired. classified by Schneider and Shiffrin (1977) as "controlled" and The mental activity included under the general categories consciousness. For a discussion of this analogy see Broadbent of conscious experience and voluntary control is still not well · (1982), Posner (1978), Logan (1978), and Laberge (1981). The defined and understood in cognitive psychology (see Carr, 1980; major contribution of Schneider's work is not in relabeling the Posner, 1978). From the point of view of our own experience distinction between the conscious (controlled) and the non­ I consciousness is primary, and we view ourselves as guided by conscious (uncontrolled), but rather in demonstrating that it is the content of our consciousness. It is, therefore, remarkable possible to specify those attributes of a task that would allow that conscious experience does not play a formal functional role transforming it~ performance from the controlled to the auto­ in any of the new approaches describing information found matic mode. within the organism. Indeed, when treated at all, consciousness In conclusion, what are the general guidelines of the the­ plays the role of a specific functional element (see Donchin, oretical analysis carried out in this chapter to the development McCarthy, Kutas, & Ritter, 1983). Thus some authors equate of a methodology to assess workload? The original objective of consciousness with the content of the short-term, primary this assessment has not changed. We still aim at uncovering memory (e.g., Atkinson & Shiffrin, 1971). Others assign to it the limits of the central processing mechanisms and argue that the role of an internal programmer, executive, supervisor of such a measurement is essential to improved prediction of be­ behavior, and go on to specify its role in specific control of variety havior. What have been changed substantially are the scope of processing and execution tasks (e.g., Logan, 1979, 1980; and details of the proposed assessment. A global measure of Mandler, 1978, 1983; Posner & Snyder, 1975). workload does not appear achievable. Instead, any claim or A growing body of experimental evidence demonstrates statement about workload should be accompanied by an attempt the obvious role played by all levels of the processing system to identify the sources of load, the mechanisms involved and WORKLOAD-AN EXAMINATION OF THE CONCEPT 41-21 affected, the levels of performers' experience, and the criteria workload will not be measured, because our knowledge of the of behavior on the task under which this statement is assumed capacity of the operator precludes this particular loop from con­ to hold. Although the generality and caution expressed in the sideration. above approach may give the impression that workload as­ It is critical to distinguish between measurements of subject sessment is back on the drawing board, this is not the case. In capacity or analyses of a task's structure and measurements of this chapter are enough leads, supported by the results of em­ workload. The critical distinction lies in the degree to which pirical tests, to direct the measurement procedure to concentrate workload measures are made in the context of, and in reference on the most effective dimensions. Clearly, a detailed analysis to, the interacting combination of subject and task. Strictly of the task in question is the basis of any measurement attempt. speaking, workload must be defined anew for each subject, and Our proposed approach is undoubtedly less concise and parsi­ for each set of prevailing circumstances. We relax this rule on monious than one would like, but is a simpler one PoSsible, the assumption that an individual can be assumed to remain given the richness and complexity of human mental activity? (subject to the effects of training and practice) essentially sta­ tionary. Thus, generalizations can be made across occasions. 3.2.1. Normative and Descriptive Approaches. Assuming 3. TECHNIQUES FOR THE MEASUREMENT OF for design purposes that human operators are a homogeneous WORKLOAD class, a properly selected group can be considered to satisfy certain capacity criteria, and workload assessment can be ac­ 3.1. Criteria for the Evaluation of Workload complished for these operators even ifonly a few representatives Measures are assessed. Moreover, if the task is then assigned only to individuals within the class for which workload was assessed, This section discusses classes of measures that have been pro­ then the measurement of workload for this class is complete. posed for, and are employed in, the measurement of workload. The problem is that the trade-off between design and selection All the procedures reviewed have been proposed explicitly for can not be avoided. Designers can develop systems that can be the measurement of workload regardless of theoretical under­ used, at a minimal workload, by most operators, or systems can pinnings of the measurement. Measures of workload have been be designed for effective use by the select few. In the first case, developed, for the most part, in a pragmatic context. This ex­ the design may be costly, but the workload is small. In the plains, in part, the lack of theoretical consistency in the analysis second case, the design may be simpler, but the cost of selection of this concept. The approach has often been intuitive and mea­ and ttaining may prove excessive. sures have rarely been evaluated for their validity and reliability. The assumption of a standard operator is typical of a nor­ Indeed, face validity (a term that lends dignity to an unques­ mative approach to the measurement ofworkload. In a normative tioning reliance on one's intuitions) is often considered para­ approach, one assumes that the operators satisfy some criteria mount in evaluating workload measurement. Thus not all pro­ for capacity, motivation, rationality, and other related attributes. cedures that purport to measure workload adequately will satisfy The task is analyzed for its challenges to the standard operator, the criteria derived from the analysis in this chapter. It is our and conclusions are drawn regarding the workload in the loop. intent to review measurement techniques within the framework This approach underlies what, in industry, is probably a prev­ of our own interpretation of workload. The reader is referred alent method for assessing workload. We refer to time-line to O'Donnell and Eggemeier, Chapter 42, for a fairly complete analysis, an inherently open-loop technique that focuses on the catalog of available measurement techniques. Here we discuss structural description of tasks. Its practitioners have acquired only a few techniques that illustrate classes of problems needing the skill to examine tasks rather minutely and to identify the attention in the design of workload measures. components of the task. The specific actions that must be taken at any instance along the epoch of task performance are ascer­ 3.2. Workload as a Property of the Operator-Task tained. The product is a. chart that serves to document the Loop structure of a task. If we assume a normative operator, the designer of a system may be affected in the choice of assignments The most fundamental assertion regarding the measurement given to the system's operators by the number of activities of workload is that workload is an attribute of the loop between called for according to the time-line analysis. The assessment an operator and a task. Workload is a hypothetical construct of the degree to which a task may or may not be more difficult, intended to capture limitations on the operator's information as judged by the time-line record, depends on our assumptions processing apparatus as these are viewed from the perspective regarding the capacity of the operators. In other words, time­ of some assigned task. The critical implication of this assertion line analysis cannot be considered a procedure for the assessment is that it is not particularly meaningful to measure workload of workload except for those restricted circumstances in which in an open loop. One cannot specify workload associated with the task will be presented solely to individuals who fit certain a task without reference to the operator, and one cannot relate selection criteria. Moreover, the assumption is also made that workload to an operator without this operator's being in the. there will be no interaction between the task and the operator matrix of the task-operator loop. that changes the character of our statement. Relevant operator and task characteristics having an im­ From the analysis presented in Section 2, it should be evident portant effect on workload characterize a closed loop. The op­ that it is unlikely that a global standardized measure will capture erator's capacities and skills can be assessed and the structure the complexity or the human information processing system. of the task desginated a priori or ascertained by means of a This is true at least as far as the limitations on the system are task analysis. When the information about the task or the op­ concerned. Appealing as the mystique of the superhuman ex­ erator suggests that the task-operator loop will accomplish perimental pilot is, the reality is that all humans are equipped nothing, we do not bother to close the loop and therefore cannot with limited-capacity processors, and there is considerable var­ measure workload. Section 1.1 mentioned leaping across the iance among persons in the effectiveness of these processors Grand Canyon as a clearly difficult task with respect to which and in the ability to use them in the service of rather complex 41-22 HUMAN PERFORMANCE tasks. A careful monitoring of workload as it appears in the task. Thus, the subjects are judging the interactions between operator-task loop is therefore mandatory to ensure proper sys­ themselves and the system. tem design. Subjective methods of measurement have gained popularity 3.2.2. Overview of Section. The remaining pages of this during the last decade. Indeed, the Cooper-Harper scale (see section will consider several classes of measures of workload, O'Donnell & Eggemeier, Chapter 42), which allows pilots to assuming that within any task-operator loop the operator can record their impressions of workload, is the technique commonly be represented as an ensemble of processing facilities (structures used in the aircraft industry to measure workload. and resources are other words used to describe the relevant There are a number of ways by which the subjective mea­ elements of the operator). These are the totality of mechanisms, sures are acquired. Usually subjects are presented with elaborate energetical sources, and limited processors discussed in previous rating scales on which they rank the demands associated with sections. In most tasks, the operator must deploy one or more the tasks along a wide variety of dimensions. Thus subjects of these facilities, and performance is related monotonically to may be asked to indicate the extent to which they felt time the level of deployment of these facilities. Task difficulty can pressure, or to evaluate the physical activity, or the mental be expressed in terms of the demands on these facilities by the effort, or task complexity (e.g., Casali & Wierwille, 1982; Hart, structural properties of the task. The interaction between task Childress, & Bartolussi, 1981; Wickens & Yeh, 1982). In a sense, and operator is manifested by the degree to which these facilities the subjects are performing a task analysis when they are mak­ can be made available given their inherent limitations. The ing these ratings. The questions asked are putatively about the assessment of these interactions is the assessment of workload. task and components of the task, rather than about workload. There are two general classes of measurement techniques: The individual rating scales may be combined into an index of those that focus on the loop and provide some global measure subjective workload. The combination can follow different rules. of the interaction, and those that attempt to be specific regarding Commonly the investigators employ a multiple, conjoint, scaling the locus of the interaction. The first body of techniques assumes technique for combining the measures (e.g., O'Donnell & Eg­ that, despite the structural complexity of the operator and the gemeier, Chapter 42; Reid, Shingledecker, & Eggemeier, 1981). detailed form in which limitations on the processors can manifest Despite its "subjective" aspects, this remains a task analysis themselves, it is possible to obtain global measures of workload. procedure because it requires a decomposition and a separate These approaches are based on theoretical positions discussed evaluation of the components of the task. Quite a different ap­ in Section 2 under the single-channel, single-bottleneck, or sin­ proach takes as its object of measurement the perceived work­ gle-resource concept. As seen in O'Donnell and Eggemeier, Chap­ load. The subjects' responses are given as a magnitude esti­ ter 42, such global measures have active and persuasive ad­ mation. This procedure is adopted from classical psychophysics herents. In this category fall measures ];,ased on subjective where the psychological dimension is identified as workload judgment and performance-based measures that focus on the while the physical dimension is anchored in the structure of performance of the assigned task as an indicator of the workload the task (e.g., Borg, 1978; Gopher & Braune, 1984). in the system. There is also a class of "physiological" mea­ Subjective measures are easy to obtain and excel in face sures, based primarily on arousal, that have been used as global validity; there is a compelling sense of relevance in a measure measures of workload. that seems to depend directly on the subject's actual experience The second category includes a variety of procedures de­ of workload. Presumably, subjects are keenly aware of the effort signed with an interest in diagnostic (Wickens, 1984) measures needed to cope with task demands. Indeed, panelists, in a sym­ of workload. Here the commitment is to view the information posium held in 1979 in which the measurement of workload processor as a complex structure, with emphasis on its stages, was examined, concluded: "If the person tells you that he is whether serial or cascading. There is a commitment to multiple loaded and effortful, he is loaded and effortful whatever the resources in this class of measures and to an interest in the behavioral and performance measures may show" (Moray, Jo­ fine structure of the interactions. We will review in this category hanssen, Pew, Rasmussen, Sanders, & Wickens, 1979, p. 105). a family of secondary-task techniques, beginning with an anal­ These brave words are clear enough, but the theoretical and ysis of the conceptual basis of secondary-task measures, and empirical status of subjective measures of workload remains outline some of the advantages and disadvantages of the pro­ vague. Particularly obscure is the validity of the measures. cedures. We shall then discuss the use within the secondary­ Advocates of subjective measures appear to assume that face task paradigm of what are often considered ''physiological" validity is sufficient. Thus, there are a very few studies in which measures-the event-related brain potentials (ERPs). Within actual rather than face validity is assessed. The degree to which the context of workload assessment, the ERPs are used as a any subjective measures can be used to account for variance in form ofnonovert behavior, and they are logically equivalent to performance remains unknown. Equally obscure are the re­ other measures of secondary-task performance. The key ad­ lationship between subjective measures and psychological pro" vantage of these covert responses is that they allow an assess­ ceases and the behavioral phenomena of which they are supposed ment of workload in domains that are opaque to assessment by to be a manifestation. more traditional techniques. To appreciate the problems with the use of subjective mea­ sures, consider the following hypothetical example. You can 3.3. Subjective Measures ask a cab driver in any major city to assign a number describing the workload associated with driving his cab during the rush Subjective measurements of workload are made whenever sub­ hour. You may facilitate the cab driver's task by suggesting jects are asked for a direct estimate of the workload they ex­ that the load level of driving during early morning hours, when perience during the performance of a task. The term workload the streets are empty, equals 10. The driver is likely to ponder is rarely used when instructing the subjects; instead they are for a while and then produce a number. Assuming that this usually asked to report the "difficulty" of the task. However, person also plays chess, he can also be asked to assign a number the target is the experience of difficulty during execution of the to the workload experienced during a recent game. Again, the WORKLOAD-AN EXAMINATION OF THE CONCEPT 41-23 driver will generate a number with little difficulty. But what auditory monitoring task. In both studies, performance was not is the meaning of these two numbers? What do they tell us correlated with the workload ratings despite an organized and about the processing mechanisms and the degree to which they consistent ordering of the experimental conditions according were pushed to the limit in the two tasks? Are these numbers, to workload and performance. Some investigators obtain in­ so easily gotten, valid and reliable estimates of workload? What termediated results. They report a correspondence between the is the theoretical and practical significance of a person's ability subjective estimates and measures of performance on some of to compare and assign numbers expressing the result of the their independent variables, yet for other variables there is comparison to such divergent tasks as driving a car and playing partial dissociation between the workload index and performance chess? (Wickens & Yeh, 1983; Vidulich & Wickens, 1983). 3.3.1. Consistency of Subjective Estimates. Any attempt It is noteworthy that Gopher and Browne (1984) report a to examine the theoretical and practical basis of subjective es­ high correlation (0.93) between subjective measures and a task timates must consider a striking aspect of the relevant data, analysis that leads to an index of task difficulty. This analysis, the remarkable consistency with which the estimates are given. developed by Wickens, is based upon salient features of the Clearly subjects do not assign workload values to tasks at ran­ task. The index has four dimensions: (1) familiarity of stimulus dom. On repeated exposures to the same tasks, even ifembedded (e.g., letters are familiar, random dot patterns are not); (2) in a wide variety ·of other and very different tasks, subjects concurrency of tasks (single- vs. dual-task conditions); (3) dif­ manifest an impressive level of consistency in the choice of ficulty (e.g., number of elements in a memory set, delay in numbers to describe the difficulty of the tasks. For example, recall); and (4) resource sharing (e.g., same vs. different input Gopher and Browne (1984) gave subjects a battery of21 tasks, modality, same or different coding principle). It is possible that varying input modality (visual, auditory), type of mission the subjective estimate captures an objective and external as­ (tracking, memory search, dichotic listening, etc.), and mode sessment of a task. It does not, however, serve to predict how of response (verbal, manual). Fourteen were single-task and well, or how poorly, a subject will do. One may infer from this seven were dual-task conditions. The whole battery was per­ observation that the responses that subjects give when asked formed three times, and subjects were asked to give their es­ for an estimate of the difficulty of tasks are based hirgely on a timate of load following performance on each task. In addition "cognitive" analysis of their knowledge of the task rather than they were asked at the midpoint and at the end of the experiment on an assessment of the way they actually interacted with the to evaluate the load of each of the conditions in the battery in task. a single instance. The intercorrelation coefficients for the sub­ 3.3.2. Theoretical Considerations in the Interpretation and jective load profile of the 21 tests, in the five rating instances, Use of Subjective Measures of Workload. It is of interest to were all above 0.90. Reliability coefficients of 0.95 and above consiqer the processes manifested by the subjective measures were also found in a random split-half of the subjects. Similar and to assess their relation tot~ processes of interest in work­ levels of consistency were reported by Hallsten and Borg (1975), load measurement. By their very nature, subjective measures who compared subjective ratings of items from a standardized depend on the content of consciousness. The operator, after all, intelligence test. Supportive evidence can also be found in Casali is instructed to report the perception of interaction with the and Wierwille (1984) and in Hauser, Childress, and Hart (1982). task. This report can only be based on those aspects of the The consistency of the relationship between the task and the interaction of which the subject is aware. Evidently, these mea­ subjective value assigned to its workload is preserved under sures will serve their purpose only to the extent that the critical constrained and well-structured rating methods, but also if aspects of the interaction between the subject and the task are magnitude estimation techniques are employed, and the subject available to consciousness. is generating his or her values ad lib. In fact, Gopher and Browne, There is considerable controversy regarding what is and using ratings that subjects made on the 21 different tasks, were what is'not available to consciousness. In this discussion it is able to describe the complete set by fitting a single power function useful to recall the categorization of the contents of consciousness to the data. The roots of this consistency remain to be determined. proposed by G. Mandler (1983). Three main categories of events More important, how is the workload indicated by these measures are identified by Mandler: (1) We are conscious as we acquire related to the overt performance? new knowledge and behavior; (2) conscious processes are active The overall correspondence between subjective measures during exercise of choice and judgment; and (3) conscious pro­ and performance measures is not very high, with equal pro­ cesses exercise an important function during ''troubleshooting." portions of positive and negative reports from experimental This view is reminiscent ofthe functional relation that Broadbent works. Consider a few illustrations. Hallsten and Borg (1975) (1982) proposed between consciousness and controlled processes, required subjects to rate workload in the solution of problems as the term was defined by Schneider and Schiffrin (1977). extracted from an IQ test. They report that the ratings correlated If one accepts these classifications of the nature of con­ reasonably well with problem solving (R was in the range 0.70- sciousness, one has to accept that they constrain the information . 0.80). Another example is a study by Bratfisch, Borg, and Dornic processing activities-variations in which are most likely to (1972), who report high correlations between subjective mag- be captured by the subjective measures. It seems clear that the nitude estimations of Workload and the objective difficulty of measures are likely to reflect those aspects oftasks that require problems in the Raven matrices test. A strong association be­ a guided (voluntary) involvement in coping with novelty, gen­ tween subjective ratings and performance variations in a difficult eration of new action plans, selection among alternatives, and manual control task was reported by McDonnell (1968). In con­ commitment of the limited-capacity working memory. In short, trast, many investigators report a dissociation between sub­ those activities that regulate the allocation ofvoluntary attention jective estimates and measures of performance. Hauser, Chil­ include problems in performance to which the service of this dress, and Hart (1982) assigned subjects a tracking task using mechanism is called. four levels of difficulty; they also included four levels of difficulty These are indeed important components of task perform­ on a memory search task, as well as a tiine estimation and an ance, but the relative share of this range of activities is small _ 41-24 HUMAN PERFORMANCE

in the totality of the human information processing system as try to instruct the subjects to produce estimates during per­ it copes with the demands of a specific task. Clearly, there is a formance, but this procedure may constitute a demanding sec­ substantial ensemble of routine, highly practiced tasks such as ondary task that will interfere with performance (Ericcson & driving, speaking, or eating that use a very small component Simon 1980, 1984). of processes involving consciousness. At the same time, virtually every task of intermediate complexity includes elements that 3.4. Performance Measures, Primary Task require conscious activity. It is reasonable to assume that the degree to which the subjective ratings predict performance is In our discussion of subjective measures, we emphasized the a function of the proportion of controlled, conscious activities frequent lack of correlation between these measures and the necessary for task performance. This hypothesis explains the manner in which the operator performs the task. Those com­ fact that subjective measures sometimes correlate with and ments implied that the operator's ability to perform a task sometimes are dissociated from performance. This view is also ought to serve as a validating measure for any other measure consistent with the high correlation found by Gopher and Browne of workload. This attitude is actually quite common. It appears (1984) between subjective measures and Wickens's index of straightforward that the level of performance an operator analysis of task difficulty. The four dimensions of the Wickens achieves on any task is a rather direct measure of the difficulty, · index may represent those features of the task that attracted and by implication, the workload, associated with the task. It the attention ofsubjects and influenced their voluntary allocation is recognized that fluctuations in performance may reflect of resources. However, this component of processing is only a changes in motivation. However, in this discussion we assume fraction of the total processes that determine actual performance. that neither motivation nor the basic ability changes. We are The relationship might have been different with problem:solving concerned only with changes in performance due to limitations tasks, or if variations in the standard of desired performance on the information processing system. In this case, it would had made the distinguishing attributes among experimental appear natural to use the performance on the primary task as conditions (e.g., Gopher, Navon, & Brickner, 1982; Gopher & a measure of workload. Navon, 1980). We will not review here specific measures of primary task In the measurement of workload, it would appear that sub­ performance. Their implementation is fairly obvious, and the jective measures have limited but important functions. They reader is referred to O'Donnell and Eggemeier, Chapter 42, for provide converging information, they may help clarify the di­ more detail. This section focuses on the factors that reduce the mensions of the tasks, and they would be almost entirely suf­ utility of direct measures of performance in the study of work­ ficient in tasks that depend primarily on controlled processes. load. However, their limitations must be recognized. An alternate When the difficulty of a task is increased, more resources view (e.g., Logan 1978; Norman & Shallice, 1983; Schneider, are required by default to maintain the same level of perform­ 1984) is that all the effects of attention should be looked upon ance. Ifthese resources are available, performance may remain as a directed (or controlled) biasing force of limited power, unchanged. Thus, even though no change in behavior is observed, sometimes also described metaphorically as a gain factor. Only workload has increased, and therefore the evaluation of observed those processes that use these biasing factors ''load" the infor­ "direct" measures is quite different. mation processing systems; the rest are data driven, automatic, Consider a study in which the speed of a target is increased and can flow in parallel at no cost to the central processor. Such for a subject tracking the movement of a target on a computer automatic processes do not limit the information processing screen with a dual-axis hand controller. We can expect an in­ system and therefore are not relevant to the measurement of crease in target speed to be accompanied by an increased demand workload. According to this logic, a dissociation between sub­ for processing and response efforts. If tracking accuracy is not jective measures and performance is due largely to failures in changed, we can assume that additional resources were indeed performance that do not result from limits on the central pro­ deployed to maintain this level of accuracy, but we have no cessor. clue from the behavior itself as to how much and how loaded 3.3.3. Methodological Considerations. The use ofsubjective the new tracking condition is. If an increase in tracking error measures presents yet another problem. Current techniques of is observed, we still do not know whether those decrements subjective measurement are retrospective in nature. That is, reflect all the "costs" or only part of the "costs" to the system the operators are asked to evaluate the task some time after (see Gopher & Navon, 1980; Navon, Gopher, Spitz, & Chillag, performance. Thus, even accepting the theoretical validity of 1984, for a discussion of attention demands of tracking tasks). these measures,their utility is constrained by the limited ca­ Moreover, under the multiple-resource notion, decrements may pacity of working memory. It is reasonable to assume that a result from exceeding the limits of one or several different pro­ certain portion of the information available to the operator cessors. We usually have little indication in direct performance l while performing the task either is not available or may have measures as to which of the underlying capacities has been been distorted by the time the information is sought. This prob­ overloaded. lem is common to all measures that depend on verbal reports A complementary problem to the evaluation of task difficulty as data (for a review and theoretical treatment see Ericcson & manipulation is the need to scale changes in required standards Simon, 1980, 1984; Nisbett & Willson, 1979). It may very well of performance so that they match the amounts of invested be that the subjective measurements of a task are determined resources. Logically, the marginal costs of every additional unit by some salient features that are compared to general knowledge of performance will not be the same across the whole range of bases made of the subject's past experience. Those knowledge task performance, and a greater effort per unit of performance bases are relevant, but only in a general sense, to the load of will be demanded at higher levels. We follow Norman and Bobrow the currently performed task. This may be another source of (1975) and Navon and Gopher (1979) in proposing the idea of dissociation between performance and subjective measures. a hypothetical performance resource function (PRF) that relates There are no data that pertain directly to this issue. We may actual performance to the amount of invested resources and WORKLOAD-AN EXAMINATION OF THE CONCEPT 41-25

the suggestion of changing rather than fixed costs for additional A key problem in the interpretation of these data and in units of performance. Within this framework, comparisons be­ the utility of these measures of workload is related to the spec­ tween the resource costs of different levels of performance should ificity of the response. In a way, the problem is quite similar consider their location on the PRF. In the example of the tracking to the problem one encounters in the attempt to use psycho­ task, if a subject is able to comply with our increased accuracy physiological measures as indexes of deception. It is quite clear requirement, we can assume that it is done with additional that the stress associated with deception does manifest itself costs to the system, but we are still unable to determine the in an ensemble of observable physiological changes reflecting amount and nature of these additional costs. If performance changes in "arousal." It is however equally clear that exactly does not improve, it is again unclear whether no additional the same changes may be observed in connection with stress resources can be recruited to tracking performance or whether caused by many factors other than deception. Thus, it is not performance reached an upper ceiling such that additional re­ the physiological measure per se but rather the context within sources would not improve accuracy. which it is recorded that determines the value of any psycho­ In summary, direct measures of performance on the task physiological measure. The investigator must create a setting of interest are usually a poor indicator of mental workload within which the physiological changes, and the arousal they because they often do not reflect variation in resource investment signify, can be interpreted in an unambiguous fashion. It is for due to difficulty changes, they do not diagnose the source of this reason that we are not entirely persuaded of the value of load, and they do not make possible a systematic conversion of global measures of arousal in the assessment of workload. A performance units into measures of relative demands or load discussion of the manner in which psychophysiological measures on the p~cessing system. The attempts to develop a workload can be used in a more specific manner is provided in Section measurement technique based upon performance measures have 3.9. been almost exclusively limited to the strategy of studying pat­ terns of interference in the concurrent performance of tasks 3.6. Specific Measures under dual-task conditions. There are two main variants of this experimental paradigm: the secondary-task technique and the It is not surprising that approaches to workload measurement performance operating characteristics (POC) methodology. (See based on the assumption that workload is an entity that globally also O'Donnell & Eggemeier, Chapter 42; Sperling & Dosher, characterizes the interaction betw€len an operator and a task Chapter 2.) have been found wanting. We have reviewed the many attempts to define the concept and to map the limitations on the human 3.5. Arousal Measures information processing paradigms. The conclusion seems clear. Because the information processing system is diverse in its A class of direct measures that we shall mention briefly in this structure, it can go about its tasks in many different ways. section is the ensemble of psychophysiological measures. These Multiple and not altogether predicatble strategies may be used measures are obtained by recording, in general noninvasively, by different operators under different circumstances, for different signals generated by the activity of some bodily system. Beatty tasks. This diversity allows system limitations to appear in and his colleagues, for example, have reported that the diameter different guises and to be circumvented by operators who will of the pupil is particularly sensitive to variations in mental employ all handy stratagems to ensure that they cope with effort (Beatty, 1979). Various cardiovascular measures of effort their assigned tasks. have been examined, and advocated, by numerous investigators Once the system is so conceived, it is natural to examine (e.g., Mulder, 1979; see also Hart, in press). O'Donnell and Eg­ the possibility that measurement of the limitations would ad­ gemeier (Chapter 42) review this literature in some detail. dress specifically the full repertoire of possible limitations. If In general, the assumption underlying this work is that as workloa~ is a consequence of a diverse set of interactions taking the demand for mental effort increases various bodily systems place in different "structures,'' it is critical that workload be are activated, or "aroused" in the process of marshalling resources measured in a structure-sensitive manner. Measures must in the service of this increased effort. This arousal may be man­ therefore be designed to allow monitoring of the different struc­ ifested through increased cardiovascular activation. It may also tures whose stress under excessive demands brings forth the be manifested through activation of the parasympathetic system increase in workload. We review here two closely related ap­ so that pupil dilation is evident. The signals are readily re­ proaches, both from the secondary-task technique. cordable and can be obtained with minimal disruption to the In this measurement paradigm, the workload task associ­ performance of the task. The recording of cardiovascular activity ated with a given task, the "primary" task, is measured by requires merely the attachment of a few electrodes to the body. assigning the operator another task to perform concurrently Pupillary activity can be recorded without any ~ttachments to with the primary task. The operator is told that this .new task the body, though the equipment required to monitor the pupil is "secondary" in importance. The primary task must be per­ may be somewhat cumbersome (see Hamilton, Mulder, Strasser, formed to the best of the operator's ability, even if this means & Ursin, 1979). neglecting performance on the secondary task. Fluctuations in The validation ofthese "physiological" measures ofworkload performance of the secondary task are therefore assumed to has been based on the recording of changes in the measure reflect fluctuations in workload associated with the primary under conditions in which control variation of workload is in­ task. An assumption underlying this technique is that per­ duced. Beatty's work has been particularly elegant, as he has formance on any task depends on the measure of the resources been able to exercise rather tight control over the pupillary allocated to that task. It is further assumed that the pool of diameter by varying the cognitive load imposed on the subject. resources is fixed in magnitude. From these two assumptions, Thus, the pupil will dilate in a rather precise manner as a it follows that there is a reciprocal relationship between the function of the requirement to perform mental arithmetic, or performance in each of the two tasks. Deterioration in the per­ as a function of the selective attention demands of the situation. formance of the secondary task must be due to an increase in 41-26 HUMAN PERFORMANCE the resources drawn to maintain the level of performance on formance criteria, and so on. They may concentrate on the char­ the primary task. So presented, this class of measures assumes acteristics of the secondary task, the primary task, or both. a general, undifferentiated pool of resources. It is possible to Examples of this logic can be found in Gopher, Brickner, and extend this logic to a system characterized by pools of specific Navon (1982), (R.eisberg (1983), and Wickens and Sandry (1983). resources by selecting secondary tasks assumed to draw from The original logic of the secondary-task technique is maintained one or another specific pool (Knowles, 1963; Rolfe, 1973.) within each of the comparisons of patterns of interference among 3.6.1. Selection of a Secondary Task. Thus the selection a single pair of concurrently performed tasks. At the same of a secondary task depends on the view adopted regarding the time, the researcher examines the emerging pattern across all nature of the central procesSor. Under the single-capacity, un­ dual-task combinations. differentiated resource models that dominated the theoretical 3.6.2. Types of Interference and Lack of Interference. Given thinking during the fifties and sixties (Broadbent, 1958; Kahne­ the above considerations of selective interference, an important man, 1973; Moray, 1967), it was logical to search for a single decision for an experimenter when selecting a secondary task standard secondary task that could constitute a common ruler is whether the interest is in maximizing the interference between along which all other tasks could be scaled and compared. The the concurrently performed tasks or in searching for a paired hope to find such a task guided the work of Michon (1966) when task that can be performed in parallel uninterrupted. Max­ he proposed his tapping task (Michon & von Doore, 1967), the imization of interference appears to be more consistent with research with critical tracking tasks conducted by Jex (1967, the original secondary-task rationale, in which the second task 1976; Jex et al., 1966), and the development of a workload index is added to saturate the capacity ofthe system, create an overload, based upon pupil dilation as attempted by Kahneman (1973). and enable one to scale the demands of the primary task. It is, However, it has proven impossible to develop a standard task. therefore, somewhat surprising that a lack of obtrusiveness of It was particularly difficult to find a task whose assignment the introduction of a secondary task to the performance of a would not interfere with the performance of all primary tasks. primary task has been identified by several authors as a highly It has also become increasingly clear that tasks are selectively desired property of a good secondary task (e.g., Casali & Wier­ sensitive to interference by one group of tasks and indifferent willa, 1982; O'Donnell & Eggemeier, Chapter 42; Williges & to members of other groups; this fact was an important con­ Wierwille, 1979). How can this aspiration coexist with the main tributor to the refutation of a single-capacity view and the de­ thrust of a technique that advocates the study of interference velopment of the "structural" approaches to the measurement patterns as its main tool? There appears to be an additional, of workload (e.g., Gopher, Brickner, & Navon, 1982; McLeod, implicit assumption by these authors. It is assumed that per­ 1977; Wickens, 1976).Within the structural framework, these­ formance on the primary task can be completely secured when lection of a secondary task is much more complex. The exper­ the secondary task is introduced, such that all consequences of imenter accepts in advance the argument that different sec­ the overload would show up only in secondary-task performance. ondary tasks can tap different components of the task under This is, of course, a very convenient state of affairs from a consideration. The demands imposed by some secondary tasks methodological viewpoint, but it also requires the assumptions can be totally uncorrelated with those of the task in question. that: (1) subjects are in full control of the allocation of their The burden is on the experimenter to clarify the exact aspect processing efforts among the two tasks (see Navon & Gopher, of the workload being measured. It is then necessary to identify 1979); (2) the introduction of the second task does not cause an a secondary task that can best tap this dimension. Several au­ important change in the nature of the performance conditions thors provided extensive reviews ofthe secondary-task literature of the primary task (e.g., Duncan, 1979); and (3) there are no that demonstrate clearly the pronounced consequences of such minimal processing costs of the secondary task that are created a selection strategy over a wide variety of primary- and sec­ by its mere introduction (e.g., Gopher, 1981; Norman & Bobrow, ondary-task pairings (e.g., O'Donnell & Eggemeier, Chapter 1975). In most situations it is quite likely that one or more of 42; Odgen, Levine, & Eisner, 1979; Wickens 1980; Williges & these assumptions are violated. The hope that the performance Wierwille, 1979). on the primary task will be protected and the secondary task One problem with trying to fit the selection of a secondary will tap a relevant common dimension has turned out to be task to probe the processing limit of interest is that we do not rather naive. To uncover the demand composition of a task, it have a good model of the central processor. Hence, all hypotheses seems more realistic to start with a situation where the existence on the mapping of task dimensions and performance require­ of a strong interference between tasks has been demonstrated ments to the limits of this processor are highly speculative at for both tasks and work our way through the origins of this this time. Favorite dimensions are: modality of stimuli, mode interference. of response, coding principles (spatial/verbal), memory load (both A different perspective of the dual-task pradigmis emerging working memory and long-term memory), time characteristics when the main objective of the measurement is the ability to (self-paced/externally paced, continuous/discrete), and level of perform, in parallel, two tasks with minimal interference. Here; practice. A second problem arises from the fact that every task, the main thrust is not a study of the processing demands of one whether primary or secondary, is a compound of demands that of the tasks in the pair, but rather the global efficiency of the can be described along many dimensions. It is impossible to dual-task situation. To phrase it differently, how many things construct a secondary task that taps only a single component can be done simultaneously? In the selection of tasks, the effort as it is unreasonable to assume that a primary task depends is to minimize the overlap in demands and eliminate the in­ only on one kind of processing facility. An attempt to study the terference. In principle, this approach is complementary to the demand structure of a task employing a secondary-task pro­ secondary-task technique, but it marks both a strategic and a cedure is, therefore, likely to require a series of tests, each theoretical shift in emphasis. Theoretically, a lack of interference including a slightly different primary-task-secondary-task is less amenable to association with a specific dimension and pairing. These variations are directed toward partialing out should be considered an as overall integral property of the sit­ structural issues, timing overlap, rate of presentation, per- uation. Strategically, subjectS are encouraged to put equal em-

~ I WORKLOAD-AN EXAMINATION OF THE CONCEPT 41-27 phasis on the performance of both tasks, integrate them if pos­ their nature; and (2) the dual-task situation does not have dif­ sible, and maximize their overall performance. Experimental ferent or emergent properties not present in the single-task examples for this approach are a study by Allport, Antonis, condition. As has been argued by several authors, the validity and Reynolds (1972) in which subjects were required to time­ of these assumptions is rather uncertain (Duncan, 1979; Neisser, share the playing of one musical piece on a piano and the reading 1976). With training under the dual-task conditions, many tasks of another piece; a study by Hirst, Spelke, Reaves, Charack, are not viewed as independent components of a pair. Subjects and Neisser (1979) in which subjects read text presented on a often integrate the two tasks to reduce demands. The problems screen while writing down dictations presented orally; and a encountered in the performance of this new task cannot be study by Schneider and Fisk (1982) in which subjects performed easily reduced to the demand profiles of the original combining a category classification and a choice reaction task simulta­ tasks (Hirst et al., 1979; Neisser, 1976). neously. In all of these studies, initial interference disappeared Even in the event that task integration did not occur, new after prolonged practice. The experimental approach and the factors that increase or temper interference in concurrent per­ interpretation of data in these studies have a different focus. formance of tasks may appear as the sole contribution of the The aim of the workload analysis should therefore be specified dual-task condition and be absent or irrelevant to the processing and stated clearly in advance. demands of the component tasks themselves. Imagine that the Another concern in generating and interpreting patterns primary task of interest is a tracking task, and the secondary of dual-task interference is related to our distinction between task requires subjects to classify words presented in an adjacent structural and energetical dimensions ofprocessing and response window. The requirement to move the visual fixation point from limitation. When the effects of vocal response are contrasted the tracking display to the word presentation may impose heavy with manual response and found to be different, they are at­ constraints on the ability to time-share the performance of the tributed to a structural factor, namely, the mode of response. two tasks, but it has nothing to do with the load imposed by Similarly, we regard as structural the inability of the eyes to each individually and the variables that can be used to manip­ move fast enough to monitor changes on a multi-instrument ulate the processing difficulty on each. This is a rather clear display, or the problems of coordinating actions with asyn­ example; others may be harder to detect or account for. Duncan chronous rhythms. These factors may all create large differences (1979) brings several experimental examples of the effects of in our ability to time-share the performance of tasks. The struc­ such emergent properties on dual-task performance. In one ex­ tural label extends to the typology of processing stages (e.g., periment, the perception of letters in each of several attended Gopher & Sanders, 1984; Sanders, 1983), or to the sequence at spatial locations was shown to be affected by properties combined which relevant processing structures are employed (e.g., Kerr, across locations and not solely by the local stimulus presented 1983; Treisman, 1969). Energetical concepts have been intro­ at each attended location. In a second experiment, special dif­ duced to describe interference in those cases in which the struc­ ficulties were demonstrated when stimulus-response mappings ture of the tasks remains unchanged but the intensity of in­ were different for two choice reaction tasks performed in close volvement of one or all mechanisms is manipulated. This is succession [psychological refractory period (PRP) paradigm]. A usually done by increasing the difficulty of one task variable, third experiment showed that when the two hands performed say, the speed of movement of a target in a tracking task (Gopher different rhythmic actions (internally programmed sequences & Navon, 1980), or using several levels of difficulty in a number­ of taps) there was some tendency of each hand to carry out the counting task (Reisberg, 1983). A second technique is to change action assigned to the other (see also Kelso, Southard, & Good­ the level of required performance on a task (e.g., the speed of man, 1979). In all of these examples, extra costs and considerable response, or the tolerance level for errors, Gopher, Brickner, & interference were added due to factors that did not exist and Navon, 1982). were not relevant to the performance of the single tasks them~ Operationally, the two manipulations are translated into selves. qualitative and quantitative changes of the properties of con­ The theoretical consideration is whether we want to elim­ currently performed tasks. Because the results of such changes inate or separate the factors that emerge from the combination are regarded as complementary rather than conflicting, many ofthe tasks. Emergent processes reflect the reality ofthe system's studies include a manipulation of both. One example of this attempt to coordinate and overcome the problems of two de­ approach is a study of McLeod (1977), in which the effects on manding missions. As such, they should be of as much interest tracking of two structurally different choice reaction tasks were to the researcher of the system limits as are the demands imposed studied jointly with two levels of difficulty of a mental arithmetic by the performance of each of the composing tasks. This ar­ task. Another example is a study by Gopher, Brickner, and gument is even more compelling when one considers the vague Navon (1982), in which two versions of a typing task were status of our notion of a task. Every task is composed of many combined with a tracking task under three different levels of elements, and subjects can frequently trade them in their per­ desired performance. formance. In a two-dimensional tracking task with a single 3.6.3. The Problems of Concurrency. Within the second­ display and a single hand controller, subjects can easily trade ary-task methodology, the secondary task is of no interest by tracking along vertical and horizontal axes (Gopher & Navon, itself; it serves only as a tool to make possible the study of the 1980; Navon, Gopher, Spitz, & Chillag, 1984). Is it then a single­ demand composition and attention requirement of another task or a dual-task situation? There is no good answer to this question, of interest. Its employment includes the 'additional implicit as­ because there does not exist a clear definition of a task. Note, sumption that the basic structure of these requirements remains however, that an acceptance of the spirit of this argument unchanged. This implies that the demands on the operator's changes the general locus of emphasis of the methodological resources imposed in the dual-task conditions are a linear sum­ approach. While the main focus in the original approach was mation of the demands imposed by each task when performed on the difference between single-task and dual-task performance by itself. This "task invariance" assumption comprises two levels, our intereat now concentrates upon the dual-task situation 1 components: (1) When combined the two tasks do not change and we compare single to dual tasks only as a secondary source 41-28 HUMAN PERFORMANCE

of information. This is the approach proposed by Kantowitz and Navon & Gopher, 1979, 1980). Several experimental works have Knight (1976), and it is strongly advocated by Navon and Gopher shown that attention control is a skill that can be acquired and (1979) and Gopher and Sanders (1984). The dual-task situation improved. It was also shown that, without training, subjects is the focal point for these researchers, and manipulation of tend to converge on suboptimal solutions (Gopher, 1981; Gopher task variables under dual-task conditions is the main vehicle & Brickner, 1979; Gopher & North, 1977). This issue deserves for uncovering the demand structure of tasks. Experimental explicit treatment. examples of this approach can be found in Kantowitz and Knight The main problems of the secondary-task methodology can (1979) and in Gopher, .Brickner, and Navon (1982). be summarized as follows: (1) Single- to dual-task performance 3.6.4. Allocation Policy. If the introduction of a secondary comparisons are risky because of emerging properties and task fulfills its purpose, the capacity of some processor is sat­ structural changes; (2) the primary-task protection requirement urated, and a shortage in processing facilities is created. The is highly susceptible to subjective interpretation. It is often subject is confronted with the need to develop an allocation violated and experimenters are required to deal with perform­ policy. That is, the subject must decide how to divide the scarce ance changes on both tasks, with little capability to disambiguate resources. The subject's abilities to control this allocation policy, the reasons for these changes. (3) The paradigm does not make and the degree to which options are open to the subject, are possible the assessment of problems related to attention control aspects of the secondary-task paradigm that should not be ig­ and strategic planning. The next section examines the ways in nored. The common approach has been to designate one task which these problems are approached in the performance op­ as primary and ask subjects to protect its performance, thereby erating characteristics (POC) methodology. creating a priority difference between the concurrently per­ formed tasks. In previous paragraphs, we described some of the 3.7. Performance Operating Characteristics (POC) difficulties that may prevent subjects from protecting primary­ task performance. These difficulties may result in an unavoidable A POC is a curve depicting all possible dual-task combinations impairment of performance on both the primary and the sec­ arising from splitting a common and limited pool of resources ondary tasks, leaving the experimenter with the problem of among concurrently performed tasks. Given the structure of weighting decrements on one task and improvement on the tasks and the capabilities of the system, some levels of joint other. However, there are further complications in the imple­ performance are feasible while others are not. A POC depicts mentation of this paradigm. the set of all dual-task combinations produced when the system One problem is that the instructions are rather vague. The operates at its full capacity. The term full capacity is used to precise meaning of primary and secondary may be unclear to refer only to those facilities competed for by a pair of concurrently the subject. How should the efforts be allocated between tasks? performed tasks. The overlap in processing demands between How should the instructions to protect primary-task performance tasks may be partial, and some resources may be r~levant to be interpreted? The subject may not have clear answers to these the performance ofone task only. That is, a POC is a performance questions. The introduction of a secondary task complicates the trade-off function that describes the improvement ofperformance situation. Since a second task is introduced, it is clear that some on one task due to added resources released from lowering the performance on it is expected. This expectation, however, con­ standard of performance on another task with which it is time­ flicts with the instruction to fully protect primary-task per­ shared. The idea of constructing POCs, as a general approach formance. Subjects must decide how much to protect, when the to the analysis of the behavior of the human processing system, protection requirement is satisfied, and how much effort (or was first introduced by Norman and Bobrow (1975), with ref­ ''resources") can be devoted to the secondary task. The subjects erence to a metaphor of a computer system. It was elaborated may be motivated to maximize their secondary-task performance by Navon and Gopher (1979, 1980), who adopted concepts from .. ~ because this presents a real challenge. They may be driven to microeconomic theory, thus drawing an analogy between a per­ adopt a lenient interpretation of the instruction, or sacrifice son performing two tasks and a manufacturer trying to optimize aspects of performance that are not directly monitored by the investments in the production of two products. Sperling and experimenter (e.g., reducing their accuracy when only speed is Dosher (Chapter 2) use the term attention operating character­ monitored, or limiting their efforts to store information in istics (AOC) to label these curves and discuss them in the general memory for future reference; see also Sperling & Dosher, Chapter framework of the theory of signal detectability. Irrespective of 2). It may also be the case that subjects do not have self-knowl­ the background metaphor, the technique employed by all re­ edge, or good external feedback on their performance and thus searchers is the same: Subjects are instructed to perform with do not know what to protect. graded changes in the relative priorities of tasks under dual­ Another implicit assumption of the instructions is that task conditions (different emphasis levels are indicated verbally, '1 subjects have sufficient control on their processing efforts that and often augmented by feedback information and differential they are able to determine for each pair of tasks how much to rewards). Our discussion is based mainly on the terminology give and. readjust their policy if it does not lead to the expected and arguments developed by Navon and Gopher (1979, 1980). :t outcomes. But what if some levels of allocation are mandatory Figure 41.11 depicts several hypothetical POCs describing and imposed by the tasks themselves? How much control do the possible trade-off's in the performance of task X and task I subjects have on the allocation of their processing facilities? Y. Each POC traces the bounds of joint performance and thus l Can they detect deviations from optimal allocation? It is clear can be considered to represent the production frontier line, to that these questions are an integral part of the measurement follow economics terminology. Each combination on this curve paradigm, as much so as selecting secondary tasks and deciding or inside the area bounded by it is feasible [e.g., C1, C2, Ca, and l about the types of difficulty manipulations. Moreover, they need c5 in (a)]; all points outside the area are impossible [e.g., c4 in to be examined in a broader, theoretical perspective, because (a)]. Values in brackets indicate the relative priorities assigned they influence the degrees of freedom that the processing system to each task at that point. Note that the intersections of the has in coping with situational demands (Moray, Chapter 40; functions with the X and Y axes are given the values 0.0 and WORKLOAD-AN EXAMINATION OF THE CONCEPT 41-29

features of the POC technique before turning to consider ex­ perimental examples. Single y (0.0,~ Recall that a POC is obtained by changing the relative emphasis on concurrently performed tasks holding all other variables at a constant level; it is basically a test of the influence of a change in processing efforts under fixed-difficulty conditions. >­ If, in addition, the variables or the structure of tasks is manip­ w ulated, a new POC has to be constructed for every new condition, (.) z <{ leading to a family of POCs like the one depicted in Figure ::;;: a: 41.12. 0 u.. Quadrant I in this figure presents a family of three POCs. a: w They were obtained by manipulating priorities in the joint per­ c.. formance of task X with task Y, holding constant the difficulty of task X and changing the difficulty of Y in three levels (easy, medium, and difficult). It can be seen that as the difficulty of task Y increases, more units of performance on task X have to (1.0, 0.0) be sacrificed to improve each unit of performance on task Y. Single x PERFORMANCE X Consequently, the curve intersects the Y-axis at a lower point and has a shallower slope. Quadrants II and IV in Figure 41.12 Figure 41.11. Three hypothetical POC curves depicting performance trade­ demonstrate how the joint POCs are related to the individual offs between concurrently performed tasks. Each curve traces the bounds of performance resource functions of each task. The third quadrant joint performance under different levels in intertask priorities. Combinations depicts the allocation policy line and the influence of two priority c,, C2, C3, and Cs in (a) are feasible: C4 is not. Numbers in brackets indicate the priority level of each task at that point. The three curves describe different levels (A, B) on concurrent performance levels. types of overlap between tasks in demands for processing resources: (a) total The POC methodology has broadened considerably the scope overlap; (b) partial overlap; and (c) complete independence. of usage of interference patterns among concurrently performed tasks as a method for studying workload. Its major assertion is that, in order to analyze the demands of a pair of tasks, one 1.0, respectively. They are assumed to correspond to single-task cannot use just one condition in which demands are imposed, performance levels on the two tasks (i.e., when one task is fully as this would be considering a single point, and at an unknown attended to, 1.0, and the other is at 0.0 emphasis). location, from a complete curve. An analogous problem would· The three POCs in Figure 41.11 indicate that a POC may be the selection of a single point on an ROC curve in signal assume different shapes that reflect differences in the nature detection theory. The limits ofsuch an approach are quite obvious of the trade-offs between performance on the two tasks. Curve from examining the hypothetical POCs in Figures 41.11 and 41.11(a) represents the case of a complete and even trade-off. 41.12 (for a further discussion of this point see Navon & Gopher, Every unit of performance on one task can be traded with per­ 1979, 1980; Sperling & Dosher, Chapter 2). The adoption of the formance on the other tasks, and the "costs" of each performance POC method entails a systematic consideration of allocation unit are equal across the whole range of performance on the policy effects and motivational factors neglected by the tradi­ two tasks. Curve 41.11(c) depicts the opposite case, a complete tional secondary-task paradigm. The value of collecting these absence of trade-off. Each task can progress and regress across data is not only to control the influence of such variables, but its total performance range with no effect on the performance also to make possible a better assessment of the contribution of the other task. Both (a) and (c) are demonstrations of a clear of processing effort (energy investments) to the performance of relationship easily explained. The more complex and difficult tasks aad the degree to which tasks interfere with each other, (and more frequent) to interpret is curve 41.11(b), which shows holding all other variables constant. Given the inherent am­ the convex curve reflecting changing rates and uneven processing biguity of the specification of task structure, the power and costs at different regions of the performance range. There are importance of such an additional comparison are much increased. several possible causes for this type of trade-off function, and This is especially so if a family of POCs is constructed by ma­ they are not mutually exclusive. One possibility is that one or nipulating variables of task difficulty. both tasks have reached a performance ceiling or a data limi­ The main drawback of the approach is that it is considerably tation such that resources released from one task cannot be more complex in the design of experiments. It is also more time­ used to improve performance on the other. Another reason is consuming in data collection and analysis. However, when one that the marginal costs of performance on each task may be considers the richness and importance of the information, and different at different levels of performance and the POC will the present state of our knowledge of workload, the additional be sensitive to this change of costs. A third possibility is that effort seems worthwhile. We next review one example from a tasks overlap only partially in their demand for common re­ paper by Gopher, Brickner, and Navon (1982) to demonstrate sources. The influence of this overlap on joint performance is the application of this approach. The reader can find another restricted to one region of performance or change in different example in Sperling and Melchner (1978; see also Sperling & regions. The convex curve and its degree of curvature in this Dosher, Chapter 2). · case represent an intermediate class between a complete trade­ To test the notion of multiple resources, a two-dimensional off and the total independence depicted in Figure 41.11(a) and pursuit-tracking task was paired with a letter-typing task. The (c). An interpretation of a POC may be beneficial and lead to relative priorities of the two tasks under dual-task conditions a considerable broadening of the scope of questions examined and the difficulty of the typing task were manipulated. Each relative to the limited range provided by the traditional sec­ task was also performed singly. A schematic diagram of the ondary-task approach. It seems instructive to review a few more experimental display is presented in Figure 41.13. 41-30 HUMAN PERFORMANCE

Perf (y)

II E Perf-res functions (y J : POCs { .. I Rlim Resources (y J Perf (xJ I Allocation policy l Perf-res function (x) m Rlim

Resources (x)

Figure 41.12. A hypothetical example of a family of performance operating characteristics describing dual-task performance of tasks X and Y with manipulation task priorities (allocation policy) and task difficulty. Difficulty of Y is varied at three levels: easy (f); medium (M); and difficult (0): Task X is unchanged. Quadrant I depicts the three performance operating characteristics (POCs) resulting from the combined performance of task X with the three variants of task Y. Quadrants II, Ill, and IV illustrate how the POCs of joint performance are related to the performance-resource function of each task (quadrant II for task Y and quadrant IV for task X). Points A and B in quadrant Ill show how two different priority levels would affect joint performance. (From D. Gopher & A. F. Sanders, S-Oh-R Oh stages, Oh resources. In W. Printz & A. F. Sanders (Eds.), Cognition and motor processes, Springer-Verlag, 1984. Reprinted with permission.)

In the tracking task, subjects controlled the movement of lated into commensurate increases and decreases of performance a control symbol on the screen via a two-dimensional hand demands along the POC curve assuming that subjects operate controller with a relatively difficult control dynamic. Target with full capacity. Desired performance levels were computed movement was driven by a random forcing function governed relative to a standardized level representing top effort obtained by the computer. The typing task required the subject to type in single-task conditions. The moving bar graphs were contin­ the appropriate chord combinations of Hebrew letters presen~ uously updated, based upon a running average, and reflected within the moving target of the tracking task. This task was the momentary difference between actual and desired perform­ based on a single-hand chord typewriter developed by Gopher ance on each task. Three levels of priority (0.3, 0.5, 0.7) were and Eilam (Gopher, 1984). The system comprises three keys, employed under dual-task conditions, in addition to single tasks and each letter is entered by typing two successive chords of _ considered as 1.0 level on the priority scale. one to three keys pressed together. Letter codes provide spatial The main findings of this experiment are presented in Fig­ mnemonics corresponding to the shape of letters in print. Dif­ ures 41.14 and 41.15. Figure 41.14 depicts the effect of the ficulty on this task was manipulated in two ways. Under the difficulty and priority manipulations on typing performance. cognitive manipulation the number ofletters in the set presented Figure 41.15 describes the family of POCs resulting from the to the subject was increased from four to 16. In the motor ma­ joint performance of tracking and typing under all experimental nipulation a group of four letters coded by motorically difficult conditions. chord combinations was selected. The difficulty of the tracking From examining Figure 41.14 it is clear that both typing taskTemained constant under all conditions. Priorities of tasks difficulty and the change of priority levels had marked effects were varied through a feedback display composed from a static on typing performance. Response times to enter letters decreased vertical line and two moving bar graphs, each representing one monotonically with. an increase in the relative priority of the task (Figure 41.13). The vertical line represented desired per­ typing task. Note that single-task levels lie roughly on the formance. When it was moved away from the side of one task, same line. There was also a step increment in response time performance demands on it were increased while simultaneously as a result of increasing either motor or cognitive difficulty. decreasing on the other task. Priority changes were thus trans- However, only motor difficulty interacted with priority changes WORKLOAD-AN EXAMINATION OF THE CONCEPT 41-31 and had a steeper slope with response time as task priority of efforts with those that result from a manipulation of the varied, relative to the effect of priority change on the easy task characteristics of tasks. version. The same pattern of results is present in the POC family plotted in Figure 41.15. This figure also shows how per­ 3.8. Basic Assumptions of the POC Methodology formance changes on the typing task were accompanied by commensurate changes in tracking accuracy. These results were The POC approach does make a number of implicit assumptions, r interpreted to indicate that tracking and typing compete with as do other techniques. The first and major one is the sensitivity each other for motor resources but not for cognitive resources. of performance to processing efforts. Performance on tasks is f " The existence of interaction between motor difficulty and assumed to be monotonically related to the amount of invested t priority change and the absence of such interaction when cog­ resources. It is further assumed that this amount is the main t nitive load was manipulated were crucial for the above inter­ source of variability in time-sharing performance. Problems pretation. If one member of a concurrently performed pair of that arise from data limitations or performance ceiling (e.g., tasks is made more difficult on a dimension that taps a resource Norman & Bobrow, 1975) are acknowledged but considered to for which the two compete, the slope of the POC is predicted to be only a secondary source of variation. A related assumption change, because a larger sacrifice ofperformance will be required is that the human is able to control resources and apportion c;m the other task to make possible an improvement on the now them at least across the main portion of the response-sensitive more difficult task. If, however, the increase of difficulty is made range. A third important assumption is that the POC reflects on a dimension that is not relevant to joint performance, per­ the boundary of system performance at full capacity, and that formance on the task on which the manipulation was conducted this capacity is fixed. That is, there is a fixed upper limit on may be affected, but the slope of the POC should not change the rate of recruitment of processing facilities. If subjects do because the load on the shared resource has not been changed. not operate at maximum capacity, or if capacity can expand A differential diagnostic of the type demonstrated in this ex­ and shrink (e.g., Kahneman, 1973), the interpretation of a POC periment cannot be achieved with the conventional secondary­ is impossible. task approach. They are demonstrative of the power of the POC Two additional assumptions are independence of tasks and methodology. A second benefit of this approach is the ability to process invariance. A manipulation of relative priorities under compare the relative sensitivity of performance to reallocation dual-task conditions is meaningless unless the component tasks I ( Desired performance Line Q f Letter typing ~. Tracking bargraph I ==:J I t bar graph f f. ~

! Q ;etter Target~

X Control symbol

i DOD I Letter typing keyboard •t ~ i Tracking hand controller Figure 41.13. Subject's display in concurrent performance of tracking and letter typing with manipulation Ir of priorities. Tracking required to follow the target driven by the computer, with the X controlled by the right-hand controller. Typing was performed by using the left-hand keyboard to enter the motor chord I codes of Hebrew letters, presented within the tracking target square. Continuous feedback on performance was given through the two horizontally moving bar graphs in the upper section of the display. A short vertical line indicated the desired level of performance and was moved to the right and left to inform subjects on changes in the emphasis on tasks. The size of the performance bar graphs reflected the instantaneous difference between actual and desired levels of performance. (From D. Gopher, M. Brickner, & D. Navon, Different difficulty manipulations interact differently with task emphasis: Evidence for multiple resources, Journal of Experimental Psychology: Human Perception and Performance, 1982, 8. Copyright 1982 by American Psychological Association. Reprinted with permission.)

~ !. ~ t 41-32 HUMAN PERFORMANCE

LETTERS ONLY 3.9. Psychophysiological Measures of Workload 'iii' 1400 "0 <: 3.9.1. Introductory Comments on the P300 Component. The 8 1350 .. ERP is a transient series of voltage oscillations in the brain ,;g 1300 that can be recorded from the scalp in response to the occurrence 11250 of a discrete event (Donchin, 1975). The ERP is viewed as a w sequence of components commonly labeled with an N or a P ~ 1200 1- denoting polarity, and a number that indicates their minimal z 115 0 ~ ,. = Difficult motor 0 latency measured from the onset of the eliciting event (e.g., i= • =Difficult cognitive u 1100 "=Easy N100 is a negative-going component that occurs at least 100 < msec after a stimulus). Since ERPs are small, relative to the ~ 1050 ongoing EEG, their study became practical only after the de­ ~ 1000 0:: velopment of reliable signal averagers. These capitalize on the >- 1- 950 fact that the ERP is, by definition, time-locked to ·the eliciting event. 900 .3 .5 .7 It is crucial to recognize the componential nature of the ERP. The effects of the experimental manipulations tend to be PRIORITIES quite specific to a few components, and a combination of the Figure 41.14. An empirical example of obtained performance resource measures of the entire epoch may obscure the relevant variance. functions. Average response times for typing letters are plotted as a function There is a degree of controversy as to the proper identification of the relative priority of this task in concurrent performance with 'a tracking and definition of components (Donchin, Ritter, & McCallum, task. Priorities were changed by instructing subjects to change their standards 1978; Picton & Stuss, 1980). In this chapter, however, we shall of performance relative to a baseline obtained for each subject. Each curve follow Donchin et al.'s (1978) definition of an ERP component depicts the results for one of the three variants of the typing task employed in terms of the responsiveness of the waveforms to specific ex­ in this experiment. (From D. Gopher, M. Brickner, & D. Navon, Different perimental manipulations. A component is thus mapped into difficulty manipulations interact differently with task emphasis: Evidence for a cognitive space populated by psychological concepts such as multiple resources, Journal of Experimental Psychology: Human Perception and Performance, 1982, 8. Copyright 1982 by America'n Psychological As­ decisions, expectations, plans, strategies, associations, and sociation. Reprinted with permission.) memories. The subset of elements in cognitive space associated maintain their independence, in the sense that resources al­ located to the performance of one are withdrawn from the other. TYPING In the same vein, it is assumed that when the levels of emphasis • =Easy are changed tasks do not change their basic processing structure. • = Difficult motor Only the level at which these structures are engaged is assumed 900 & =Difficult cognitive to vary. If the demand composition of tasks changes with the • • = Single task level of allocated efforts, what hopes may one have to find general ' ' !i 1000 rules to relate the properties of tasks to processing demands? i= The POC methodology, like the secondary-task approach, is w ~ vulnerable to the effects of emergent properties and task in­ lQ 1100 O'iii' ... Q.-o tegration. However, it is better equipped to detect and isolate VI C ' the effect of such variables, in the context of a family ofPOCs. w 0 0:: ~ 1200 This is another justification yet for the construction of a family (.!)z:: .. ~-'o,\ of trade-off functions. As we define workload in terms of the O::'E interactions between operators and tasks, and as the POC curves ~ '-' 1300 0:: portray these interactions, their value within this framework w 1- is evident. t;j 1400 Note that, while the POC curves do provide a very useful ....J analytic tool ofthe limits on human capacity, the straightforward structure of the tests employing secondary-task methodology is abandoned. This is due largely to the fact that the secondary 0 .65 .70 .75 task may interfere with the primary task. There is thus a need for secondary tasks that do not interfere, or that interfere only TRACKING PERFORI.lANCE slightly, with the primary task. A category of such tasks has (1 - RMS error) been proposed and developed by Donchin and his colleagues Figure 41.15. An experimental example of a family of performance operating (Donchin, 1975; Donchin, Kramer, & Wickens, 1983; Donchin, characteristics. The curves depict trade-offs under time-sharing conditions 1984) by using event-related brain potentials (ERPs) as the with manipulation of priorities, between a constant difficulty tracking task source of the data from which variations in primary-task work­ and three variants of a letter-typing task. Dashed lines were used to connect performance in dual-task conditions with single-task levels on the typing load can be inferred. The ERP, in particular the ERP component task. (From D. Gopher, M. Brickner, & D. Navon, Different difficulty ma­ called the P300, is recorded off the scalp of an awake subject, nipulations interact differently with task emphasis: Evidence for multiple and its generation requires no overt action by the s11bject. The resources, journal of Experimental Psychology: Human Perception and Per­ next section introduces the ERP and reviews the way it is used formance, 1982, 8. Copyright 1982 by American Psychological Association. in studies of workload. Reprinted with permission.) WORKLOAD-AN EXAMINATION OF THE CONCEPT 41-33 with a particular component thus contributes to the definition a subroutine is an appropriate metaphor for the activity ofERP of the ERP component. components (Donchin, Kubovy, Kutas, Johnson, & Heming, The specific attributes of a waveform that are examined 1973; Donchin, 1975). Thus, ERP components may be associated in defining a "component" are the amplitude, latency, and scalp with specific information ,processing functions that are activated distribution. It is the sensitivity of these attributes to experi­ in a variety of different tasks. In the case of the P300, the mental manipulations that defines an ERP component. Although "subroutine" may be invoked whenever there is a need to eval­ no reference has been made to the underlying neural source of uate surprising, task-relevant events. This interpretation of components, it is gen~rally assumed that a scalp distribution the changes in P300 amplitude is strengthened by the evidence that is invariant across repeated stimulus presentations implies that has accumulated in the past decade regarding the factors a specific and fixed set of neural generators (Goff, Allison, & that control the latency of the P300. As the use we make of Vaughan, 1978). Thus the scalp distribution that is related to P300 in the analysis of workload depends strongly on the the­ the underlying neural population responsible for the generation oretical interpretation of the component, it will be useful to of the component is assumed to be a crucial defining charac­ provide a brief review and interpretation of the latency data. teristic. The ERP components discussed in this chapter are endog­ 3.9.2. The Latency of the P300 Component. The peak la­ enous and are distinct from another class of ERPs called ex­ tency of the P300 component appears to depend on the time re­ ogenous. The exogenous components represent an obligatory quired to recognize and evaluate a task-relevant event. The latency response of the brain to the presentation of a stimulus. These is in the range of 300-750 msec following the presentation of a components are primarily sensitive to such physical attributes discrete stimulus. For example, in fairly simple tasks calling for of the stimuli as intensity, modality, and rate. The seven peaks a discrimination between two tones that differ in pitch, the stimuli or ''bumps" that occur in the first 8-10 msec after the presen­ elicit relatively short-latency P300s. More difficult discriminations tation of an auditory or somatosensory stimulus are a proto­ result in increases in the latency of P300. typical example of the exogell.ous category (Jewett, Romano, & Assuming that manual or vocal reaction time terminates Williston, 1970). processing, and that P300 is a manifestation of a process that Endogenous components, typically, are not sensitive to precedes the response, then it would be expected that P300 changes in the physical characteristics of the eliciting stimuli. latency and reaction time should positively covary. This pre­ On the other hand, these components are very sensitive to diction has been supported by numerous studies (Wilkinson & changes in the processing demands of the task imposed on the Morlock, 1967; Bostock & Jarvis, 1970; Rohrbaugh, Donchin, subject. The endogenous components are nonobligatory responses & Eriksen, 1974). Other investigations, however, failed to detect to stimuli. The strategies and expectancies of the subject as a relationship between P300 latency and reaction time (Karlin, well as other psychological aspects of the task account for the Martz, & Mordkoff, 1970; Karlin & Martz, 1973). variance in the endogenous components. A typical example, Donchin et al. (1978) proposed an interpretation of the pro­ and one to which we shall devote the remainder of this chapter, cesses underlying the P300 that may reconcile these contra­ is the P300 component. dictory findings. They suggested that P300 latency is determined This ERP component is elicited by rare, task-relevant by the time required to evaluate the stimulus but is largely stimuli. A task in which it is readily elicited is often called the independent of response selection and execution time. The cor­ "oddball" paradigm. In a study by Duncan-Johnson and Donchin relation between reaction time and P300 latency would, ac­ (1977) using this paradigm, the subject was instructed to count cordingly, vary as a function of the percentage of reaction time covertly the total number ofhigher-pitched tones in a Bernoulli variance that is accounted for by stimulus evaluation processes. series. In different blocks of trials the relative probability of This percentage would be affected by the strategies employed the two tones was manipulated. It can be seen from Figure by the subject. The strategies, therefore, should influence the 41.16 that the amplitude of the P300 increases monotonically relationship between P300 latency and reaction time (see also as the probability of the stimulus decreases. This occurs re­ Ritter et al., 1972). Evidence that P300 is determined by the gardless of which of the two stimuli is being counted. When the amount of time required to recognize and evaluate a stimulus subjects were solving a word puzzle and were not required to has been reported by several investigators who employed process the tones, the P300s were not elicited. Sternberg's (1966) additive factors methodology (Ford, Roth, Note that the ERPs in Figure 41.16 that were obtained in Mohs, Hopking, & Kopell, 1979; Ford, Mohs, Pfefferbaum, & this "ignore" condition show no P300 at all levels of probability. Kopell, 1980; Gomer, Spicuzza, & O'Donnell, 1976). Thus, the amplitude of P300 is determined by a combination Other investigators employing different paradigms also of the task relevance and the subjective probability of the elic­ report that P300 latency and reaction time are positively cor­ iting event. This basic finding plays a crucial role in the use of related when stimulus evaluation time is manipulated. Squires, P300 in the assessment of workload. Donchin, Squires, and Grossberg (1977) found that P300 latency The demonstration that P300 is elicited by unexpected, and reaction time covaried with the difficulty of auditory and task-relevant stimuli led Donchin, McCarthy, Kutas, and Ritter visual discriminations. Furthermore, P300 latency varied with (1983) to suggest that "the P300 is a manifestation, at the scalp, the manipulation of stimulus discriminability while reaction of neural action that is invoked whenever the need arises to time was influenced by both stimulus evaluation and response update the 'neuronal model' (Sokolov, 1969) that seems to un­ selection factors. Heftley, Wickens, and Donchin (1978) per• derlie the ability of the nervous system to control behavior" (p. formed an experiment in which subjects were required to monitor 105). The neural or mental model is continually assessed for a dynamic visual display for intensifications of one or two classes deviations from inputs and revised when the discrepancies exceed of targets. P300 latency was found to increase monotonically some criterion value. The frequency with which the mental with the number of elements on the display. Since subjects were model is revised is based on the surprise value and task relevance not required to make an overt response, the differences in P300 of the stimuli. Donchin (1981) also argued that the concept of latency were attributed to stimulus evaluation processes. -- Count high · · · · · • Ignore

High tone Low ·tone

10% ~ 90% ~-~

20% -~~80% ...... ·························' ~-········ .. ·····...... ~~ t

~70% 30% ~...... ··· ...... - ~ .. ~ 60% 40%~,

~······················-····· 50% ~ 50%~)

. . 40% ~ 60%.~················· .. ... •

30% ~ 70% ~ .. ··· . . . ~ 20% 80%~ 10%

90%~

ro 20.Uv I - I I 1 4001 I 600I I 200 400 600 200 - [ MILLISECONDS· + MILLISECONDS

Figure 41.16. Event-related brain potentials (ERPs) recorded from a group of subjects. Each trace represents the ERP obtained by averaging over trials and over subjects. These data were recorded at an electrode at the parietal site (Pz) referred to a linked ear electrode. Positivity at the scalp electrode is reflected by a . downward deflection. Each of the superimposed pairs was recorded in experimental series in which the subject was presented with a Bernoulli sequence of tones. The solid lines in the left column were recorded when the subject was instructed to count and report, after the series was presented, the number of times the high tone was presented. This is the so-called oddball task. There were nine such series and they differed in the probability that the high tone would be presented. The probability was varied from .10 to .90 in increments of .1 0, as indicated by each trace (the percentage figure by each trace indicates the percentage of high tones in the series). Note that when the high tones are rare, the ERP is characterized by a large positive-going wave. This is the P300. The ERPs represented by the dotted lines superimposed on the solid lines were elicited by precisely the same tones that elicited the solid lines, except that in this case the subject was not counting tones but rather was trying to solve a word puzzle. Note the absence of the P300 from these ERPs. These data indicate that the P300 is not an obligatory response to the tone, that its elicitation depends on the fact that the tone is task relevant and on the rareness of the tone. The data in the right column were elicited by the low tones that were presented in the series used in the corresponding pair in the left column. Note that when low tones were rare they nevertheless elicited a larger P300 even though they were not counted. (From C. C. Duncan-johnson & E. Donchin, On quantifying surprise: The variation in event-related potentials with subjective probability, Psychophysiology, 1977, 19. Copyright 1977 by the Society for Psychophysiological Research. Reprinted with permission.)

41-34 WORKLOAD-AN EXAMINATION OF THE CONCEPT 41-35 lfP300 latency is determined by stimulus evaluation time different stages of processing. P300 latency was increased by and is largely independent of the time required for response the addition of the noise to the target matrix but was not affected selection and execution, then experimental variables that have by the incompatibility between the stimulus and the response. a different effect on processing time in the two stages should These results support the conclusion that P300 latency is affected influence the relationship between P300 latency and reaction' by a subset of the set of processes that affect reaction time. The time. For example, when subjects are instructed to respond P300 is elicited only after the stimulus has been evaluated. quickly with a low regard for accuracy, their responses are Subsequent processing required for the selection and execution probably emitted without full evaluation of the stimulus of the response does not appear to influence the latency of the (Wickelgren, 1977). On the other hand, if subjects are instructed P300. It should be noted that the phrase stimulus evaluation is to respond accurately, they are likely to perform a more thorough used here to denote all of the processes that precede response analysis of the stimuli prior to responding. This analysis leads selection and execution. It is not implied, in the theoretical to the prediction that the correlation between P300 latency and position outlined above, that P300 latency is related solely to reaction time would vary with the subjects' strategies. Specif­ the detection and encoding of physical stimuli. It is more than ically, the correlation would be high and positive when the likely that the processing that leads to a P300 extends beyond subjects are instructed to be accurate. Low correlations would strict "stimulus" evaluation and encompasses all aspects of the be observed under speed instructions. situation that affect, in some way, the system's need to update Kutas, McCarthy, and Donchin (1977) tested this hypothesis working memory. by requiring subjects to distinguish between two stimuli under The P300 component of the ERP provides a metric for the both speed and accuracy instructions. In one experimental con­ decomposition of stages of information processing that comple­ dition, subjects were required to discriminate between two ments the traditional behavioral measures. In terms of appli­ names, Nancy and David, presented on a CRT (with relative cations to system design and workload evaluation, ERPs used frequencies of20 and 80%, respectively). In a second condition, in conjunction with behavioral and subjective measures permit female names comprised 20% of the items and male names the assessment of stage-specific task interference effects. For 80%. In the third condition, subjects were required to discrim­ example, if two time-shared tasks interfere with each other, it inate between synonyms of the word prod that occurred with is usually desirable to know the locus of this interaction. Only a relative probability of 20% and unrelated words that were by discovering the stage at which tasks interact can systems presented with the complementary probability. The average be designed that minimize operator workload. P300 latency was shortest for the first condition, intermediate for the second, and longest for the third condition. The more 3.9.3. P300 and Perceptual/Central Processing Resources. complex the discrimination, the longer the P300 latency. A The ~tudies reviewed above provided evidence that the P300 detailed analysis of the single trials revealed that the correlation component is a manifestation, at the scalp, of a processing entity between P300 latency and reaction time was larger for the that is invoked, inter alia, whenever task-relevant, surprising accuracy condition (.617) than the speed condition (.257). Kutas stimuli are present. The routine appears to be performing a et al. (1977) concluded that the data supported the hypothesis role in the context-updating activities that occur whenever an that P300 latency reflected the termination of a stimulus eval­ event calls for the revision of the neuronal model or schema of uation process while reaction time indexed the entire sequence the environment. This model makes certain predictions re­ of processing from encoding to response selection and execution. garding the relationship between the recall of stimuli and the Thus, under the accuracy condition when response selection is amplitude of the P300 they elicit. Such predictions were con­ contingent on stimulus evaluation processes, P300 latency and firmed by Karis, Fabiani, and Donchin (198'4). Also consistent reaction time are tightly coupled. However, when subjects per­ with this model is the demonstration by Klein, Coles, and Don­ form the discrimination under the speed instructions, the pro­ chin (1984) that people with perfect pitch do not invoke a P300 cesses of stimulus evaluation and response selection are more when they make auditory comparisons. loosely coupled, and, hence, the relationship between P300 la­ It is noteworthy that the subroutine manifested by P300 tency and reaction time is not as high. is invoked only if the stimuli are associated with a task that Additional evidence bearing on the issue of the P300's sen­ requires that they be processed. Ignored stimuli do not elicit a sitivity to the manipulation of stimulus evaluation processes P300. But what if the stimuli are only partially ignored? What has been obtained in a study by McCarthy and Donchin (1981), if the subject is instructed to perform the oddball task concur­ who manipulated orthogonally two independent variables in rently with another task? Would the amplitude of the P300 an additive factors design (Sternberg, 1969). The subject's task reflect the centrality of the oddball task? Would it, perhaps, was to decide which of two target stimuli, the words right and change with the amount of resources allocated to the oddball left, was presented in a matrix of characters on a CRT. The task? Clearly, if so, the P300 may serve as a very useful measure characters were presented either within a four-by-four matrix of the amount of resources demanded by the two tasks. It is of# signs (no noise condition) or within a four-by-four matrix this series of questions that lies at the core of the usage that of letters chosen randomly from the alphabet (noise condition). can be made of P300 in the assessment of workload. Stimulus-response incompatibility was manipulated by pre­ It was the basi'c assumption of this research program that ceding the target matrix either with the cue same or with the the oddball task can be used as a nonintrusive secondary task cue opposite. The cue same signaled a compatible response. The sin~e the ERP-eliciting tones occur intermittently, are easily cue opposite indicated an incompatible response; the right hand discriminable, and do not require an overt response. Another would respond to the word left and the left hand .to the cue advantage of this procedure is that it could be applied uniformly right. Reaction time increased when the command word was across different operational settings. In other words, the oddball embedded in noise and when the response was incompatible task could be inserted into virtually any operational setting with the stim.ulus. The effect of the two variables on the reaction without requiring modifications in the system associated with time was additive, implying that these manipulations influenced the primary task. Wickens, Isreal, and Donchin (1977) reported 41-36 HUMAN PERFORMANCE

PDP 11/40 Tracking display Disturbance + output -.J Error I-- Control system

Man-machine system output

Figure 41.17. A schematic representation ofthe experimental arrangements used in studies of workload utilizing the event-related brain potentials (ERPs). A primary task is controlled by the PDP 11/40 computer. In the instance shown, the subject controls a cursor's movement on a screen while the computer controls the movement of the target. As the computer monitors the subject's performance and controls the display the task can be made adaptive. That is, the difficulty of the task can be changed as the subject becomes more proficient. The electroencephalogram is recorded and digitized in a conventional manner. Probe stimuli used in the oddball task are also generated by the computer.

one ofthe first studies in the series using a compensatory tracking the tracking task was manipulated. The bandwidth was in­ task as the primary task and the oddball paradigm as the sec­ creased gradually until the cursor's speed reached the highest ondary task. level the subject could tolerate without exceeding a preset error Figure 41.17 illustrates the experimental procedures used criterion. The results are shown in Figure 41.18. in this and several other studies to be discussed. The subjects Again, P300 amplitude is diminished by the introduction sat in front of a CRT and were instructed to cancel computer­ of the tracking task, but increases in the bandwidth of the generated cursor movements by keeping the cursor superimposed forcing function did not produce systematic changes in the am­ on a target in the center of the display. This was accomplished plitude of the P300. These results cannot be explained easily by movement of a joystick mounted on the left-hand side of the within the framework of an undifferentiated-capacity theory if subject's chair. Levels of tracking difficulty were manipulated we assume that P300 amplitude indexes the demands placed by requiring the subject to track in either one or two dimensions on the subject by the primary task. Increasing the bandwidth (horizontal and/or vertical). The compensatory tracking task clearly affects the performance of overt secondary tasks (McDon­ was defined as the primary task. ·In addition to the tracking, ald, 1973; Wierwille, Gutman, Hicks, & Muto, 1977). The fact the subjects were instructed to count one of two tones presented that P300 did not change, even though a dramatic drop in am­ in a Bernoulli series of high- and low-pitched tones. Control plitude was observed with the introduction of the task, required conditions were also included in which the subjects performed explanation. each of the two tasks separately. One interpretation of the results is that the P300 is not The data indicate that the introduction of the tracking task sensitive to the, processing demands of the task, but, instead, drastically diminishes the amplitude of the P300. However, no reflects the motor activity required by tracking. This hypothesis further reduction in P300 amplitude could be observed as was tested by Isreal et al. (1980), who instructed subjects to tracking difficulty increased by requiring tracking in two di­ manipulate a joystick with one hand concurrently with the mensions. Even though tracking difficulty, assessed by root oddball task. The amplitude of the P300 component elicited by mean squared error (RMS) as well as by reaction time to the the tones was not affected by the motor demand. Thus it would tones, definitely increased with the addition of a tracking di­ seem that hand movements did not decrease the amplitude of mension; P300 amplitude did not change. Isreal, Chesney, the P300. Wickens, and Donchin (1980) conducted a similar study requiring Another interpretation ofthe results can be developed within subjects to perform a compensatory tracking task concurrently the framework of the multiple-resource theory we reviewed in with a counting task. In this case, however, the bandwidth of Section 2.6. The notion that P300 is sensitive to a specific aspect the random forcing function rather than the dimensionality of of information processing is consistent with the data, reviewed WORKLOAD: AN EXAMINATION OF THE CONCEPT 41-37

above, regarding the relation between P300 latency and reaction amplitude of the P300 elicited by a secondary task would be time. P300 latency appears to be sensitive to .a subset of the expected to covary with primary-task difficulty. processes that determine reaction time. Furthermore, P300 la­ Isreal, Wickens, Chesney, and Donchin (1980) tested the tency is influenced by manipulations offactors that are assumed latter hypothesis by combining the oddball task as a secondary li to affect relatively early stimulus evaluation processes while task with a visual monitoring task that served as the primary being insensitive to changes in variables that produce their task. The subjects were instructed to monitor a simulated air effect on the later response selection and execution processes. traffic control display either for course changes or for intensi­ I'{ Ifthe manipulation of the dimensionality and bandwidth of the fications of one of two classes of stimuli (triangles or squares). jt tracking task demands resources associated largely with re­ Primary-task difficulty was manipulated by increasing the sponse selection and execution processes, then P300 amplitude number of elements traversing the CRT (Sperando, 1978). The It should not reflect fluctuations in performance. On the other numerosity variable did have a systematic effect on reaction I hand, if the perceptual aspects of a task were manipulated, the time to the tones when subjects were monitoring for course i{,.~· Ascending Descending

Interval ~ 7 ~ 6 ~ 5

i ·~ 4 10.uv I I ~ 3 j + ~ ~ ": 2 ~ ~

0 (No tracking)

Count. only control

0 370 740 0 370 740

MILLISECONDS MILLISECONDS

Figure 41.18. Event-related brain potentials (ERPs) elicited in response to probes in the arrangement shown in Figure 41.17. Subjects were tracking a target in a study in which tra~king difficulty was varied in a graded fashion by varying the bandwidth of the signal that controlled the t~rget's movement. Subjects were counting high tones in a task similar to that described in connection with Figure 41.16. Ncite that when tracking is introduced the amplitude of the P300 is decreased relative to its amplitude wllen the subject's sole task requires counting tones. Further increases in task difficulty, however, do riot further decrease P300 amplitude. (From J. B. Isreal, G. L. Chesney, C. D. Wickens, & E. Donchin, P300 and

~: tracking difficulty: Evidence for multiple resources in dual-task performance, Psychophysiology, 17. Copyright :Ji:l ~I 1980 by the Society for Psychophysiological Research. Reprinted with permission~) 11 41-38 HUMAN PERFORMANCE changes. Reaction time increased monotonically from the control culty in the monitoring task when subjects were detecting course condition to the condition in which subjects were required to changes. In the flash detection condition, P300s decreased with monitor eight elements simultaneously. However, in the flash the introduction of the monitoring task, but increases in the detection condition, reaction time did not increase significantly number of display elements failed to attenuate P300 amplitude as a function of the number of elements displayed. further. This result is also consistent with the reaction time As can be seen from Figure 41.19, the P300 elicited by the data. Since the primary task did not require a response, the · counted tones decreased monotonically with increases in diffi- data of Isreal et al. (1980) have demonstrated that P300 am-

Course-change detection Flash detection

S1 -./\ __ .,...... ,.. .. ~-.... ---' ~

S2 .. - .··. S3 v . ~"< ...... ~ ., ~..lr. ~

S4 . . _, y-...... ~ S5 -.. ,_ "' S6 Y . ~- ~~~fj" . S7 .....J\ .~ ~. "'" ~-·····

sa .... ~ ~ ~. ""' 20~.;v -r •••.••.•.. 8 display elements + ~ I I I I I I 0 200 400 600 800 1000 1200 --- 4 display elements --- Count only MILLISECONDS

Grand average --'\,.:;;...... ~ v .... ~

Figure 41.19. These event-related brain potentials (ERPs) were obtained when the subject was monitoring a display in which numerous targets were moving. On occasion, one of the targets could briefly intensify. On other occasions, the target would change course. This experiment is similar to that described in Figure 41.18, except that the primary task in this case required the processing of a visual target. Task difficulty was varied by varying the number of elements in the monitored display. The secondary task assigned was again the counting of tones in an oddball paradigm. Data are shown for eight subjects, as well as a "grand average" for all eight subjects. Note that when the subjects are monitoring the display for course changes the amplitude of the P300 is. reduced when the monitoring assignment is introduced. The amplitude is further reduced when the number of elements in the display is increased. The effect is specific to the course detection task. When the subject is monitoring the display for flashes, the effect is much diminished. (From ). B. Isreal, C. D. Wickens, G. L. Chesney, & E. Donchin, The event-related brain potential as an index of display-monitoring workload, Human Factors, 1980, 22. Copyright 1980 by Human Factors Society. Reprinted with permission.) WORKLOAD-AN EXAMINATION OF THE CONCEPT 41-39

Target

~ tj!; ~t ~ J

Manipulator

......

I I Capture 0

~: -f§)- - ®- Capture button Orientation control t Tracking control Figure 41.20. The display used by Kramer et al. Subject's task is to align the manipulator with the target, the latter moved in a linear track. The cursor is controlled by the subject using a joystick. When the manipulator nears the target, the target begins to rotate and the subject must align manipulator with target using a joystick. Task difficulty could be manipulated by varying the order of control for the tracking joystick, in this case, the secondary task utilized as probe stimuli brightenings of either the cursor or the target at different phases of the study. (From A. F. Kramer, C. D. Wickens, & E. Donchin, An analysis of the processing requirements of a complex perceptual-motor task, Human Factors, 1983, 25. Copyright 1983 by Human Factors Society. Reprinted with permission.)

plitude is sensitive to the perceptual demands of a primary by the cursor is called the acquisition phase. Acquisition was task. accomplished by manipulating the two-axis joystick mounted \ Kramer, Wickens, and Donchin (1983) performed a com­ on the right side of the subject's chair. Successful acquisition ponential analysis of the demap.ds of controlling higher-order initiated the·alignment phase. The target began to rotate at a systems, well validated in the literature, to impose a greater constant velocity in either a clockwise or counterclockwise di­ load on information processing resources (Baty, 1971; Fuchs, rection. The subjects had to rotate the c~r at the same velocity 1962). By order of control we refer to the number of time in­ as the target while also keeping the two elements superimposed. tegrations of the output of a controller (i.e., joystick) and the The rotation was accomplished by manipulating the single­ output of the system. In a first-order or velocity-driven system, axis joystick mounted on the left side ofthe subject's chair. A a deflection of the joystick corresponds to a change in the velocity deflection of the stick to the right produced a clockwise rotation of the controlled element. A second-order or acceleration-driven of the cursor at an angular velocity proportional to the angle system produces a change in the acceleration of the controlled of deflection; a deflection to the left produced a counterclockwise element proportional to the movement of the control stick. As­ rotation. Deviation from the initial acquisition criterion for suming that P300 amplitude is sensitive to the perceptual aspects more than 1000 msec necessitated a realignment of the elements. of a task, then a reduction in P300 amplitude by higher-order Once the subject decided that all of the criteria had been flatisfied control should localize the influence of the order variable at and the target and cursor were aligned, a capture button could the earlier processing stages. be pressed and the trial terminated. Figure 41.20 illustrates the subject's task. The target ap­ It was assumed that the alignment phase would be more peared on the screen and moved in a straight line, but at a difficult than the acquisition phase due to increased perceptual randomly selected angle, in the direction ofits exit. The subject demands imposed by the requirement to control the additional had to move the cursor into the neighborhood of the target. The rotational axis. It was predicted, therefore, that the P300 am­ time between the appearance of the target and its acquisition plitude elicited by the tones, associated with an oddball task

'~ 41-40 HUMAN PERFORMANCE run concummtly with the tracking task, would be larger during 3.9.4. P300 and Resource Reciprocity. The studies cited the acquisition than during the alignment phase. above have demonstrated a robust relationship between P300 The ERP results presented in Figure 41.21 confirm these amplitude and the allocation of processing resources in a sec­ predictions. The P300 amplitude is attenuated both as a function ondary task. P300s elicited by secondary-task probes decrease of phase, larger-amplitude P300s being elicited in the acquisition in amplitude with increases in the perceptualJcentral processing phase, and system order, larger P300s elicited during the easier, difficulty of primary tasks. As outlined previously, one of the first-order tracking. Other investigators employing a compen­ basic assumptions of the secondary-task technique is that in­ satory tracking task have also found a systematic relationship creases in primary-task difficulty divert processing resources between P300 amplitude and system order (Wickens, Gill, Kra­ from the secondary task. The decrement in secondary-task per­ mer, Ross, & Donchin, 1981). These studies, along with additive formance is believed to reflect this shift of resources' from the factors investigators of manual control parameters, have pro­ secondary to the primary task. Thus, it is assumed that there vided converging evidence that system order has a salient per­ is a reciprocal relationship between the sources allocated to the ceptualJcentral processing component (Wickens & Derrick, 1981; primary and secondary tasks. Ifthis assumption is correct, then Wickens, Derrick, Micallizi, & Berringer, 1980). The results it should be possible to demonstrate that P300s elicited by task­ might also be useful in the design and evaluation of complex relevant, discrete events embedded within the primary task tracking tasks. If operators are required to perform a tracking are directly related to primary-task difficulty. task with higher-order system dynamics, then concurrently Kramer, Wickens, Vanasse, Heffiey, and Donchin (1981) performed tasks should be designed to minimize perceptual! conducted an experiment in which ERPs were elicited by task­ central processing load. We see here, again, how the ERPs pro­ relevant events embedded within a tracking task. The subjects vide data that increase the theoretical depth with which one were required to perform a single-axis pursuit step tracking can draw conclusions about the human information processing task with either first-order (velocity) or second-order (acceler­ system. ation) control dynamics. In this task, the horizontal position of

,f

1 s1 order targe1s 2nd order targets 1st order cursors 2nd order cursors

Target monitoring task Cursor monitoring task Phase I (acquisition) Phase I (acquisition) ,.·f"->_~~~-:~~~/7;;~:<:.~-::~;.~~ ··...... ------

Phase II (alignment) Phase II (alignment)

, ... ~ ... -··::":':::.-:::.~:::; v::·-...... ·~::::..:::::~···· • J ',_,-1·':.~.-::::)>::.- \ ·.. / ; ·· ..... ······ ' \ . , ', ... /:,­ ' \.,.• I '·\ I \ ,_, /

0 200 400 600 BOO 1000 0 200 400 600 800 1000

Figure 41.21. Representative event-related brain potentials (ERPs) in the study described in Figure 41.20. In each quadrant we are shown ERPs elicited by both target and cursor. However, in the left column. are shown data recorded when the target was relevant, and in the right column data recorded when the cursor was relevant. Note that in each case only the relevant stimulus elicited a P300 even though both were in the monitored part of the visual field. Note also how P300 amplitude elicited during the acquisition phase exceeds the amplitude elicited during the more demanding alignment task. Furthermore, the order of control for the joystick also affected P300 amplitude as predicted. (From A. F. Kramer, C. D. Wickens, & E. Donchin, An analysis of the processing requirements of a complex perceptual-motor task, Human Factors, 1983,25. Copyright 1983 by Human Factors Society. Reprinted with permission.) WORKLOAD-AN EXAMINATION OF THE CONCEPT 41-41 a target was determined by a random series of step displacements occurring at 3-sec intervals. The subject's task was to keep the o Auditory Visual cursor superimposed on the target. Difficulty was varied by 260j 1•0 Counted steps a: l> Track only manipulating two variables: the degree of predictability of the 0 250 a: series of steps and the system order. In the high-predictability a: w 240 condition, the step changes alternated in a regular right-left w a: pattern. In the low-predictability condition the sequence of step <( 230 changes was random. The magnitude of the changes was un­ ;:) 0en t predictable in both conditions. The two dimensions of difficulty, z 220 <( I system order and input predictability, were crossed to create w three conditions of increased difficulty: first-order control of ::::; 210 1- predictable input, first-order control of unpredictable input, and 0 0 200 second-order control of unpredictable input. a: Three different types of probes were employed as ERP­ 150 I eliciting events. In one condition, subjects performed the tracking I i task while also counting the number of occurrences of a low­ 140 pitched tone from a Bernoulli series of high- and low-pitched I 130 tones. In the second condition, subjects counted the dimmer of two flashes in a Bernoulli sequence. The flash appeared as a o Auditory 1.'•. horizontal bar along the path traversed by the target. In the >- 1- 6 • Visual ..J 0 Counted steps primary-task condition, subjects counted the total number of ;:) ~ l> Track only I step changes to the left. Two control conditions were also in­ ()u:: 5 I cluded: one in which the subjects counted the probes but did u.. 1 15 not track, and a second in which subjects performed the tracking w 4 ~ > ~ task without counting the probes. f= ' () 3 The important findings to note in the data presented in ...,w CD Figure 41.22 are the monotonic relations between the tracking ;:) en 2 difficulty manipulations and the subject's perceived ratings of ~ difficulty, as well as those between tracking difficulty and JtMS 1st order error. Both the subjective and behavioral indexes converge on 1st order random 2nd order the same ordering of task difficulty. However, these measures regular random do not provide information concerning the underlying resource TASK DIFFICULTY - structure of the task. Figure 41.22. Subjects in this study performed a step tracking task as a The effect of tracking difficulty on P300 amplitude in the "primary" task concurrently with different oddball tasks serving as "secondary" auditory condition provides results consistent with previous tasks. A target box made discrete jumps to the right or to the left and the research (Isreal et al., 1980; Wickens et al., 1980), as can be subject was required to move, by means of a joystick, a cursor onto the seen in Figure 41.23. Thus, in the auditory condition, an increase moving target The difficulty of the task was varied by manipulating the relationship between the movements of the joystick arid the movement of in the difficulty of the primary task resulted in a decrease in the controlled cursor. This was combined with changes in the degree to the amplitude of the P300 elicited by the secondary-task probes. which the movement of the target was regular (Le., movements in alternate In the visual condition, the introduction of the tracking task directions) or random. The root mean squared error, averaged over subjects, resulted in a reduction in the amplitude of the P300. However, is plotted against task difficulty, as is the subject's estimate of the error increases in tracking difficulty failed to produce any further involved' in each of the experimental conditions. The different lines in each attenuation. In the step conditions, the amplitude of the P300 frame were recorded while different secondary-task conditions were used. elicited by the discrete changes in the spatial position of the In the "auditory" and "visual" series, the subjects counted either visual or controlled element increased wit)t increments in the difficulty auditory probes that were not associated with the step tracking. In the "counted of the primary task. Thus, the hYPothesis of resource reciprocity steps" condition, the subject counted the number of times the step was made between the primary and secondary tasks was confirmed. One in one of the two directions. Thus, in this condition, the oddball stimuli were embedded in the primary task. It was predicted that in this case the amplitude final aspect of the step tracking study has considerable potential of the P300 associated with the secondary task would increase with increases practical utility. The sensitivity of the P300 elicited by visual in primary-task difficulty. (From C. D. Wickens, A. Kramer, l. Vanasse, & E. steps to resource allocation was observed independent of whether Donchin, Performance of concurrent tasks: A psychological analysis of re­ or not the subjects were required to count the stimuli. These ciprocity of information processing resources, Science, 1983, 221. Copyright data suggest that inferences from the P300 about resource al­ 1983 by American Association for the Advancement of Science. Reprinted location and therefore workload can be made in the total absence with permission.) of a secondary-task requirement, a considerable advantage if workload is to be assessed unobtrusively in real-time environ­ As a secondary task, however, the probe task is not entirely 'ments. unobtrusive, and interpretation of the measures still requires 3.9.5. Summary and Conclusions. The investigations re­ the investigator to make certain assumptions about the nature ported above demonstrate that the P300 as a secondary task of the primary-task-secondary-task interaction to make infer­ measure can diagnostically reflect primary-task workload ences conceming operator workload. For this reason the obser­ variations of a perceptual/cognitive nature, uncontaminated by vation that P300 elicited by primary-task stimuli also reflect response factors. The absence of overt response requirements resource allocation are. particularly encouraging to the utility provides it with a considerable advantage over the secondary of the ERP as a measure of workload in extralaboratory envi­ task in that the oddball count task is considerably less intrusive. ronments. c"" 41-42 HUMAN PERFORMANCE

Count only 1st order regular 1st order random 2nd order random

Step count

i; 1"1 I; li i! I; Ill i: Visual Step no count ""':\. ¥'=\'-'.\ I 1./. I ~·-·-,_ I/~··············· \ I ·: 10/-lV ~v.. l· \ \J ""'\""• ·J + ,..

0 327 753 1180 0 327 753 1180

MILLISECONDS MILLISECONDS

Figure 41.23. The event-related brain potentials (ERPs) acquired in the step tracking task described in Figure 41.22. Each frame presents the ERPs elicited by secondary-task probes, as indicated, for the four different levels of difficulty of the primary task (one being total absence of the primary task). Note that for the auditory and visual conditions the data replicate previous results and PJOO amplitude decreases with increasing tracking difficulty. This pattern reverses when the probes are embedded in the primary task, as they do in the "step count" condition. (From A. Kramer, C. D. Wickens, l. Vanasse, E. G. Heffley, & E. Donchin, Primary and secondary task analysis of step tracking: An event-related potentials approach, Proceedings of the Twenty-Fifth Annual Meeting of the Human Factors Society. Copyright 1981 by Human Factors Society. Reprinted with permission.)

4. EPILOGUE concerned with the attempt to clarify the nature of the dimen­ sions along which workload varies. We have tried to explicate Our examination of the workload concept reiterated the frequent the attributes that should be considered in the selection of a statement that workload is a multidimensional, multifaceted measurement procedure. construct. It was unlikely, we suggested, that the manifestations The general goal of the workload analysis has been defined of workload would be captured by one unique, representative as the assessment of the processing and response limitations measure. Our review of the literature confirms this conclusion. of the human information processing system. Our primary thesis We have seen how, time and again, the diverse searches for a is that these limitations are revealed only through the inter­ uniform metric were forced, after initial successes, to admit the actions between an operator and the assigned tasks. We con­ overriding complexity and multidimensionality of the infor­ sidered the nature of the limitations on two levels. dn the more mation processing system. It is impossible to escape the fact theoretical level, we examined efforts to model the limitations that any operator brings to virtually every task a large repertoire on the system by characterizing the invariant, open-loop prop­ ofstructures and processes that can be deployed in a very flexible erties of the human processing system. On a $eCOnd and perhaps manner in the service of the task. Any measurement technique more practical level, we developed the argument that workload must be sensitive to these factors. Much of our discussion was at any specific instant of measurement should be regarded as WORKLOAD-AN EXAMINATION OF THE CONCEPT 41-43 the joint, closed-loop property of the human and the assigned performance. We described throughout this chapter the richness task. and flexibility of the central processing activity, and we cited The discussion of general sources of limitations emphasized data from many experiments to show that the path from stimulus the close affinity between the study ofworkload and the literature and response can be traversed in a variety of ways. The strategy that focuses on attention. Investigators of both workload and used by the subject at the time of evaluation is therefore another attention are interested in the limitations that are placed on determinant to be considered in the analysis of workload. the central processor, and in the cost of mental operations. In Finally, the influence of practice should not be ignored. In

j' the course of studying these two related phenomena, investi­ several sections we discussed the development of automatic gators were forced to develop a detailed account ofthe energetical segments of tasks and the reduction of resource costs in a system and structural characteristics of the central processor. The oftasks and the reduction of resource costs in a system driven linkage between the areas of workload and attention has been by automatic as compared with controlled processes. The actual neglected in the past. We endeavored to show in this chapter level of practice that an operator has at the time of assessment, that it will be useful if investigators ofboth workload and at­ or the desired level of practice that was specified by the assessor tention become aware of each other's concepts and findings. The as a part of the prerequisites of the task, is therefore an impor­ ~·. goals of workload assessment can be better identified and mea­ tant consideration in the estimate of workload. This factor may I,.• sured with reference to the structure of the processing system reduce workload dramatically with the passage of time or in­ ~f i. and to the interrelation between its variables, as they emerge crease it instantly when novelty is encountered. within the attention and information processing research. At­ In the discussion of measurement approaches, we empha­ tention research, in turn, can benefit from and be enriched by sized the importance oftheory-based measures. The underlying a deeper consideration of the costs of mental operations and assumptions, limitations, and major strength of the main classes l~{ ~ their impact on global measures of performance. of measurement procedures were reviewed and evaluated. The In other words, there is a need for developing a more com­ limits of subjective measures were addressed in detail, and a prehensive view of the ways in which the system utilizes its cautionary note was sounded against awarding the subjective degrees of freedom. This question is recognized as crucial for measures a privileged status due to their compelling face validity workload evaluation, but it is almost completely neglected in and technical convenience. Our recommendation for a mea­ research on attention. The vast majority of experiments on surement procedure follows the inclusive definition of workload attention concentrate on isolated and specific comparisons within concept that encompasses both conscious and nonconscious pro­ a limited segment of the system. In. many experiments, the cessing activity. Accordingly, we favor the use of detailed task question of costs is not raised at all, as there is no attempt to activity procedures to uncover the major components of tasks, interpret the effects of the experimental manipulations within followed by a battery of performance-based measures designed the framework of the subject's effort to cope with task demands. to evaluate the load on each component. Matters are equally in need of clarification when we consider the situation in which the measurement of workload takes place. Usually, an individual, or a group, is assigned a task. It has ACKNOWLEDGMENT generally been thought that estimate of workload can be derived from knowledge of properties of the task as well as from the Preparation of this paper, during the long period of its gestation, capabilities of the performer. This argument has been a second has been aided by several organizations and many colleagues. pivot in our discussion. We have emphasized in this chapter Lloyd Kaufman, Ken Boff,Jackson Beatty, Chris Wickens, and the need for a closed-loop assessment of workload to reflect Sandra Hart have commented on different versions of the paper. three basic features of the measurement situation: (1) the fuz­ Maya Weil and Nira Arazi at the Technion and Sharon Cum­ ziness and arbitrariness of the task concept; (2) the degrees of mings at the University of Illinois helped with the preparation freedom available to a person in adopting a specific strategy to of the bibliography. Barbara Mullins prepared the manuscript, cope with task demands and considerations of allocation policy; and Barbara Hartman and Alfreda Mitchell aided in a variety and (3) the influence of practice levels at the time of measure­ of ways. The authors' research programs have been aided by ment. AFOSR, ONR, and DARPA. Additional support was provided The concept of a task in the human performance literature by NASA, Ames Research Center, Man-Vehicle Research Di­ designates a composite entity that can be specified on many vision, under grant NAGW-494 to Daniel Gopher. We appreciate dimensions. These include its formal properties (e.g., modality Lloyd Kaufman's patience while this paper was written tran­ and rate of information presentation, mode of response), the satlantically. Parts of Section 3.9 are revisions of material pre­ selective emphasis on different aspects of performance (e.g., sented by Donchin, Kramer, and Wickens (1982). reaction time, accuracy, retention, comprehension), and expec­ E. Donchin is particularly appreciative of the office provided tations regarding the level of performance on each of the selected at NYU, which allowed him to escape the distraction of Cham­ measures. Deficits in performance and levels of workload are paign-Urbana and concentrate on writing in the serene atmos­ described in terms of these specifications. They are used to phere of Washington Square. The authors have shared the effort define the criteria for sufficiency and adequacy in the assessment and the responsibility in the preparation of this chapter. The of performance. The analysis is, naturally, not sensitive to effects order of authorship was determined by.the toss of a coin. that are outside of the defined features, and the estimate of load may change drastically when any of the specifications is changed. REFERENCE NOTE In a similar manner, it is reasonable to assume that the performer may have more than a single way to cope with the 1. Schneider, W. A distributed processing architecture for attention complexity of tasks, and that different strategies may yield and skill development. Invited paper, Midwestern Psychological similar achievements when evaluated by global measures of Association meeting, May 1983. 41-44 HUMAN PERFORMANCE

REFERENCES Overview and new instrumentation. Behavior Research and In­ strumentation, 1983, 15, 401-405. Allport, D. A., Antonia, B., & Reynolds, P. On the division of attention: Casali, J. G, & Wierwille, W. W. A sensitivity intrusion comparison of A disproof of the single-channel hypothesis. Journal ofExperimental mental workload estimation techniques using flight task empha­ Psychology, 1972, 24, 225-235. sizing perceptual piloting activities. In Proceedings of the 1982 Anderson, J. R. Cognitive skills and their acquisition. Hillsdale, N.J.: International IEEE Conference on Cybernetics. New York: IEEE, Erlbaum, 1981. 1982, pp. 598--aD2. Atkinson, R. C., & Shiffrin, R. M. Human memory: A proposed system Cherry, E. C. On human communication: A review, a summary, and a and its control processes. InK. W. Spence & J. T. Spence (Eds.), criticism. Cambridge, Mass.: M.I.T. Press. New York: Wiley, 1957. The psychology of learning and motivation (Vol. 2). New York: Conrad, R. Speed stress. In W. F. Floyd & A. T. Welford (Eds.), Symposium Academic, 1968. on Human Factors in Equipment Design. London: H. K. Lewis & Attneave, F. Application ofinformation theory to psychology. New York: Co. for the Ergonomic Research Society, 1954. Holt, 1959. Coombs, C. H., Dawes, R. M., & Tversky, A. Mathematical psychology: Baddeley, A. D., & Liberman, K. Spatial working memory. In R. S. An elementary introduction. Englewood Clift's, N.J.: Prentice-Hall, Nickerson (Ed.), Attention and performance VIII. Hillsdale, N.J.: 1970. Erlbaum, 1980. Davis, R. The limits of the "psychological refractory period." Quarterly Baty, D. Human transformation rates in one- to four-axis tracking. Journal ofExperimental Psychology, 1956, 8, 24-38. Proceedings of the Seventh Annual NASA Conference on Manual Davis, R. The human operator as a single-channel information system. Control (NASA, SP A281), 1971. Quarterly Journal ofExperimental Psychology, 1957, 9, 119-129. Beatty, J. Pupillometric methods of workload evaluation: Present status Davis, R. The role of "attention" in the psychological period. Quarterly and future possibilities. In R. Auffret (Ed.), Survey of methods to Journal ofExperimental Psychology, 1959, II, 211-220. assess workload (Agard Proceedings 246). London: Harford House, Deutsch, J. A., & Deutsch, D. Attention: Some theoretical considerations. 1979, pp. 103-110. Psychological Review, 1963, 70, 80-90. Beatty, J. Task-evolved pupillary responses processing load and the Dixon, N. F. Preconscious processing. New York: John Wiley & Sons, structure of processing resources. Psychological Bulletin, 1982, 91(2), 1981. 276-292. Donchin, E. On evoked potentials, cognition, and memory. Science, Berlyne, D. E. Attention and change. British Journal of Psychology, 1975, 190, 1004-1005. 1951,42,269-278. Donchin, E. Brain electrical correlates of pattern recognition. In G. F. Berlyne, D. E. Conflict, arousal and curiosity. New York: McGraw-Hill, Inbar (Ed.), Signal analysis and pattern recognition in biomedical 1960. engineering. New York: John Wiley & Sons, 1975. Berlyne, D. E. Attention as a problem in behavior theory. In D. E. Donchin, E. Event-related brain potentials: A tool in the study of human Mostofsky (Ed.), Attention: Contemporary theory and an analysis. information processing. In H. Begleiter (Ed.), Evoked potentials New York: Appleton-Century-Crofts, 1970. and behavior. New York: Plenum, 1979. Bertelson, P. S-R relationships and reaction time to new versus repeated Donchin, E. Surprise! ... Surprise? Psychophysiology, 1981, 18, 493- signals in a serial task. Journal ofExperimental Psychology, 1963, 513. 65, 478-484. Donchin, E. The dissociation ofelectrophysiology and behavior: A disaster Borg, G. Psychological and physiological studies of physical work. or a challenge? In E. Donchin (Ed.), Cognitive psychophysiology: In W. T. Singleton & D. Whitfield (Eds.), Measurement of man at Event-related potentials and the sudy of cognition. The Carmel work. London: Taylor & Francis, 1971. Conferences (Vol. 1). Hillsdale, N.J.: Erlbaum, 1984. (a) Borg, G. Subjective aspects of physical and mental load. Ergonomics, Donchin, E. The use of ERPs to monitor nonconscious mentation. In 1978, 21, 215-220. S. G. Hart & E. J. Hartswell (Eds.), Proceedings of the Twentieth Borg, G., & Noble, B. S. Perceived exertion. In J. H. Wilmore (Ed.), Annual Conference on Manual Control (NASACP-2341), 1984. (b) Exercise and sports sciences reviews (Vol. 2). New York: Academic, Donchin, E., Kramer, A., & Wickens, C. Probing the cognitive infra­ 1974. structure with event-related brain potentials. In M. C. Frazier & Bostock, H., & Jarvis, M. J. Changes in the form of the-cerebral evoked R. B. Crombie (Eds.), Proceedings of the workshop on flight testing response related to the speed of simple reaction time. Electroen­ to identify pilot workload and pilot dynamics (AFFTC-JR-82-5), cephalography and Clinical Neurophysiology, 1970, 29, 137-145. 1982,pp.371-387. Bratfisch, 0., Borg, G., & Dornic, S. Perceived item difficulty in three Donchin, E., Kubovy, M., Kutas, M., Johnson, R., & Herning, R. I. tests ofintellectual performance capacity (Report No. 29). Stockholm: Graded changes in evoked response (P300) amplitude as a function Institute of Applied Psychology, University of Stockholm, 1972. of cognitive activity. Perception and Psychophysics, 1973, 14, 319- Broadbent, D. E. Perception and communication. London: Pergamon, 324. 1958. Donchin, E., McCarthy, G., Kutas, M., & Ritter, W. Event-related brain Broadbent, D. E. Decision and stress. London: Academic, 1971. potentials in the study of consciousness. In R. Davidson, G. Schwartz, Broadbent, D. E. Task combinations and selective intake of information. & D. Shapiro (Eds.), Consciousness and self regulation (Vol. 3). Acta Psychologica, 1982, 50, 253-296. New York: Plenum, 1983. Broadbent, D. E., & Gregory, M. On the interaction ofS-R compatibility Donchin, E., McCarthy, G., & Kutas, M. Electroencephalographic in­ with other variables affecting reaction time. British Journal of vestigations of hemispheric specialization. In J. E. Desmedt (Ed.), Psychology, 1965,56, 61--a7. Language and hemispheric specialization in man: Cerebral ERPs. Brooks, L. R. The suppression of visualization by reading. Quarterly Progress in Clinical Neurophysiology (Vol. 3). Basel, Switzerland: Journal ofExperimentalPsychology, 1967,19,289-299. Karger, 1977. Brooks, L. R. Spatial and verbal components ofthe act ofrecall. Canadian Donchin, E., Ritter, W., & McCallum, C. Cognitive psychophysiology: Journal of Psychology/Review of Canadian Psychology, 1968, 22, The endogenous components of the ERP. In E. Callaway, P. Tueting, 349-368. & S. Koslow (Eds.), Brain event-related potentials in man. New Cannon, W. B. The wisdom ofthe body. New York: W. W. Norton, 1932. York: Academic, 1978. Carr, T. H. Consciousness in models of human information processing: Donders, F. C. [On the speed ofmental processes.] In W. G. Koster (Ed. Primary memory executive control and input regulation. In and trans.), Attention and performance II. Amsterdam: North-Hol­ G. Underwood & R. Stevens (Eds.), Aspects of consciousness, Vol. land, 1969. (Reprinted from Acta Psychologica, 1969, 30, 412-431.) I, Psychological Issues. New York: Academic, 1979. Duncan, J. Divided attention: The whole is more than the sum of its Casali, J. G., Wierwille, W. W., & Cordes, R. E. Respiratory measurement: parts. Journal ofExperimental Psychology: Human Perception and WORKLOAD-AN EXAMINATION OF THE CONCEPT 41-45

Performance, 1979, 5, 216-228. on Cybernetics and Society, 1981, pp. 609--614. Duncan.Johnson, C. C., & Donchin, E. On quantifying surprise: The Gopher, D., & Brickner, M. On the training of time-sharing skills: An variation in event-related potentials with subjective probability. attention viewpoint. In G. Goonick, M. Hazeltine, & R. Durst (Eds.), Psyclwphysiology, 1977, 14, 456-467. Proceedings of the Annual Meeting ofHuman Factors Society, Santa Easterbrook, J. A. The effects of emotion on cue utilization and the Monica, CaJ., 1980. organization ofbehavior. PsyclwlogicalReview, 1959, 66, 183-201. Gopher, D., Brickner, M., & Navon, D. Different difficulty manipulations Ericcson, K. A., & Simon, H. A. Verbal reports as data. Psyclwlogical interact differently with task emphasis: Evidence for multiple re­ Review, 1980, 87, 215-251. sources. Journal of Experimental Psyclwlogy: Human Perception Ericcson, K. A., & Simon, H. A. Protocol analysis: Verbal report as and Performance, 1982,8, 146-157. data. Cambridge, Mass.: M.I.T. Press, 1984. Gopher, D., & Browne, R. On the psychophysics of workload: Why Fairbank, G., Guttman, N., & Miron, M.S. Effects of time comparison bother with subjective measures? Human Factors, 1984. upon the comprehension of connected speech. Journal of Speech Gopher, D., & Navon, D. How is performance limited-Testing the and Hearing Disorders, 1952, 22, 10-19. notion of central capacity. Acta Psyclwlogica, 1980, 46, 161-180. Fisk, A. D., & Schneider, W. Category and word search: Generalization Gopher, D., & North, R. A. Manipulating the conditions of training in search principles to complex processing. Journal ofExperimental time-sharing performance. Human Factors, 1977, 19, 583-593. Psyclwlogy: Learning, Memory, and Cognition, 1983, 9, 177-195. Gopher, D., & Sanders, A. F. S-OH-R OH stages, OH resources. In Fitts, P. M. The information capacity of the human motor system in W. Printz & A. F. Sanders (Eds.), Cognition and Motor Processes. controlling the amplitude of movement. Journal ofExperimental Berlin: Springer-Verlag, 1984. Psyclwlogy, 1954,42, 381-391. Hallsten, L., & Borg, G. Six rating scales for perceived difficulty (Report Fitts, P. M., & Posner, M. I. Human performance. Belmont, Cal.: Brooks/ No. 58). Stockholm: Institute of Applied Psychology, University of Cole, 1967. Stockholm, 1975. Ford, J. M., Mohs, R. C., Pfefferbaum, A., & Kopell, B.S. On the utility Hamilton, P., Mulder, G., Strasser, H., & Ursin, H. Final report of ofP300 latency and reaction time for. studying cognitive processes. physiological psychology group. InN. Moray (Ed.), Mental workload: In H. H. Kornhuber & L. Deecke (Eds.), Motivation, motor and Its theory and measurement. New York: Plenum, 1979. sensory processes ofthe brain: Electrical potentiols, behavioral and Hart, S. G. Theory and measurement ofhuman workload. InJ. Zeidner clinical use. Amsterdam: North-Holland Biomedical Press, 1980. (Ed.), Human productivity enhancement: Cognitive processes in Ford, J. M.; Roth, W. T., Mohs, R. C., Hopking, W. F., & Kopell, B.S. system design. New York: Praeger Scientific, in press. Event-related potentials recorded from young and old adults during Hart, S. G., Childress, M. E., & Bartolussi, M. Defining the subjective a memory retrieval task. Electroencephalography and Clinical experience of workload. Proceedings of the Twenty-Fifth Annual Neurophysiology, 1979, 47, 450-454. Meeting of the Human Factors Society, 1981, pp. 527-531. Fraise, P. La periode refractoire psychologique. Annee Psyclwlogique, Hauser, J. R., Childress, M. E., Hart, S. G. Rating consistency and 1957,57,315-328. component salience in subjective workload estimation. Proceedings Frazier, M. L., & Crombie, R. B. Proceedings of the workslwp on flight of the Eighteenth Annual Conference on Manual Control, Dayton, testing to identify pilot workload and pilot dynamics (AFFTC-TR- Ohio, 1982. 82-5). Air Force Flight Test Center, Edwards Air Force Base, Cal­ Heffley, E., Wickens, C. D., & Donchin, E. Intramodality selective at­ ifornia, 1982. tention and P300---Reexamination in a visual monitoring task. Friedman, A., & Polson, M. C. The hemispheres as independent resource Psyclwphysiology, 1978, 15, 269-290. systems: A limited capacity processing and cerebral specialization. Hick, W. E. On the rate of gain of information. Quarterly Journal of Journal ofExperimental Psyclwlogy: Human Perception and Per­ Experimental Psyclwlogy, 1952, 4, 11-26. formance, 1981, 7, 1031-1058. Hillyard, S. A. Event-related potentials and selective attention. In E. Friedman, A., Polson, M. C., Defoe, C., & Goskill, S. Diverting attention Donchin (Ed.), Cognitive psyclwphysiology: Event-related potentials within and between hemispheres: Testing a multiple resources and the study of cognition, The Carmel Conferences (Vol. 1). Hills­ approach to limited capacity information processing. Journal of dale, N.J.: Erlbaum, 1984. ExperimentalPsyclwlogy: Human Perception and Performance, 1982, Hirst, W., Spelke, E. S., Reaves, C. C., Charack, G., & Neisser, U. 8, 625--650. Dividing attention without alternation of automaticity. Journal Frowein, H. W. Selective effects of barbiturate and amphetamine on ofExperimental Psyclwlogy: General, 1980, 109, 98-117. information processing and response execution. Acta Pscylwlogica, Hyman, R. Stimulus information as a determinant of reaction time. 1981, 47, 105-115. Journal ofExperimental Psyclwlogy, 1953, 45, 188-196. Fuchs, A. The progression-regression hypothesis in perceptual-motor Isreal, J. Structural interference in dual task performance: Behavioral skill learning. Journal ofExperimentalPsyclwlogy, 1962, 63, 177- and electrophysiological data. Unpublished doctoral dissertation, 182. University oflllinois, 1980. Garner, W. R. Uncertainty and structure as a psyclwlogical concept. Isreal, J. B, Chesney, G. L., Wickens, C. D., & Donchin, E. P300 and New York: John Wiley & Sons, 1962. tracking difficulty: Evidence for multiple resources in dual-task Garner, W. R. The processing' of information and stimulus. Potomac, performance. Psychophysiology, 1980, 17, 259-273. Md.: Erlbaum, 1974. Isreal, J. B, Wickens, C. D, Chesney, G. L., & Donchin, E. The event­ Goff, W. R., Allison, T., & Vaughan, H. G. The functional neuroanatomy related brain potential as an index of display monitoring workload. of the event-related potentials. In E. Calloway, P. Tueting, & S. Human Factors, 1980,22,212-224. Koslow (Eds.), Brain event-related potentials in man. New York: Jagacinski, R. J., Reppenger, D. W., Ward, S. L., & Moran, M.S. A test Academic, 1978. of Fitts' law with moving targets. Human Factors, 1980, 22, 225- Gomer, F. E., Spicuzza, R. J., & O'Donnell, R. D. Evoked potential 233. correlates of visual item recognition during memory scanning tasks. James, W. The principles of psyclwlogy. New York: Holt, 1890. Physiological Psyclwlogy, 1976, 4, 61--65. Jewett, D., Romano, H. W., & Williston, J. S. Human auditory evoked Gopher, D. The contribution of vision-based imagery to the acquisition responses: Possible brain stem components detected on the scalp. and operation of transcription skills. In W. Printz & A. F. Sanders Science, 1970, 167, 1517-1518. (Eds.), Cognition and motor performance. Berlin-Heidelberg: Spring­ Jex, H. R. Two applications of the critical instability task to secondary er-Verlag,.1984. task workload research. IEEE Transaction of Human Factors in Gopher, D. Performance tradeoff under time-sharing conditions: The Electronics, 1967, HFE-8, 279-282. ability of human operators to release resources by lowering their J ex, H. R. A proposed set of standardized subcritical tasks for tracking standard of performance. Proceedings ofthe International Conference workload calibration. In N. Moray (Ed.), Mental workload. New 41-46 HUMAN PERFORMANCE

York: Plenum, 1976. Kutas, M., McCarthy, G., & Donchin, E. Augmenting mental chronom­ Jex, H. R., McDonnei, J.P., & Phatac, A. V. A critical tracking task metry: The P300 as a measure of stimulUs evaluation time. Science, for manual control research. IEEE Transaction ofHuman Factors 1977, 197, 792-795. in Electronics, 1966, HEF-7, 138-144. Laberge, D. Acquisition of automatic processing in perceptual and as­ Kahneman, D. Remarks on attention control. Acta Psychologica: At­ sociation learning. In P.M. A. Rabbitt & S. Dornic (Eds.),Attention tention and Performance III, 1970, 33, 118-131. and performance V. New York: Academic, 1975. Kahneman, D. Attention and effort. Englewood Clift's, N.J.: Prentice­ Laberge, D. Automatic information processing: A review. In S. Long Hall, 1973. & A. Baddeley (Eds.), Attention and performance IX. Hillsdale, Kahneman, D., Beatty, J., & Pollack, I. Perceptual deficit during a N.J.: Erlbaum, 1981. (a) mental task. Science, 1967, 157, 218-219. Laberge, D. Utilization and automaticity in reading. In M. Larsman Kahneman, D., Peavler, W. S., & Onuska, L. Effects of verbalization & H. Hunt (Eds.), Proceedings of the Lake Wilderness Attention on incentive on the pupil response to mental activity. Canadian Conference, July 1981. (b) Journal ofPsychology, 1968,22, 186-196. Laberge, D., & Samuels, S. J. Toward a theory of automatic information Kahneman, D., Peavler, W. S. Insensitive effects and pupillarity changes processing in reading. Cognitive Psychology, 1974, 6, 293-323. in association learning. Journal ofExperimental Psychology, 1969, Levison, W. H. A model for mental workload in tasks requiring contin­ 79, 312-318. uous information processing. InN. Moray (Ed.), Mental workload: Kahneman, D., & Treisman, A. Changing views of attention and auto­ Its theory and measurement. New York: Plenum, 1979. maticity. In R. Parasuraman, J. Beatty, & E. Davis (Eds.), Varieties Lindsley, D. B. Emotion. InS. S. Steven (Ed.), Handbook ofexperimental of attention. New York: John Wiley & Sons, 1983. psychology. New York: John Wiley & Sons, 1951. Kahneman, D., Tversky, B., Shapiro, D., & Crider, A. Pupillary, heart Logan, G. D. Attention in character classification: Evidence for the !~: rate and skin resistance changes during a mental task. Journal of automaticity of component stages. Journal of Experimental Psy­ Experimental Psychology, 1967, 27, 187-196. chology: General, 1978, 107, 32-63. Kantowitz, H. B., & Knight, J. L. On experimental limited processes. Logan, G. D. On the use of a concurrent memory load to measure at­ Psychological Review, 1976, 83, 502-507. tention and automaticity. Journal of Experimental Psychology: Kantowitz, H. B., & Knight, J. L. Testing tapping time-sharing. Journal Human Perception and Performance, 1979, 5, 189-202. ofExperimental Psychology, 1979, 103, 331-336. Logan, G. D. Attention and automaticity in Stroop and priming task: Karis, D., Fabiani, M., & Donchin, E. "P300" and memory: Individual Theory and data. Cognitive Psychology, 1980, 12, 523-553. differences in the von Restorff effect. Cognitive Psychology, 1984, MacCorquadale, K., & Meehl, P. E. On a distinction between hypothetical 16, 177-216. constructs and intervening variables. Psychological Review, 1948, Karlin, L., & Kastenbaum, R. Effects of number of alterations on the 55,95-107. psychological refectory period. Quarterly Journal ofExperimental Mandler, G. Mind and emotion. New York: John Wiley & Sons, 1978. Psychology, 1968, 20, 167-178. Mandler, G. Consciousness: Its function and construction. San Diego: Karlin, L., & Martz, M. J. Response probability and sensory evoked University of California, 1973. potentials. In S. Kornblum (Ed.) Attention and performance N. Mane, A. M., Coles, M. G. H., Wickens, C. D., & Donchin, E. The use New York: Academic, 1973. of the additive factors methodology in the analysis of skill. Pro­ Karlin, L., Martz, M. J., & Mordkoff, A.M. Motor performance and ceedings ofthe Twenty-Seventh Annual Meeting ofthe H umanFac­ sensory evoked potentials. Electroencephalography and Clinical tors Society, 1983, pp. 407-411. Neurophyisology, 1970,28, 307-313. Marcel, G. Conscious and preconscious recognition of polysemous words: Kaufman, L. Sight and mind. New York: Oxford, 1979. Locating the selective effects of prior verbal context. In R. S. Nick­ Keele, S. W. Attention and human performance. Pacific Palisades, Cal.: erson (Ed.), Attention and performance VIII. Hillsdale, N.J.: Erl­ Goodyear, 1973. baum, 1980. Keele, S. W. Behavioral analysis of movement. In V. B. Brooks (Ed.), Martin, D. W. Residual processing capacity during verbal organization Handbook of physiology: Vol. II, Motion control. Baltimore, Md.: in memory. Journal of Verbal Learning and Verbal Behavior, 1970, American Physiological Society, 1981. 9, 391-397. Kelso, S. A., Southard, D. L., & Goodman, D. On the coordination of McCarthy, G., & Donchin, E. A metric for thought: A comparison of two-handed movements. Journal ofExperimental Psychology: Hu­ P300 latency and reaction time. Science, 1981, 211, 77-80. man Perception and Performance, 1979, 5, 229-259. McClelland, J. L. On the time relations and mental processes. An ex­ Kerr, B. Preplanning for aimed movements: Disruption from preliminary amination of systems of processes in cascade. Psychological Review, task. Journal ofExperimental Psychology: Human Perception and 1979, 86, 287-330. Performance, 1983, 9, 596-606. McClelland, J. L., & Rumelhart D. Distributed memory and the rep­ Kinsbourne, M. The mechanism of hemispheric control of the lateral resentation of general and specific information. Journal ofExper­ gradient of attention. In P.M. A. Rabbitt & S. Dornic (Eds.), Attention imental Psychology, 1985, 114, 159-188. and performance V. New York: Academic, 1975. McDonald, L. B. A model for predicting driver workload in the freeway Kinsbourne, M., & Hicks, R. Functional cerebral space. In J. Requin environment: A feasibility study. Unpublished doctoral dissertation, . (Ed.), Attention and performance VIII. Hillsdale, N.J.: Erlbaum, Texas A & M University, 1973. 1978. McDonnell, J.D. Pilot rating techniques for the estimation and evaluation Klein, M., Coles, M.G. H., & Donchin, E. People with absolute pitch of landing qualities (AFFDL-TR-68-76). Dayton, Ohio: Air Force process tones without producing a P300. Science, 1984,223, 1306- Flight Dynamics Laboratory, Wright-Patterson Air Force Base, 1309. 1968. Knowles, W. B. Operator loading tasks. Human Factors, 1963, 5, 155- McLeod, P. A dual-task response modality effect: Support for multi­ 161. processor models of attention. Quarterly Journal ofExperimental Kramer, A. F., & Wickens, C. D., & Donchin, E. An analysis of the Psychology, 1972,29, 651-667. processing requirements of a complex perceptual-motor task. Human McLeod, P. Does probing ·RT measure central processing demand? Factors, 1983, 25(6), 597-621. Quarterly Journal ofExperimental Psychology, 1978, 30, 83-89. Kramer, A., Wickens, C. D., Vanasse, L., Heffiey, E. F., & Donchin, E. Michon, J. A. A note on the measurement of perceptual motor load. Primary and secondary task analysis of step tracking: An event­ Ergonomics, 1964, 7, 461-463. related potentials approach. In R. C. Sugarman (Ed.), Proceedings Michon, J. A. Tapping regularity as a measure of perceptual motor of the Twenty-Fifth Annual Meeting of the Human Factors Society, load. Ergonomics, 1966, 9, 401-412. Rochester, N.Y., 1981. Michon, J. A., & von Doore, H. Equipment note: A semi-portable ap- WORKLOAD-AN EXAMINATION OF THE CONCEPT 41-47

paratus for the measurement of perceptual motor load. Ergonomics, Posner, M. 1., & Snyder, C. R. Attention and cognitive control. In R. L. 1967,10,67-72. Solso (Ed.), Information processing and cognition. Hillsdale, N.J.: Miller, G. A. The magical number seven, plus or minus two.Psychological Erlbaum, 1975. Review, 1956, 63, 87-97. Pribram, K. H., & McGuinness, D. Arousal, activation and effort in the Moray, N. Attention in dichotic listenting: Affective cues and the in­ control of attention. Psychological Review, 1975,82, 116-149. fluence ofinstruction. Quarterly Journal ofExperimental Psychology, Regan, D. Evoked potentials in psychology, sensory physiology, and 1959, 11' 56-60. clinical medicine. London: Chapman & Hall, 1972. Moray, N. Where is· capacity limited? A survey and a model. Acta Reid, G. B., Shingledecker, C. A., & Eggemeier, F. T. Application of € Psychologica, 1967, 27, 84-92. conjoint measurement to workload scale development. Proceedings i Moray, N. Mental workload: Its theory and application. New York: ofthe Human Factors Society Annual Meeting, 1981, pp. 522--526. Plenum, 1979. Reisberg, D. General mental resources and perceptual judgment. Journal I: Moray, N. Subjective mental workload. Human Factors, 1982, 24, 25- ofExperimental Psychology: Human Perception and Performance, I 40. 1983, 9, 966-979. Moray, N., Johansen, J., Pew, R. W., Rasmussen, J., Sanders, A. F., & Reisberg, D., Rappaport, L., & O'Shaughnessy, M. The limits of working Wickens, C. D. Report of the experimental psychology group. In memory: The digit span. Journal of Experimental Psychology: N. Moray (Ed.), Mental workload: Its theory and measurement. Learning and Memory, in press. Itij-~ New York: Plenum, 1979. Ritter, W., Simson, R., & Vaughan, H. G. Association cortex potentials ~ Moscovitz,. M., & Klein, D. Material specific perceptual interference and reaction time in auditory discriminations. Electroencephal­ "' for visual words and faces: Implications for models of capacity lim­ ography and Clinical Neurophysiology, 1972, 33, 547--557. t itations attention and laterality. Journal of Experimental Psy­ Rohrbaugh, J. W., Donchin, E., & Eriksen, C. W. Decision making and chology: Human Perception and Performance, 1980, 6, 590-604. the P300 component of the cortical evoked response. Perception •s; Mulder, G. Sinusarthymia and mental workload. InN. Moray (Ed.), and Psychophysics, 1974, 15, 368--374. Mental workload: Its theory and measurement. New York: Plenum, Rolfe, J. M. The secondary task as a measure of mental load. In W. T. ~t: 1979. Singleton, J. A. Fox, & D. Whitfield (Eds.), Measurement of man Mulder, G., & Mulder, L. J. M. Information processing and cardiovascular at work. London: Taylor & Francis, 1973. control. Psychophysiology, 1981, 18, 392-402. Sanders, A. F. Some variables affecting the relation between relative Murrell, K. F. H. Ergonomics. London: Chapman & Hall, 1969. stimulus frequency and choice reaction time. In A. F. Sanders (Ed.), Navon, D., & Gopher, D. Task difficulty resources and dual-task per­ Attention and performance III. Amsterdam: North-Holland, 1970. formance. In R. S. Nickerson (Ed.) Attention and performance VIII. Sanders, A. F. Some remarks on mental load. InN. Moray (Ed.), Mental Hillsdale, N.J.: Erlbaum, 1980. workload. New York: Plenum, 1979. Navon, D., Gopher, D., Spitz, G., & Chillag, N. On separability and Sanders, A. F. Stage analysis ofreaction processes. In G. E. Stelmach interference between tracking dimensions in dual-axis tracking. & J. Requin (Eds.), Tutorials in motor behavior. Amsterdam: North­ Journal of Motor Behavior, in press. Holland, 1980. Neisser, U. Cognition and reality: Principles and implications ofcognitive Sanders, A. F. Stress and human performance. In G. Salrendy & E. psychology. San Francisco: Freeman, 1976. Smith (Eds.), Machine pacing and occupational stress. London: Nisbett, R. E., & Willson, T. D. Telling more than we can know: Valid Taylor & Francis, 1981. reports on mental processes. Psychological Review, 1977,84,231- Sanders, A. F. Ten symposia on attention and performance: Some issues 259. and trends. In H. Bowma & D. Bauhier (Eds.), Attention and per­ Norman, D. A. Toward a theory of memory and attention. Psychological formance X. Hillsdale, N.J.: Erlbaum, 1983. (a) Review, 1968, 75, 522--536. Sanders, A. F. Toward a model of stress and human performance. Acta Norman, D. A., & Bobrow, D. J. On data limited and resources limited Psychologica, 1983, 53, 61-97. (b) processes. Cognitive Psychology, 1975, 7, 44-64. Sanders, A. F., Wijmen, J. L. C., & Van Arkel, A. E. An additive factors Norman, D. A., & Shallice, T. Attention to action: Willed and automatic analysis of the effects of sleep loss on reaction time. Acta Pscyho­ control of behavior. In M. Lansman & E. Hunt (Eds.), Proceedings logica,1982,51,41-59. of the Lake Wilderness Attention Conference, July 1981. Schneider, W., Dumais, S. T., & Shiffrin, R. M. Automatic and control Ogden, G. D., Levine, J. M., & Eisner, E. J. Measurement of workload proressing and attention. In R. Parasuraman, R. Davis, & J. Beatty by secondary task. Human Factors, 1979, 5, 529-548. (Eds.), Varieties of attention. New York: Academic, 1983. Pew, R. W. Human perceptual motor performance. In B. H. Kantowitz Schneider, W., & Fisk, A. D. Degree of consistent training: Improvements (Ed.) Human information processing: Tutorials in performance and in search performance and automatic process development. Per­ cognition. New York: Erlbaum, 1979. ception and Psychophysics, 1982, 31, 160-168. Picton, T. w., & Stuss, P. T. The component structure of the human Schneider, W., & Fisk, A. D. Attention theory and mechanisms of skilled event-related potentials. In H. H. Kornhuber & L. Deecke (Eds.), performance. In R. A. Magill (Ed.), Memory and control of action. Motivation, motor and sensory processes of the brain: Electrical New York: North-Holland, 1983. potentials, behavioral and clinical use. Amsterdam: North-Holland Schneider, W., & Shiffrin, R. W. Controlled and automatic human in­ Biomedical Press, 1980. formation processing: Decision research and attention. Psychological Pollack, I. The information of elementary auditory display I. Journal Review, 1977, 84, 1-66. of the Acoustical Society ofAmerica, 1952,24, 745-749. Shannon, C. E., & Weaver, W. A mathematical model of communication. Pollack, I. The information of elementary auditory display II. Journal Urbana, Ill.: University of Illinois Press, 1949. of the Acoustical Society ofAmerica, 1953,25, 765-769. Sheridan, T. B. Mental workload-What is it? Why bother with it? Pollack, I., & Ficks, L. Information of elementary multidimensional Human Factors Society Bulletin, 1980, 23, 1-2. auditory display. Journal of the Acoustical Society ofAmerica, 1954, Sheridan, T. B., & Ferrel, W. R. Man-machine system: Information 26, 155-158. control and decision models of human performance. Cambridge, Posner, M. I. Chronometric exploi-ation ofmind. Hillsdale, N.J.: Erlbaum, Mass.: M.I.T. Press, 1974. 1978. Shiffrin, R. M., & Domais, S. T. The development of automatism. In Posner, M. 1., & Boies, S. J. Components of attention. Psychological J. R. Anderson (Ed.), Cognitive skills and their acquisition. Hillsdale, Review, 1971, 78, 391-401. N.J.: Erlbaum, 1981. Posner, M. I., & McLeod, P. Information processing models: In search Shiffrin, R. M., & Schneider, W. Controlled and automatic human in­ of elementary operators. Annual Review of Psychology, 1982, 33, formation processing: Perceptual learning, automatic attending, 477--514. and a general theory. Psychological Review, 1977,84, 119-127. 41-48 HUMAN PERFORMANCE

Schwartz, S. P., Pomerantz, J. R., & Egeth, H. E. State and process Wickelgren, W. Speed-accuracy tradeoff and information processing limitations in information processing: An additive factor analysis. dynamics. Acta Psychologica, 1977, 41, 67-85. Journal of Experimental Psychology: Human PerceptWn and Per­ Wickens, C. D. The effects of divided attention in information processing formance, 1977,3, 402-410. in tracking. Journal ofExperimental Psychology: Human PerceptWn Slater-Hammel, A. T. Psychological refractory period in simple paired and Performance, 1976, 2, 1-13. responses. Research Quarterly ofAmerican Association of Health, Wickens, C. D. The structure of processing resources. In R. Nickerson Physical EducatWn, and Recreatwn, 1958, 29, 468-481. & R. Pew (Eds.), AttentWn and performance VIII. Hillsdale, N.J.: Sokolov, E. W. The modeling properties of the nervous system. In Erlbaum, 1980. I. Maltzman & K. Cole (Eds.), Handbook of contemporary Soviet Wickens, C. D. Workload: In defense of the secondary task. Personnel psychology. New York: Basic Books, 1969. Training and Selectwn Bulletin, 1981, 2, 119-123. Sperando, J. C. The regulation of working methods as a function of Wickens, C. D. Processing resources in attention. In R. Parasuraman, workload among air traffic controllers. Ergonomics, 1978,21, 193- J. Beatty, & R. Davies (Eds.), Varieties of attentWn. New York: 202. John Wiley & Sons, 1983. Sperling, G., & Melchner, M. J. The attention operating characteristic: Wickens, C. D. Engineering psychology and human performance. Co­ Examples from visual search. Science, 1978, 202, 315-318. (a) lumbus, Ohio: Charles F. Merrill, 1984. Sperling, G., & Melchner, M. J. Visual search attention and the attention Wickens, C. D., & Derrick, W. The processing demands of higher-order operating characteristics. In J. Requin (Ed.), Attentwn and per­ manual control: Applicatwn ofadditive factors methodology (EPL- formance VII. New York: Academic, 1978. (b) 81-1/0NR-81-1). Engineering-Psychology Research Laboratory, Squires, N. K., Donchin, E., Squires, K. C., & Grossberg, S. Bisensory University oflllinois, 1981. stimulation: Inferring decision-related processes from the P300 Wickens, C. D., Derrick, W. D., Micallizi, J ., & Berringer, D. The structure :~ component. Journal ofExperimental Psychology: Human PerceptWn of processing resources. In R. E. Corrick, E. C. Haseltine, & R. T. !~: .,, and Performance, 1977, 3, 299-315. Durst (Eds.), Proceedings of the Twenty-Fourth Annual Meeting of !?: Sternberg, S. High-speed scanning in human memory. Science, 1966, the Human Factors Society, Los Angeles, Cal., 1980. 153, 652-654. Wickens, C. D., Gill, R., Kramer, A., Ross, W., & Donchin, E. The Sternberg, S. On the discovery of processing stages: Some extensions cognitive demands of second-order manual control: Applications of Donders' method. Acta Psychologica, 1969, 30, 276-315. (a) of the event-related brain potential. Proceedings ofthe Seventeenth Sternberg, S. Memory scanning: Mental scanning: Mental processes Annual NASA Conference on Manual Control, NASA TM, 1981. revealed by reaction time experiments. American Scientist, 1969, Wickens, C. D., Israel, J. B., & Donchin, E. The event-related rortical 57,421-457. (b) potential as an index of task workload. Proceedings of the Twenty­ Sutton, S., Braren, M., Zubin, J., & John, E. R. Evoked potential cor­ First Annual Meeting ofthe Human Factors Society, Santa Monica, relates of stimulus uncertainty. Science, 1965, 150, 1187-1188. Cal., 1977. Teleford, C. W. The refractory phase of voluntary and associative re­ Wickens, C. D., & Kessel, C. Processing resource demands of failure sponse. Journal ofExperimental Psychology, 1931,14, 1-35. detection in dynamic systems. Journal ofExperimental Psychology: Tichner, E. B. Lectures on the elementary psychology of feeling and Human PerceptWn and Performance, 1980, 6, 569-577. attentwn. New York: Macmillan, 1908. Wickens, C. D., & Kessel, C. Failure detection in dynamic systems. In Treisman, A.M. Contextual cues in selective listening. Quarterly Journal J. Rasmussen & W. B. Rouse (Eds.), Human detectWn and diagnosis ofExperimentalPsychology, 1960,12,242-248. of system failure. New York: Plenum, 1981. Treisman, A.M. Verbal cues: Language and meaning in selective at­ Wickens, C. D., Kramer, A., Vanasse, L., & Donchin, E. Performance tention. American Journal ofPsychology, 1964, 77, 206-219. of concurrent tasks: A psychological analysis of reciprocity of in­ Treisman, A. M. Monitoring and storage of irrelevant images in selective formation processing resources. Science, 1983, 221, 1080-1082. attention. Journal of Verbal Learning and Verbal Behavwr, 1965, Wickens, C. D., Mountford, S. J., & Schreiner, W. S. Time-sharing 3, 449-459. efficiency: Evidence for multiple resources, task hemispheric in­ Treisman, A. M. Strategies and models of selective attention. Psycho­ tegrity and against a general ability. Human Factors, 1982, 23, logical Review, 1969, 76, 282-299. 211-229. Treisman, A.M., & Gelade, G. L. A future integration theory of attention. Wickens, C. D., & Sandry, D. L. Task hemispheric integrity in dual­ Cognitive Psychology, 1980, 12, 97-136. task performance. Acta Psychologica, 1982, 52, 227-248. Trumbo, D., & Milone, F. Primary task performance as a function of Wickens, C. D., Sandry, D. L., & Micolizzi, S. A validity of the spatial encoding retention and recall in secondary task. Journal of Ex~ variant of the Sternberg memory search task (Tech. Rep. EPL-81- perimental Psychology, 1971, 91, 273-278. 2/0NR-81-2). Engineering-Psychology Research Laboratory, Uni­ Underwood, G. Memory system and conscious processes. In G. Underwood versity oflllinois;June 1981. & R. Stevens (Eds.), Aspects of conscwusness, Vol. I, Psychological Wickens, C. D., Sandry, D. L., & Vidulich, M. Compatibility and resource issues. New York: Academic, 1979. competition between modality of input, central processing, and Vidulich, M. K., & Wickens, C. D. Time sharing, manual control and output: Testing a model of complex task performance. Human Fac­ memory search: The effects of input and output modality competitWn tors, 1983, 25, 227-248. prwrities and control order (Tech. Rep. EPL-81-41/0NR-81-4). Wickens, C. D., & Yeh, Y. Y. The dissociation between subjective work­ University of lllinois, December 1981. load and performance: A multiple resource approach. Proceedings Wachtel, P. -L. Conception of broad and narrow attention. Psychological ofthe Twenty-Seventh Annual Meeting ofthe Human Factors Society, Bulletin, 1967, 68, 417-429. Norfolk, Va., 1983. Welford, A. T. The "psychological refractory period" and the timing of Wierwille, W. W., Gutman, J. C., Hicks, T. G., & Muto, W. H. Secondary high-speed performance: A review and theory. British Journal of task measurements of workload as a function of simulated vehicle Psychology, 1952,43, 2-19. dynamics and driving conditions. Human Factors, 1977,19, 557- Welford, A. T. Evidence of a single-channel decision mechanism limiting 566. performance in a serial reaction task. Quarterly Journal of ]$x­ Wierwille, W. W., Skipper, J. H., & Rieger, A. C. Decision free rotating perimental Psychology, 1959, 2, 193-210. scales for workload and estimation: Theme and variations. Pro­ Welford, A. T. A single channel operation in the brain. Acta Psychologica, ceedings of the Twentieth Annual Conference on Manual Control, 1967' 27, 5-22. Sunnyvale, Cal., 1984. Whitaker, L. Dual-task interference as a function of cognitive processing Wilkson, R. T., & Morlock, H. C. Auditory evoked response and reaction load. Acta.Psychologica, 1979,43, 71-84. time. Electroencephalography and Clinical Neurophyswlogy, 1967,

,,t' WORKLOAD-AN EXAMINATION OF THE CONCEPT 41-49 23,50-56. Woody, C. D. Characterization of an adaptive filter.for the analysis of Williges, R. C., & Wierwille, W. W. Behavioral measures of aircrew variable latency neuroelectric signals. Medical and Biological En­ mental workload. Human Factors, 1979,21, 549-574. gineering, 1967,5,539-553. Wood, C. C., & Allison, T. Interpretation of evoked potentials: A neu­ Yerkes, R. M., & Dodson, J.D. The relation of strength of stimulus to rophysiological perspective. Carwdian Jourru:d ofPsychology, 1981, rapidity of habit formation. Journal of Comparative Neurology of 35, 113-135. Psychology, 1908, 18, 459_:_482.

I'11 1""'