Acta Psychologica 139 (2012) 471–485

Contents lists available at SciVerse ScienceDirect

Acta Psychologica

j ourna l homepage: www.e lsev ie r .com/ locate /actpsy

Spatial and temporal task characteristics as stress: A test of the dynamic adaptabilitytheory of stress, workload, and performance

James L. Szalma ⁎, Grace W.L. TeoUniversity of Central Florida, USA

⁎ Corresponding author at: Performance Research Lament, University of Central Florida, 4000 Central Flor1390, USA. Tel.: +1 407 823 0920; fax: +1 407 823 58

E-mail address: James.Szalma@ucf.edu (J.L. Szalma).

0001-6918/$ – see front matter © 2012 Elsevier B.V. Alldoi:10.1016/j.actpsy.2011.12.009

a b s t r a c t

a r t i c l e i n f o

Article history:Received 21 September 2010Received in revised form 16 December 2011Accepted 21 December 2011Available online xxxx

Keywords:StressWorkloadPerformanceStress theoryTask demand

The goal for this study was to test assertions of the dynamic adaptability theory of stress, which proposes twofundamental task dimensions, information rate (temporal properties of a task) and information structure(spatial properties of a task). The theory predicts adaptive stability across stress magnitudes, with progres-sive and precipitous changes in adaptive response manifesting first as increases in perceived workload andstress and then as performance failure. Information structure was manipulated by varying the number ofdisplays to be monitored (1, 2, 4 or 8 displays). Information rate was manipulated by varying stimuluspresentation rate (8, 12, 16, or 20 events/min). A signal detection task was used in which critical signalswere pairs of digits that differed by 0 or 1. Performance accuracy declined and workload and stress increasedas a function of increased task demand, with a precipitous decline in accuracy at the highest demand levels.However, the form of performance change as well as the pattern of relationships between speed and accuracyand between performance and workload/stress indicates that some aspects of the theory need revision.Implications of the results for the theory and for future research are discussed.

© 2012 Elsevier B.V. All rights reserved.

1. Introduction

Task induced stress is a ubiquitous phenomenon that can substan-tially impair an individual's performance, safety, andwell-being. Theoryand research on human performance under stress have tended toemphasize either the stimulus properties (e.g., temperature, vibration,noise; for recent meta-analytic reviews, see Conway, Szalma, &Hancock, 2007; Hancock, Ross, & Szalma, 2007; Szalma & Hancock,2011) or physiological mechanisms of response such as the generaladaptation syndrome (Selye, 1976) or unitary arousal (Hebb, 1955).However, these single mechanism approaches proved to be insufficientto account for the relationship between stress and performance(Hockey, 1984; Hockey&Hamilton, 1983). The failure of unitary arousaltheory in particular led to the development of an energetic resourceperspective (Hockey, 1984; 1997; Hockey, Gaillard, & Coles, 1986)in which stress effects result from a person–environment relation thatis appraised by an individual as potentially or actually exceeding his/her resources to effectively adapt to the event (Lazarus, 1999; Lazarus& Folkman, 1984).

boratory, Psychology Depart-ida Blvd. Orlando, FL 32816-62.

rights reserved.

One of these theories, the Dynamic Adaptability Theory (DAT;Hancock & Warm, 1989), was developed specifically for understand-ing the mechanisms underlying the effects of stress and workloadon adaptive response as it relates to task performance. This theory in-corporates both the stimulus and response elements of stress into asingle framework, the trinity of stress (see Fig. 1). The environmentaldemands are represented as the input, which are the deterministicproperties of the task and the immediate physical and social environ-ment that can be specified independent of an individual's response.As most environments impose multiple demands, the input may beviewed as an external demand profile or ‘signature’ (c.f., Hockey &Hamilton, 1983). Note that for DAT ‘input’ refers to the demands im-posed by the environment (including the task itself), and not only thesensory/perceptual input that is an element of most information pro-cessing theories (e.g., Wickens & Hollands, 2000). That is, ‘input’ inthis model reflects each step of the information processing stream(i.e., sensory/perceptual input, the processing requirements of thetask, and response options; Wickens & Hollands, 2000). In addition,use of the term ‘stress’ in Fig. 1 refers to environmental demand rath-er than the response of the organism (the latter is represented in theother two elements described below). The second component isadaptation, which consists of a general (nomothetic) response of theorganism to the input. These responses are common to members ofthe species (e.g., release of stress hormones, mechanisms of arousal,appraisal and coping). Finally, the output is the response of the organ-ism that is dependent upon the characteristics of the individual and is

Fig. 1. The three elements of stress (the ‘trinity of stress’) incorporated into the dynam-ic adaptability model (from Hancock & Warm, 1989). Note. See footnote 1 regardinguse of the term, ‘idiographic’.

472 J.L. Szalma, G.W.L. Teo / Acta Psychologica 139 (2012) 471–485

thus idiographic1 (e.g., coping strategy selected, appraisal content,levels of arousal or stress hormones). Note that psychomotorresponse can be either adaptation (e.g., stereotypical responses) or‘idiographic’ output (e.g., dependent on skill level).

A fundamental tenet of the dynamic adaptability model is thatacross a relatively wide range of stress magnitude individuals suc-cessfully adapt to the demands imposed upon them (see Fig. 2). Thetheory specifies three general modes of adaptive function: dynamicstability (successful adaptation to demands), dynamic instability(vulnerability to failure to effectively adapt to demands), and thetransition between these two states (the dotted lines in Fig. 2). Notethat the model incorporates both increases (hyperstress) and de-creases (hypostress) in environmental stimulation. Inclusion of thelatter accounts for the stress associated with the relative absence ofstimulation or environmental demand, although to date direct evi-dence for such effects in non-sensory, cognitively complex tasks hasbeen lacking.

The theory predicts adaptive stability across a wide range of envi-ronmental demand, represented as the plateau in Fig. 2. However,increase or decrease in environmental demand to more extremelevels results in instability in adaptation. As can be seen in Fig. 2,this instability is progressive, such that declines in subjective comfort(e.g., perceived stress and workload; the regions bounded by pointsAC and AΨ in Fig. 2) begin to occur at levels of stress exposure thatare lower than the levels at which performance decrements areobserved (the regions bounded by points AΨ and AP in Fig. 2).When the level of stress exceeds the capacity of an adaptive response(i.e., comfort, performance) the resulting range of stress input is azone of dynamic instability in which the form of adaptation becomesunstable and less effective. For comfort these are manifested as in-creased perceived workload and stress, compensatory physiologicalresponse (e.g., autonomic arousal, stress hormone release) and com-pensatory effort (c.f. Hockey, 1997).

Note that in their original conception, Hancock and Warm (1989)regarded hyperstress as an excessive amount of stimulation (toomuch of something) and hypostress as an insufficient magnitude ofenvironmental stimulation or demand. For instance, they noted thatthermal stress occurs at both very high temperatures (hyperstress)and very low temperatures (hypostress). Performance impairmentoccurs at both extremes (Hancock et al., 2007). In contrast, theyidentified acoustic noise as exerting its negative effects only in thehyperstress region, as low levels of acoustic noise weakly impactsperformance, although recently it has been established that this de-pends on duration of exposure (Szalma & Hancock, 2011).

Transition from stable to unstable psychological adaptation (AΨ)results in a zone of dynamic instability that manifests as a perfor-mance decrement. The outermost region is failure of physiologicaladaptation (AP) which can manifest as unconsciousness (Harris,

1 Note that Hancock and Warm (1989) did not view ‘idiographic’ as unsystematic, acompletely idiosyncratic pattern of response, or the absence of nomothetic mecha-nisms. They argued that ‘output’ consists of systematic (i.e., nomothetic) individual dif-ferences in response to stress. It does not mean that one cannot predict output, butrather that prediction requires understanding the person factors that influence the re-sponse. Perhaps the original theoretical exposition should have used the term ‘personfactors’ or ‘individual differences’ instead of ‘idiographic.’ However, the meaning of thelatter term has often been misunderstood (Lamiell, 2003).

Hancock, & Harris, 2005) or the failure of the bodily mechanisms forphysical adaptation, as occurs in conditions of prolonged thermalstress above 85 °F effective temperature (Hancock et al., 2007) orwhen there is sufficient noise intensity to cause sensory damage.The model explicitly defines psychological adaptation in terms ofthe individual's attentional resource capacity, and physiological adap-tation is considered a homeostatic regulation mechanism. The bound-aries or transition modes between the zones of adaptability may becontinuous, in which case the functions shown in Fig. 2 representthresholds of adaptive stability. However, Hancock and Warm (1989)argued that the boundaries may instead represent discontinuities or‘catastrophe cusps’ in the transition between states of adaptation.This contrasts with the argument by Hockey (1997, 2003) that task-induced stress induces graceful performance degradation rather thanprecipitous decline.

1.1. The meaning of ‘adaptive instability’

It may seem counterintuitive to consider increases in perceivedworkload and stress as ‘instability’ of adaptation. Although stressand high workload are often subjectively unpleasant, they includecompensatory responses that preserve behavioral and physiologicaladaptation (c.f., Selye, 1976). Such outcomes do not indicate manifestinstability in response and they are not if the sole criterion for effec-tive adaptation is overt performance. However, from the energetic re-sources perspective the effectiveness of adaptation includes its costsin terms of psychological and physiological energy (e.g., compensato-ry effort; Hockey, 1997). This cost is based on the assumption thatbiological systems seek equilibrium states of minimal energy expen-diture. An increase in workload or stress therefore represents anadaptive ‘instability’ even when performance is preserved becausethe maintenance of the latter increases the energetic costs to theorganism, and these costs can render the individual more vulnerableto performance failure as a result of depleted energetic resources.Thus, ranges of task load in which performance is maintained atthe cost of compensatory effort (manifested in measures of workloadand stress) are considered ‘latent performance decrements’ (Hockey,1997).

1.2. Tasks characteristics

A unique feature of the dynamic adaptability theory is that it ex-plicitly identifies the task a person is performing as the dominant

Fig. 2. The dynamic adaptability model of Hancock and Warm (1989). Note. AN =adaptive threshold of the normative zone; AC = adaptive threshold of the comfortzone; Aψ = threshold of the psychological zone of adaptability; AP = threshold ofthe zone of physiological adaptability; and AF = zone of adaptive failure at all levels.

Fig. 3. The dynamic adaptability model of Hancock andWarm (1989), with the stress dimension partitioned into two fundamental dimensions of information from the environment(task): information structure, referring to the spatial representation of the task that conveys its meaning to the individual, and information rate, the temporal characteristics of thetask. Note. AN = adaptive threshold of the normative zone; AC = adaptive threshold of the comfort zone; Aψ = threshold of the psychological zone of adaptability; AP= threshold ofthe zone of physiological adaptability; and AF = zone of adaptive failure at all levels.

2 One could consider insensitivities as simple dissociations, and dissociations as dou-ble dissociations, with the former being a subtype of the latter. However, for the sake oftheoretical clarity and empirical consistency with previous research (e.g., Oron-Gilad,Szalma, Stafford, & Hancock, 2008) we retain Hancock's (1996) original terminology.

473J.L. Szalma, G.W.L. Teo / Acta Psychologica 139 (2012) 471–485

source of environmental demand, and it specifies two basic dimen-sions of information that constitute the task component of thebroader demand signature. These dimensions are shown in Fig. 3,which is derived by rotation of Fig. 2 around the ordinate and parti-tioning the stress axis into its component vectors. Information struc-ture refers to the spatial organization of a task (and its resultantmeaning for the individual), and information rate refers to the tempo-ral properties of a task. Note that each of the two task dimensions isitself composed of multiple components. For instance, the change inperformance with time on task manifests as a task duration compo-nent of information rate, while the rate of events represents the pac-ing aspect, and stimulus duration a third component. Hence, betterlabels for these information components may be ‘spatial’ and ‘tempo-ral’ dimensions (c.f., Hancock & Szalma, 2008).

Magnitudes of information structure and rate that are too low ortoo high induce instability in adaptation. In principle, if these dimen-sions could be quantified accurately then the collective demandscould be represented as a vector that determines the adaptive stateof the individual. Such a representation would also permit inclusionof other environmental stimuli (e.g., noise, temperature). Althoughthe vector approach is potentially useful, it is premature to specify ageneral vector function for performance under stress because of thelimited understanding of interactive effects of multiple sources ofstress (Hancock & Szalma, 2008) and the difficulty in quantifyinginformation processing (McBride & Schmorrow, 2005). For instance,Hancock and Warm (1989) explicitly represented the two taskdimensions as non-orthogonal, but it is not yet clear how this interac-tivity should be represented within a vector model, or how taskdimensions interact with other sources of stress and how theseshould be represented mathematically. These theoretical refinements

require research in which these inter-relationships can be examinedempirically. The present study represents one such effort.

1.3. Performance–workload relationships

One way in which the zones of dynamic instability may manifestin the context of task-induced stress is in the relationship between per-formance (behavioral response–psychological adaptation, AΨ) and per-ceived workload (comfort, AC). Yeh and Wickens (1988) and Hancock(1996) argued that the pattern of relationships of task manipulationsto performance and to perceived workload can reveal the effects oftask characteristics on resource supply and demand, and the strate-gic allocation of resources. Yeh and Wickens (1988) defined casesin which a manipulation impairs performance and increases work-load as associations, because both measures indicate an increasein resource demand. All other cases are dissociations, because onlyone measure indicates increased resource demand. However, Han-cock differentiated between performance–workload dissociation, inwhich one measure indicated greater consumption of resources(lower performance or higher workload) but the other measured in-dicated fewer resource demands (improved performance or lowerworkload). Manipulations in which one measure increased or de-creased while the other remained stable are insensitivities.2

474 J.L. Szalma, G.W.L. Teo / Acta Psychologica 139 (2012) 471–485

Hancock (1996; see also Parasuraman & Hancock, 2001) arguedthat these patterns were diagnostic of different adaptive responsesto demands. Hence, performance insensitivity indicates either a com-pensatory response (when workload increases), such that stabilityin performance is maintained only at the expense of greater effort;or the development of task-related skills (when workload decreases)that allow the performer to reduce resource allocation to a task whilemaintaining a stable level of performance. Workload insensitivitiesreflect either insensitivity of the performer to the quality of his/herbehavioral output (when performance declines) or to a floor/ceilingeffect with respect to perceived workload. When performance im-proves workload insensitivity may indicate that the performer hasdeveloped task-related skills so that performance requires fewerresources, but the individual has chosen to maintain his/her level ofresource investment, thereby improving performance. Note thatwhen task-related skills develop, a strategic decision by the perform-er determines whether there is performance insensitivity (i.e., theperformer chooses to not allocate excess resources to the task) orworkload insensitivity (resources are allocated to the task). Dissocia-tions occur either because the person ‘gives up’ because they do notbelieve that can achieve task goals (workload and performance bothdecline), or because the demands are growing but the person isresponding successfully so that performance is enhanced at the costof greater workload (both measures increase).

The patterns of performance–workload relationships can bemapped to the zones of the dynamic instability shown in Fig. 2 (fora more detailed treatment see Oron-Gilad et al., 2008). Region AN inthe Figure represents the normative zone within which an individualis in their normal equilibrium state. Fluctuations of external demandswithin this region do not require adaptive response and therefore nochange in performance or workload is observed; the self-regulation ofbehavior and information processing is experienced as relativelyeffortless and automatic. The threshold at point AN represents thepoint at which changes in the magnitude of the stressor require com-pensatory response (at both the physiological and psychological levelof analysis) to maintain stable adaptation to changing demands byallocating effort to preserve performance. However, in region AN–AC

the individual remains within their comfort zone, so engagement ofresources in response to increased demand is not appraised as stress-ful. In this region one would expect that increased task load would in-crease engagement in the task and energetic arousal (but not tensearousal; Helton, Matthews, & Warm, 2010; Matthews et al., 2002).Regions AN and AN–AC correspond to the engaged mode of stressresponse described by Hockey (1997, 2003), and in region AN–AC

this can manifest as a dissociation in which performance improveswith greater allocation of effort. The individual is engaging in ‘effortwithout distress’ (Hockey, 1997, 2003).

At higher magnitudes of demand more resources must be allocat-ed to the task if performance is to be maintained. If the person is op-erating at or near capacity they may enter region AC–AΨ, in whichadaptation in terms of subjective state (comfort) becomes unstable(i.e., perceived workload and stress increase) but the person canmaintain performance. This corresponds to what Hockey (1997,2003) described as strain mode (‘effort with distress’) and manifestsas performance insensitivity (workload and stress increase but per-formance is maintained). Further increase in demand leads to perfor-mance failure, as the demands exceed the resource capacity of theperson to maintain performance even with increased effort (regionAΨ–AP). This can reflect either a strain mode with performance–workload association in which performance declines and workloadincreases, or to what Hockey (1997, 2003) referred to as a disengagedmode in which both performance and workload decrease (a dissocia-tion) due to disengagement from the task (‘no effort with no dis-tress’). Note that whichever of the two possible outcomes in regionAN–AC and AΨ–AP occurs is determined by whether the individualchooses to allocate more effort to task performance (Hockey, 1997).

1.4. Tests of the theory

There have been limited attempts to directly test the predictionsof the Dynamic Adaptability Theory regarding changes in perfor-mance–workload relationships as a function of variation in informa-tion structure and rate. In one early attempt, Hancock and Caird(1993) adapted the theory to test a model of mental workloadin which they proposed that workload increases occur when theperson's perceived distance from goal achievement increases and/orwhen the time available for action decreases. If the former is consid-ered a formof information structure and the latter a formof informationrate, their model of workload can be viewed as a special application ofthe more general Dynamic Adaptability Theory. Hancock and Caird(1993) tested their model using a computer-based task in which agrid of circles was presented and participants were to mouse clickon a specific sequence of circles that were decreasing in size. Goaldistance was manipulated by varying the number of movementsrequired to navigate a sequence of circles (i.e., the number of steps inthe sequence), and time for action was manipulated by modulatingthe rate at which the circles became smaller (shrink rate). They foundthat increases in task demand impaired performance and increasedworkload in a manner consistent with their model. However, Hancockand Caird (1993) did not analyze their results in terms of the relation-ship between performance and mental workload to evaluate the moregeneral Dynamic Adaptability Theory, nor did they use a multidi-mensional appraisal-based measure of perceived stress that includestask engagement as a component.

More recently, Oron-Gilad et al. (2008) reported performanceworkload relationships in a field study of police officers engaged infirearms training tasks. However, in that study the task dimensionsidentified in the Dynamic Adaptability Model could not be manipulat-ed, which limited the strength of inferences regarding the patternof effects. The goal for the present study was to evaluate the DynamicAdaptability Theory by manipulating the task structure and rate andexamining the resultant performance–workload and performance–stress relationships. Based on the model, the form of adaptation (i.e.,comfort, performance) and the level of information demand shoulddetermine the adaptive state of the individual (i.e., the degree towhich the person can effectively maintain comfort, performance,etc.). Stable adaptation is expected at lower levels of demand withprecipitous declines in stability at higher levels of task load. Further,as explicitly noted by Hancock and Warm (1989), the two taskdimensions will likely interact in influencing response as a functionof demand (i.e., the two dimensions will not be orthogonal). The spe-cific hypotheses are summarized below.

1) Higher levels of information demand will be associated with insta-bility in effective adaptation as measured by perceived workload,cognitive state (i.e., stress), and performance.

2) These changes will occur progressively. Based on the nested struc-ture shown in Fig. 2, workload and stress are expected to increaseat a level of demand lower than that associated with performancechange, although task engagement may also increase if the personengages compensatory resources.

3) The two information dimensions will have interactive effects, suchthat the magnitude of changes in performance, workload, andstress as a function of information rate will be larger as the de-mands on the information structure dimension are increased.

4) The relation of outcome measures (adaptive responses) to taskdemandwill be curvilinear, with a precipitous change at the thresh-olds of adaptation. An alternative hypothesis, based on Hockey(1997; 2003) is that graceful degradation will be observed (i.e., ashallower slope for the degradation function).

These hypotheses were tested by manipulating the two dimen-sions specified by the Dynamic Adaptability Theory, and evaluatingtheir effects on the different forms of adaptation defined in terms of

475J.L. Szalma, G.W.L. Teo / Acta Psychologica 139 (2012) 471–485

outcome measures. Information structure was manipulated by vary-ing source complexity (i.e., the number of displays to be monitored;Davies & Parasuraman, 1982), and information rate was manipulatedby varying the rate of stimulus presentation. Subjective comfortwas measured in terms of perceived workload and stress, and psy-chological adaptation was measured in terms of performance, perHancock and Warm's (1989) original interpretation. A short duration(12 min) signal detection task was employed, in which critical signalswere pairs of digits that differed by 0 or 1.

2. Method

2.1. Participants

Three hundred and eighteen psychology undergraduates (126males, 192 females) at a large southeastern U.S. university wererecruited for the study, and received course credit in exchange forparticipation. Participant ages ranged from 17 to 32 years (M=18.4,SD=1.5).

2.2. Task

The task consisted of an adaptation of the cognitive vigilance taskemployed by Warm, Howe, Fishbein, Dember, and Sprague (1984),in which participants monitored the visual presentation of 2-digitnumbers ranging from 01 to 99. The displays were created usingMicrosoft PowerPoint in 28 point Arial font. Participants wereinstructed to respond by pressing the spacebar on a computer key-board when a critical signal appeared on the screen. Critical signalswere defined as instances in which the two digits differed by 0 or 1(e.g. 01, 54, 99), and all other digit pairs were defined as neutralevents requiring no overt response from the observer.

Throughout the task, two 2×2 grids appeared side-by-side on thescreen and each of the 8 cells was a “display” in which the digitsappeared (see Fig. 4). The two grids were separated in order to sim-plify the manipulation of the 2- and 4-display conditions, and alsoto maintain the 2×2 grid format used in previous research on sourcecomplexity in vigilance and signal detection (e.g., Grubb, Warm,Dember, & Berch, 1995). For the 1, 2 and 4 display conditions, the lo-cations of the monitored displays were selected at random for eachparticipant, with the restriction that the displays to be monitoredwere always within the same 2x2 grid. Before each condition, therewas a ‘notification screen’, on which a red outline surrounding thedisplay(s) notified participants of the number and location(s) of thedisplay(s) to be monitored for that condition. The red outline wasnot present during the tasks.

Across all experimental conditions stimulus duration was 2500 ms,determined via a pilot study which indicated that this was the mini-mum duration required to scan all 8 displays. The number of criticalsignals was set at 10 per 3-minute block of trials across all conditions.The number of neutral events differed for the different levels of eventrate. For each condition there were four continuous 3-minute blocks, apractice block and three test blocks. The task duration was thus12 min for each condition, but only the three test blocks (9 min) wereused in the analyses of results.

Fig. 4. Example of the displays used in the present study.

Note that although holding the number of targets constant pro-vides an equivalent number of critical signal exposures, it doesconfound event rate with signal probability. One can hold signal prob-ability constant, but this would then confound event rate with thenumber of critical signals presented. Following the vigilance litera-ture, we held the number of targets constant and permitted signalrate and event rate to covary as this is common practice in vigilanceresearch (e.g., Lanzetta, Dember, Warm, & Berch, 1987; Jerison &Pickett, 1964; Loeb & Binford, 1968; Parasuraman, 1979; Taub &Osborne, 1968). Although studies of visual search have consistentlyfound that performance is negatively related to target prevalence(Wolfe, Horowitz, & Kenner, 2005; Wolfe et al., 2007), Warmand Jerison (1984) noted that the evidence indicates that “the effectsof event rate transcend factors in the simple probability of signals”(page 23) because detection probability is negatively related to eventrate even when signal probability is equated.

2.3. Experimental design

The experiment consisted of a 4 (information rate) by 4 (informa-tion structure) factorial design with repeated measures on the secondfactor. Information rate was manipulated by varying the rate of stim-ulus presentation (event rate). The four event rates were 8 (n=80),12 (n=79), 16 (n=80), and 20 (n=79) events/min. Each partici-pant was assigned at random to one of these four levels. Informationstructure was manipulated by varying the source complexity (i.e., thenumber of the displays to be monitored: 1, 2, 4 or 8; 0, 1, 2, or 3 bits ofuncertainty, respectively; Davies & Parasuraman, 1982; c.f., Grubbet al., 1995). Note that source complexity is analogous to set size invisual search paradigms, and it has been well established that perfor-mance declines as a function of increasing set size (e.g., Benjamins,Hooge, van Elst, Wertheim, & Verstraten, 2009; Cameron, Tai, Eckstein,& Carrasco, 2004; Pavlovskaya, Ring, Groswasser, Keren, & Hochstein,2001; Treisman & Gelade, 1980; Wolfe, 1994). The order in whichthese four levels were presented was balanced across participantsvia a Latin Square.

2.4. Measures

2.4.1. Perceived workloadThe NASA-Task Load Index (TLX; Hart & Staveland, 1988) served

as the measure of perceived workload. The TLX provides an index ofglobal workload as well as ratings on six subscales, three of whichreflect appraisals of the task (Mental Demand, Physical Demand,Temporal Demand) and three of which reflect appraisals of the parti-cipant's own response to the task (Perceived Performance, Effort,and Frustration). The rating scales range from 0 to 100. Each ‘subscale’is a single item, and global workload is computed as an average ofthe six scales.

2.4.2. StressPerceived stress was measured using the short twenty-item ver-

sion of the Dundee Stress State Questionnaire (DSSQ; Matthewset al., 1999, 2002) which consists of three scales that reflect the cogni-tive, affective, and motivational dimensions of stress. Task engagementreflects the cognitive (concentration) and energetic/motivational(energetic arousal, motivation) components of cognitive state; dis-tress reflects both cognitive (perceived confidence and control) andaffective (hedonic tone and tense arousal) aspects of stress; and worryreflects the cognitive dimension of stress (task-related and task-unrelated thoughts, self-focused attention, and self-esteem). Matthewset al. (2002) have identified the core relational theme (Lazarus, 1999)associated with each scale. Task engagement is linked to the theme ofcommitment of effort, distress is associated with the theme of overloadof processing capacity, and worry reflects the theme of self-evaluation.Participants rated the DSSQ items (7 task engagement items, 7 distress

Table 1aMeans and standard deviations for performance measures.

Event rate(per min)

Displays n

1 2 4 8

Proportion of correct detections8 0.95 (0.14) 0.89 (0.18) 0.87 (0.16) 0.55 (0.16) 8012 0.93 (0.19) 0.88 (0.17) 0.75 (0.18) 0.51 (0.15) 7916 0.90 (0.21) 0.86 (0.17) 0.74 (0.16) 0.49 (0.14) 8020 0.88 (0.21) 0.76 (0.21) 0.63 (0.18) 0.28 (0.12) 80

Proportion of false alarms8 0.01 (0.03) 0.03 (0.06) 0.02 (0.03) 0.10 (0.09) 8012 0.01 (0.03) 0.01 (0.03) 0.02 (0.04) 0.05 (0.04) 7916 0.02 (0.06) 0.02 (0.05) 0.03 (0.07) 0.06 (0.11) 8020 0.01 (0.04) 0.02 (0.07) 0.03 (0.10) 0.06 (0.15) 80

RT correct detections8 1.14 (0.49) 1.34 (0.34) 1.71 (0.39) 2.05 (0.56) 8012 0.97 (0.26) 1.27 (0.30) 1.63 (0.31) 1.93 (0.31) 7916 0.98 (0.34) 1.22 (0.26) 1.46 (0.25) 1.72 (0.33) 8020 0.95 (0.23) 1.28 (0.26) 1.46 (0.25) 1.62 (0.46) 80

RT false alarms8 0.74 (1.64) 0.94 (1.33) 1.33 (1.82) 2.34 (0.70) 8012 0.51 (1.05) 0.79 (1.03) 1.15 (1.16) 1.90 (0.70) 7916 0.35 (0.68) 0.64 (0.79) 1.08 (1.05) 1.96 (0.60) 8020 0.54 (0.77) 0.68 (0.89) 1.03 (0.90) 1.52 (0.58) 80

Note. Standard deviations are in parentheses.

Table 1bMeans and standard deviations for perceived workload.

Event rate(per min)

Displays n

1 2 4 8

Global workload8 23.42 (17.64) 30.07 (18.39) 34.44 (15.60) 52.62 (17.06) 8012 24.62 (17.87) 29.21 (15.96) 36.78 (22.64) 50.53 (17.61) 7916 26.97 (17.77) 30.87 (18.18) 38.89 (15.82) 52.49 (16.26) 8020 30.10 (18.57) 37.47 (17.27) 44.27 (15.37) 56.29 (14.34) 80

Mental demand8 28.03 (26.27) 39.74 (27.19) 48.38 (26.44) 67.21 (28.41) 8012 31.37 (28.14) 37.03 (27.63) 48.10 (28.74) 63.72 (29.62) 7916 34.55 (28.90) 39.01 (27.59) 53.51 (27.16) 72.19 (24.27) 8020 36.96 (28.04) 48.40 (27.46) 57.40 (24.85) 71.45 (24.55) 80

Physical demand8 8.42 (14.75) 12.56 (18.67) 12.92 (17.78) 19.16 (24.99) 8012 10.38 (16.71) 14.39 (19.20) 14.09 (17.46) 21.37 (23.62) 7916 12.22 (15.29) 12.21 (16.71) 17.34 (20.03) 15.41 (21.82) 8020 13.73 (17.17) 15.35 (18.12) 15.76 (18.98) 18.89 (20.31) 80

Temporal demand8 23.75 (25.05) 36.09 (28.87) 44.09 (26.28) 71.43 (25.75) 8012 23.68 (26.82) 29.80 (28.23) 40.57 (29.23) 64.49 (30.87) 7916 30.51 (30.30) 34.68 (29.47) 46.69 (26.68) 69.59 (29.58) 8020 31.89 (27.96) 43.38 (28.57) 51.55 (28.45) 72.11 (26.05) 80

Perceived performance8 24.28 (29.78) 26.41 (29.02) 24.05 (21.31) 41.17 (26.92) 8012 26.20 (32.35) 28.53 (30.73) 30.82 (26.63) 43.16 (26.80) 7916 26.39 (30.24) 25.45 (26.74) 30.00 (23.33) 45.07 (26.15) 8020 30.85 (31.62) 28.93 (24.79) 36.28 (23.04) 50.51 (23.96) 80

Effort8 33.22 (31.21) 39.74 (29.63) 50.78 (29.05) 66.51 (27.13) 8012 35.32 (28.60) 41.60 (27.38) 57.66 (87.51) 60.82 (27.52) 7916 36.96 (31.51) 47.92 (30.40) 55.06 (28.02) 65.68 (26.87) 8020 39.35 (30.69) 52.79 (28.04) 57.43 (28.64) 65.59 (27.18) 80

Frustration8 22.83 (30.80) 25.90 (31.01) 26.43 (29.37) 49.67 (31.92) 8012 20.81 (29.92) 24.13 (28.82) 29.43 (31.09) 44.94 (34.72) 7916 21.22 (27.52) 24.08 (26.70) 33.47 (29.40) 47.74 (31.19) 8020 27.80 (28.80) 36.29 (31.49) 47.70 (32.44) 59.21 (30.81) 80

Note. Standard deviations are in parentheses.

476 J.L. Szalma, G.W.L. Teo / Acta Psychologica 139 (2012) 471–485

items, and6worry items) on a 5-point rating scale. Scores vary from0 to35 for task engagement and distress and 0 to 30 for worry.

2.5. Procedure

Data were collected in groups of up to nine participants in a labo-ratory equipped with nine computer work stations separated by cubi-cle walls to prevent visual contact between participants. In addition,participants wore noise canceling headphones during the experimentto prevent distraction from ambient sound. At no time during thesession did any of the participants communicate with one another.Participants surrendered their cell phones, pagers, and wristwatchesprior to performing the task. Participants completed the pre-task ver-sion of the stress state questionnaire (DSSQ) immediately prior to thefirst task condition. Each condition was 12 min in duration, with thefirst 3 min serving as practice trials to familiarize the participantswith the task. After each display condition participants completedthe TLX and DSSQ, the order of which was counterbalanced acrossparticipants. The duration of the entire study was approximately 2 h.

3. Results

Performance and workload were each analyzed via a 4 (eventrate)×4 (source complexity) mixed analysis of variance (ANOVA)with repeated measures on the second factor. Each of the three DSSQscales were analyzed via an ANCOVA with the corresponding pre-task state entered as a covariate. Due to technical problems someparticipants did not complete several items on the subjective mea-sures and so were not included in the analyses for those variables.Hence, the degrees of freedom are not constant across analyses. Viola-tions of sphericity were corrected using the Greenhouse–Geisser ad-justment of degrees of freedom (Maxwell & Delaney, 2004). As theindependent variables in this studywere quantitative, statistically sig-nificant effects involving event rate were further analyzed by trendanalyses using hierarchical regression (e.g., Pedhazur, 1997), and forsignificant main effects of display the trends were evaluated usingwithin-subjects polynomial contrasts. To facilitate comparison ofpatterns of performance, workload, and stress, all analyses were com-puted using z-scores of each dependent variable (computed foreach dependent measure using the respective mean and standarddeviation for the entire sample) and, when appropriate (i.e., regres-sion analyses of trends) on the independent variable (event rate).The means and standard deviations of the untransformed scoresare summarized in Tables 1a, 1b, and 1c for each dependent measure.The figures displaying the trend analyses consist of the z-scores.

3.1. Performance

3.1.1. Proportion of signals detectedThe proportion of correct detections is plotted as a function of event

rate for the four source complexity conditions in Fig. 5a. ANOVArevealed significant main effects for source complexity, F(3,842)=921.60, pb .001, ω2=.69, and for event rate, F(3,314)=23.39, pb .001,ω2=.05. As expected, in each case higher levels of demandwere associ-ated with fewer correct detections. The source complexity by event rateinteraction was also statistically significant, F(8,842)=11.68, pb .001,ω2=.07. Trend analyses of accuracy as a function of event rate withineach source complexity condition indicated a linear (Lin) decline inthe 1-display condition, F(1,317)=6.66, p=.01, R2=.02, βLin=−.143, t(317)=2.58, p=.01, and curvilinear relationships in the otherthree conditions. Linear and quadratic (Quad) trends were observedfor the 2-display condition, F(2,316)=13.09, pb .001, R2=.07,βLin=−.252, t(316)=4.66, pb .001, βQuad=−.116, t(316)=2.14,p=.03. A cubic relationship was observed for the 4-display condition,F(3,315)=25.63, pb .001, R2=.20, βCubic=−.435, t(315)=2.41,pb .001. Statistically significant quadratic and cubic terms were

Table 1cMeans and standard deviations for stress measures.

Event rate(per min)

Pre-task Displays n

1 2 4 8

Task engagement8 13.40 (3.93) 9.26 (5.15) 10.27 (4.91) 10.81 (4.61) 11.03 (4.56) 7812 12.85 (3.88) 10.82 (4.13) 9.87 (4.71) 9.95 (5.07) 10.41 (4.81) 7916 12.42 (4.54) 10.08 (4.28) 10.58 (4.74) 10.23 (4.32) 11.03 (3.69) 7920 13.47 (3.56) 10.47 (4.30) 10.19 (4.68) 10.00 (5.03) 10.39 (4.86) 74

Distress8 5.99 (3.44) 4.54 (3.59) 5.31 (3.81) 5.81 (3.83) 7.86 (4.28) 7812 5.94 (3.51) 4.90 (3.58) 5.21 (3.66) 6.91 (3.79) 8.49 (3.99) 7916 6.15 (3.33) 4.73 (3.75) 5.75 (3.87) 6.66 (3.42) 9.29 (3.40) 7720 6.30 (3.63) 5.69 (3.98) 6.12 (3.50) 8.22 (3.71) 9.04 (3.66) 74

Worry8 11.74 (4.91) 10.33 (6.81) 9.90 (6.62) 9.22 (6.18) 8.23 (6.68) 7812 11.19 (5.74) 8.78 (5.85) 8.39 (6.61) 8.63 (6.28) 8.20 (6.26) 7916 11.43 (4.53) 10.57 (5.95) 10.64 (5.75) 9.83 (6.39) 7.85 (5.78) 7920 10.68 (5.21) 9.76 (6.05) 9.68 (6.13) 9.00 (6.19) 8.99 (6.33) 74

Note. Standard deviations are in parentheses.

477J.L. Szalma, G.W.L. Teo / Acta Psychologica 139 (2012) 471–485

observed for the 8-display condition, F(3,314)=59.08, pb .001, R2=.36,βQuad=−.240, t(314)=5.32, pb .001, βCubic=−.539, t(314)=3.35,p=.001. For each of the curvilinear relationships, the level of perfor-mance accuracy was relatively stable at lower event rates and declinedat the higher event rates.

3.1.2. Proportion of false alarmsThe proportion of false alarms is plotted as a function of event rate

for the four source complexity conditions in Fig. 5b. ANOVA indicateda significant main effect for source complexity, F(3,551), pb .001,ω2=.19, but not for event rate (p=.26). However, a statisticallysignificant event rate by source complexity interaction was observed,F(5, 551)=5.91, pb .001, ω2=.03. Trend analyses for event ratewithin each display condition revealed that the only statisticallysignificant effects were for the linear and quadratic trends in the 8-display condition, F(2,315)=4.80, p=.009, R2=.03, βLin=−.121,t(315)=2.18, p=.03, βQuad=.121, t(315)=2.18, p=.03. The pat-tern of results indicates that there was little change in false alarmsas a function of event rate when the level of display uncertaintywas low, but at the highest level of structural demand false alarmsdeclined as event rate increased to the mid-range (minimum at~16 events/min), after which false alarms increased.

-2

-1.5

-1

-0.5

0

0.5

1

1.5

2

-2 -1.5 -1 -0.5 0 0.5 1 1.5 2

Pro

po

rtio

n o

f co

rrec

t d

etec

tio

ns

(z-s

core

s)

Event Rate (z-score)

1 display 2 displays4 displays 8 displays1 display 2 displays4 displays 8 displays

A

Fig. 5. Z-scores for the proportion of signals detected (A) and the proportion of false alarmsfitted to the data points. Note. Error bars are standard errors.

3.1.3. Response time on correct detection trialsMedian response time on correct detection trials is plotted as a

function of event rate and source complexity condition in Fig. 6.Statistically significant effects were observed for source complexity,F(2,723)=650.82, pb .001, ω2=.61, event rate, F(3,307)=22.37,pb .001, ω2=.05, and the interaction between these two factors, F(7,723)=7.17, pb .001 ω2=.21. Trend analyses for response timeas a function of event rate within each display condition revealed alinear relationship for the 1-, F(1,314)=10.93, p=.001, R2=.03,βLin=−.183, t(314)=3.31, p=.001, 2-, F(1,314)=7.03, p=.008,R2=.02, βLin=−.087, t(314)=2.65, p=.008, 4-, F(1,316)=45.93,pb .001, R2=.13, βLin=−.244, t(316)=6.77, pb .001, and 8-displayconditions, F(1,313)=65.53, pb .001, R2=.17, βLin=−.380, t(313)=8.10, pb .001. In each case higher event rate predicted faster responsetimes and the linear change was generally larger as a function of in-creasing display uncertainty.

3.1.4. Response time on false alarm trialsMedian response time on false alarm trials is plotted as a function

of event rate and source complexity condition in Figs. 7a and 7b,respectively. Statistically significant effects were observed forsource complexity, F(2, 88)=2.77, p=.044, ω2=.01, and eventrate, F(3,315)=7.11, p=.001, ω2=.03. The interaction between

-2

-1.5

-1

-0.5

0

0.5

1

1.5

2

-2 -1.5 -1 -0.5 0 0.5 1 1.5 2

Pro

po

rtio

n o

f F

alse

Ala

rms

(z-s

core

)

Event Rate (z-score)

1 display 2 displays

4 displays 8 displays

8 displays

B

(B) as a function of display uncertainty and event rate. In each case the functions were

2

Res

po

nse

Tim

e to

Co

rrec

t D

etec

tio

ns

(z-s

core

)

1.5

1

0.5

0

-0.5

-1

-1.5

-2-2 -1.5 -1 -0.5 0 0.5 1 1.5 2

Event Rate (z-score)

8 displays RT = -0.38ER + 0.95

1 display

8 displays

4 displays

2 displays

1 display

4 displays

2 displays

8 displays

4 displays RT = -0.24ER + 0.33

2-displays RT = -0.09ER + 0.29

1 display: RT = -0.14ER + 0.88

Fig. 6. Z-scores for median response time for correct detections as a function of displayuncertainty and event rate. In each case the functions were fitted to the data points.Note. Error bars are standard errors.

478 J.L. Szalma, G.W.L. Teo / Acta Psychologica 139 (2012) 471–485

these factors was not statistically significant (p=.090). Trend analy-sis for event rate indicated significant linear and quadratic terms,F(1,314)=58.71, pb .001, R2=.27, βLin=−.508, t(314)=10.48,pb .001, βQuad=.116, t(314)=2.41, p=.016. For source complexitythe within-subjects polynomial contrasts indicated a marginallysignificant linear trend, F(1,40)=3.63, p=.064, ω2=.01 (see Fig.7B). Response time increased as a function of source complexitybut decreased as a function of event rate.

3.1.5. Speed–accuracy tradeoff: analysis of information throughputTo further explore the pattern of speed–accuracy relationships

as a function of information rate and information structure, therate of accurate information processing (throughput; Thorne, 2006)was computed by dividing the number of correct detections by thecumulative response time (both correct and incorrect responses). AsThorne (2006) noted, this measure may be conceptualized as thenumber of accurate decisions per unit of time available to make a de-cision. In the case of a two-choice task, as in the present study, thismeasure corresponds to bits of information processed per unit time.

ANOVA indicated statistically significant effects for source complex-ity, F(2,627)=942.81, pb .001, ω2=.69, and event rate, F(3,314)=3.80, p=.011, ω2=.01. The interaction between these factors wasnot statistically significant (p=.48). For event rate linear and quadratic

-2

-1.5

-1

-0.5

0

0.5

1

1.5

2

-2 -1 0 1

RT = 0.11ER2 - 0.38ER + 0.811

2

Res

po

nse

Tim

e fo

r Fa

lse

Ala

rms

(z-s

core

s)

Event Rate (z scores)

A

-2

-1.5

-1

-0.5

0

0.5

1

1.5

2

-2

Res

po

nse

Tim

e fo

r Fa

lse

Ala

rms

(z-s

core

s)

B

Fig. 7. Z-scores for median response time for false alarms as a function of event rate (A)Note. Error bars are standard errors.

trends were observed, F(2,316)=5.73, p=.004, R2=.04, βLin=−.138,t(316)=2.49, p=.013, βQuad=−.126, t(316)=2.28, p=.023, reach-ing a maximum at ~16 events/min (see Fig. 8a). Increasing eventrate up to that level increased throughput, with a decline in perfor-mance thereafter. That is, as the rate of information presentationis increased, channel capacity increased (i.e., the rate of informationtransfer without error increased) up to a maximum point. For sourcecomplexity within-subjects polynomial contrasts indicated significantlinear, quadratic, and cubic terms for display, as shown in Fig. 8b. Asinformation uncertainty increased throughput (channel capacity) de-creased, with relative stability in the 4-display range.

3.2. Workload

3.2.1. Global workloadMean global workload scores are plotted as a function of event rate

and display condition in Fig. 9a and b, respectively. Technical problemswith computer data collection resulted in a number of missing valuesfor the weights of the six components of the TLX. Analyses were there-fore computed using the unweighted average and unweighted ratings.A significant main effect on global workload was obtained forsource complexity, F(2,633)=227.28, pb .001, ω2=.35, and eventrate, F(3,274)=2.99, p=.031 ω2=.005. The interaction betweenthese factors was not statistically significant (p=.63). For source com-plexity the within subjects polynomial contrasts indicated statisticallysignificant linear, F(1,274)=380.25, pb .001, ω2=.78, and quadratictrends, F(1,274)=56.83, pb .001, ω2=.35. For event rate a significantlinear trend was observed, F(1,310)=11.43, p=.001, R2=.03,βLin=.189, t(310)=3.38, p=.001.

3.2.2. SubscalesAnalyses of the six components (i.e. Mental Demand, Physical

Demand, Temporal Demand, Performance, Effort, Frustration) of per-ceived workload indicated a significant main effect of source complexi-ty for each scale (Mental Demand: F(2,658)=186.98, pb .001,ω2=.31,Physical Demand: F(2,741)=17.37, pb .001, ω2=.04, Temporal De-mand: F(2,704)=195.62, pb .001, ω2=.32, Performance: F(2,772)=48.51, pb .001, ω2=.10, Effort: F(2,580)=46.04, pb .001, ω2=.10,and Frustration: F(2,753)=85.58, pb .001, ω2=.17). The only sta-tistically significant effects for event rate were associated with Tem-poral Demand, F(3,284)=3.00, p=.031, ω2=.005, and Frustration,F(3,281)=4.70, p=.003, ω2=.009. For each independent variable,when effects were observed they generally conformed to that of global

y = 0.13x + 0.75

-1-1.5 0-0.5 1.510.5 2

Display (z scores)

and source complexity (B). In each case the functions were fitted to the data points.

-1.20

-1.00

-0.80

-0.60

-0.40

-0.20

0.00

0.20

0.40

0.60

0.80

1.00

1.20

-1.5 -1 -0.5 0 0.5 1 1.5

thro

ug

hp

ut

(nu

mb

er o

f co

rrec

t/cu

mu

lati

ve R

T, z

-sco

re)

Event Rate (per minute, z-scores)

A

-1.20

-1.00

-0.80

-0.60

-0.40

-0.20

0.00

0.20

0.40

0.60

0.80

1.00

1.20

-1.5 -1 -0.5 0 0.5 1 1.5

thro

ug

hp

ut

(nu

mb

er o

f co

rrec

t/cu

mu

lati

ve R

T, z

-sco

re)

Number of Displays to be monitored (z-scores)

B

Fig. 8. Throughput (bits/min) as a function of event rate (A) and source complexity (B). In each case the functions were fitted to the data points. Note. Error bars are standard errors.

479J.L. Szalma, G.W.L. Teo / Acta Psychologica 139 (2012) 471–485

workload: Perceived workload increased as a function of task demand.A statistically significant source complexity by event rate interactionwas observed for Physical Demand, F(7,741)=2.04, p=.041,ω2=.007. Trend analyses for the Physical Demand scale as a functionof event rate indicated a significant linear trend in the 1-display condi-tion, F(1,306)=4.68, p=.031, R2=.02, βLin=.129, t(300)=2.27,p=.024. There were no significant trends for the other display condi-tions (p>.11 in each case). The display by event rate interaction wasnot statistically significant for the other five scales (p>.17 in each case).

3.3. Stress

ANOVAs indicated that there were no statistically significantdifferences in pre-task engagement, distress, or worry as a functionof event rate (p>.31 in each case). To evaluate the effects of informa-tion rate and structure on post-task stress an ANCOVA was computedfor each of the three scales, with the corresponding pre-task stateentered as the covariate. Significant source complexity effects wereobserved for perceived task engagement, F(2,811)=2.81, p=.04,ω2=.004 (Fig. 10a), distress, F(2,682)=101.02, pb .001, ω2=.19

-2

-1.5

-1

-0.5

0

0.5

1

1.5

2

-2 -1.5 -1 -0.5 0 0.5 1 1.5 2

GW

L (

z sc

ore

)

Event Rate (z-score)8/min 12/min 16/min 20/min

A

Fig. 9. Global workload (z-scores) as a function of event rate (A) and source complexity (B). In

(Fig. 10b), and worry, F(2,835)=10.06, pb .001, ω2=.02 (Fig. 10c).Post-task distress increased and worry decreased as a function of thenumber of displays to be monitored. A significant source complexityby event rate interaction was observed for perceived task engagement,F(8,811)=2.27, p=.02, ω2=.009. Hierarchical trend analyses of per-ceived task engagement as a function of event rate within each sourcecomplexity condition were computed, with the pre-task state enteredfirst. There was a statistically significant quadratic term for the 1-display condition, F(3,304)=23.80, pb .001, R2=.19, βpre=.427, t(304)=8.24, pb .001, βQuad=−.106, t(304)=2.04, p=.042 (seeFig. 10a). There were no statistically significant trends as function ofevent rate for the other display conditions (p>.16 in each case).Hence, the source complexity by event rate interaction derived fromthe change in self-reported task engagement as a function of increasingevent rate in the 1-display condition, reaching a level of relative stabilityat ~15 events/min. There were no significant interaction effects forthe distress and worry scales (p>.13 in each case). In sum, the subjec-tive stress induced by higher levels of information structure demandwas restricted to the affective component of stress. The motivational(task engagement) and cognitive (worry, e.g., task unrelated thoughts)

-2

-1.5

-1

-0.5

0

0.5

1

1.5

2

-2 -1.5 -1 -0.5 0 0.5 1 1.5 2

GW

L (

z sc

ore

)

Displays (z score)

B

each case the functions were fitted to the data points. Note. Error bars are standard errors.

-2

-1.5

-1

-0.5

0

0.5

1

1.5

2

-2 -1.5 -1 -0.5 0 0.5 1 1.5 2

Po

st-T

ask

En

gag

emen

t (z

-sco

re)

Event Rate (z scores)

Pre Task Engagement1 display2 display4 display8 display1 display

A

-2

-1.5

-1

-0.5

0

0.5

1

1.5

2

-2 -1.5 -1 -0.5 0 0.5 1 1.5 2

Po

st-t

ask

Dis

tres

s ad

just

ed f

or

pre

-sta

te (

z sc

ore

)

Displays (z scores)

Pre-Task Distress

DisplayB

-2

-1.5

-1

-0.5

0

0.5

1

1.5

2

-2 -1.5 -1 -0.5 0 0.5 1 1.5 2

Po

st-t

ask

Wo

rry

adju

sted

(z-

sco

re)

Displays (z scores)

C

Fig. 10. Post-task stress (z-scores of means adjusted for pre-task state) as a function of event rate for post-task engagement (A), and display uncertainty for post-task distress (B), andpost-task worry (C). In each case the functions were fitted to the data points. Note. Error bars are standard errors.

480 J.L. Szalma, G.W.L. Teo / Acta Psychologica 139 (2012) 471–485

dimensions showed a decline in stress as a function of the number ofdisplays to be monitored. There were no significant main effects ofevent rate for the three scales (p>.09 in each case).

3.3.1. Interactions of independent variables with pre-task statesThere were no statistically significant interactions involving pre-

task state for distress or task engagement (p>.07 in each case). A sig-nificant source complexity by event rate by pre-task state interactionwas observed for worry, F(8,835)=2.65, p=.005, ω2=.02, indicat-ing that the relationship of pre-task to post-task worry was moderat-ed by task characteristics. However, separate event rate by pre-taskworry ANCOVAs within each display condition indicated that therewere no statistically significant effects for event rate or for the interac-tion of event rate and pre-task worry (p>.06 in each case). Correla-tional analyses within each event rate condition revealed statisticallysignificant correlations between pre- and post-task worry, and theserelationships were generally greater in magnitude as event rate in-creased (see Table 2). Thus, higher levels of pre-task worry renderedparticipantsmore vulnerable to the cognitive stress effects of informa-tion demand, such that the relationship between pre-post task worrywas larger in magnitude at the three higher event rates relative to thelowest level of temporal demand.

4. Discussion

The purpose for this study was to test elements of the DynamicAdaptability Theory of stress and performance proposed by Hancockand Warm (1989). Specifically, the effects of variation in informationstructure and rate on adaptive response in terms of performance,perceived workload, and stress were investigated. Consistent withhypothesis 1 and research on the inverse relationship between targetprevalence and performance accuracy (Wolfe et al., 2005, 2007), per-formance accuracy generally declined as event rate increased, but forthe three conditions consisting of multiple displays the decline varied

Table 2Correlations between pre- and post-task worry.

Eventrate

Display N

1 2 4 8

8 .12 .32⁎ .25 .38⁎ 7712 .54⁎ .42⁎ .56⁎ .69⁎ 7916 .42⁎ .38⁎ .48⁎ .21 7320 .46⁎ .59⁎ .46⁎ .44⁎ 73

⁎ pb .05 after Bonferroni correction (α=.0125).

in form and occurred only in specific ranges of temporal demand (i.e.,the relationship was curvilinear).

In contrast, linear relationships were observed between responsetime to correct detections and event rate. Higher levels of structuraldemand were associated with steeper slopes for these functions, butresponse time was faster as a function of event rate and slower as afunction of source complexity. Further, the form of the relationshipof each dimension to performance varied as a function of the otherdimension and of the performance measure itself (accuracy vs. re-sponse time). However, perceived workload and stress functionsfor each dimension were similar across levels of the other dimension(i.e., there were no interactive effects). Thus, the functions describingadaptive instability may not have a parallel, nested structure of differ-ent thresholds of precipitous change in adaptation for subjectivestate and task performance. Further, these functions do not exhibitsymmetry across task dimensions, as indicated in the model shownin Fig. 2. In other words, hypothesis 2 was not supported.

One could make the post-hoc argument that the magnitudes ofdemands in this study were restricted to those ranges of informationrate and information structure in which both forms of adaptationexhibited instability (i.e., region AΨ–AP in Fig. 2). This is unlikely,however. The least demanding condition (1 display at 8 events/min)was associated with high performance accuracy (correct detectionsM=95%; false alarms M=1%) and with low workload (M=23.42),and distress (M=4.54), suggesting this condition lay within the com-fort zone shown in Fig. 2 (specifically, region AN–AC). In contrast, themost demanding condition (8 displays at 20 events per minute) wasassociated with very poor performance accuracy (correct detectionsM=28%; false alarms M=6%), higher workload (M=56.29), andgreater distress (M=9.04). In addition, the reduction in accuracyand response time associated with the event rate manipulation isconsistent with prior research on the effect of target prevalence on vi-sual search performance.

Hence, the current results are not likely an artifact of range effects.Instead, it appears that the progressive changes in adaptation inwhich subjective comfort declines at lower levels of demand thanperformance (see Fig. 2) may not be ubiquitously true. The evidencefrom this study does not preclude the possibility that there may betasks in which such progressive functions occur, but it does indicatethat relative positions of the nested functions may depend on thenature of the task itself (i.e., which components of information struc-ture and rate are dominant for a given task). Future research shouldvary the specific nature of the task (e.g., different kinds of informationprocessing tasks, and different manipulations of information rate andstructure) to assess the generality of these results.

481J.L. Szalma, G.W.L. Teo / Acta Psychologica 139 (2012) 471–485

Hypothesis 3, that the information dimensions would not beorthogonal, was confirmed for performance measures, but task vari-ables generally exerted independent effects upon workload andstress. The latter results were consistent with those of Hancock andCaird (1993), who reported no significant interaction of spatialand temporal dimensions of task demand for either the TLX or theSWAT, an alternative measure of perceived workload (Reid & Nygren,1988). Recall that Hancock and Warm (1989) proposed that the set ofenvironmental demands (which they referred to as a ‘stress signature’)could in principle be represented as multivariate vector in which thetwo task dimensions are combined with other environmental stimuli(e.g., noise, temperature) that affect adaptation. It may be that theway in which these variables combine itself varies according to theform of adaptive response measured (i.e., performance vs. subjectivestate). If this proves to be the case, then future refinement of the theoryshould specify conditions under which the dimensions of stress com-bine additively or multiplicatively.

The prediction for hypothesis 4 was that the relation of outcomemeasures (adaptive responses) to task demand would be curvilinear,with a precipitous change at thresholds of adaptation. This predictionwas confirmed only at the most extreme condition (i.e., 8-display)and only for accuracy. At lower levels of information structure eventrate functions were linear (1-display), curvilinear with a decline inaccuracy at the extremes of demand (4-display), or curvilinear witha gradual decline at the highest level of demand (2-display; seeFig. 5a). In addition, the loss of comfort (the transition at point AC inFig. 2) was not precipitous as predicted by the Dynamic AdaptabilityTheory but was more consistent with the gradual (‘graceful’) degra-dation predicted by Hockey (1997).3 However, both perspectives pro-pose that individuals can effectively adapt to increases in demand byallocating more resources to the task. Results indicate that for perfor-mance the strategy of resource allocation varied as a function of thenature of the demand (i.e., structural vs. temporal), which manifestedin the relationships between speed and accuracy across experimentalconditions.

4.1. Speed/accuracy tradeoffs

The relationship between speed and accuracy as a function ofevent rate varied in form across levels of source complexity. Acrossall display conditions response time to correct detections declinedas a linear function of event rate. In the 1-display condition accuracyalso declined linearly as a function of event rate, indicating a strategyof speed over accuracy to compensate for the increased temporal de-mand. This result is similar to those reported previously in the con-texts of visual search tasks (Wolfe et al., 2005). For the 2-displaycondition accuracy remained relatively stable until the highestevent rate, at which point performance declined, revealing a speedaccuracy tradeoff only at the highest levels of temporal demand(i.e., the difference between the 16 and 20 events/min conditions).

For the 4-display condition accuracy declined at the extremes butwas relatively stable in the mid-range of event rate, indicating that astrategy of favoring speed over accuracy impairs the latter only at theextremes of temporal demand (i.e., changes at the lowest and highestrates of presentation). In the 8-display condition accuracy was rela-tively stable across event rates until a precipitous decline occurredat the highest level of demand. Hence, a strategy of favoring speedover accuracy did not incur costs for the latter until a threshold oftemporal demand was reached. Across the range of event rates fasterresponse was related to a progressive decline in accuracy only in the

3 As pointed out by one of the reviewers, it is possible that gradual degradation re-sults from averaging across individuals and that individuals vary in the rate of changeand the location of the AC threshold (and see Szalma, 2008). Although inspection of theindividual data in this study did not reveal such a pattern, it nevertheless remains alogical possibility that warrants further empirical evaluation.

1-display conditions. In the other conditions the decline in accuracyeither did not occur at lower event rates or was more gradual untilthe higher levels of information rate.

With respect to changes in information structure, participantsresponded to increased task load by slowing their responding. Thisstrategy preserved accuracy at lower levels of temporal and structuraldemand, but at the higher levels of demand accuracy declined sub-stantially and response times showed a corresponding linear in-crease. Thus, a strategy shift favoring speed was adopted to copewith increased temporal demand but slower response times werenot effective in preserving accuracy as structural demand increased,and the highest levels on both dimensions were associated with thelargest changes in performance. It is noteworthy that the incrementin response time as a function of display was linear, given that theincrease in scanning requirements were larger between the 4- and8-display conditions than between the 1- and 2- or 2- and 4-displayconditions due to the spatial separation of the 2×2 grids. The dis-tances between displays or stimuli to be inspected are known toinfluence visual search, and it is also an important consideration indisplay design (Wickens & Hollands, 2000). However, the presenceof this physical separation was not sufficient to exert a correspondingnon-linear effect on response time. Although non-linearity was ob-served for performance accuracy, this was not restricted to the 8-display condition and the pattern observed is thus unlikely to bedue only to the physical separation of displays in that condition.

It is somewhat surprising that at the highest level of structural de-mand response time declined linearly as a function of event rate,while accuracy remained relatively stable (although accuracy waslow relative to lower levels of structural demand) until the highestevent rate. This pattern reflects a supra-additive synergistic interac-tion (Hancock & Szalma, 2008) for accuracy, because increases inthese demands combined to induce larger effects on performancethan if they exerted independent effects and were thus simply addi-tive. In contrast, response time was associated with a sub-additiveantagonistic interaction (Hancock & Szalma, 2008), such that increas-ing structural demand increased response time while increasing tem-poral demand decreased response time. Further, the slope of thefunction for each dimension increased as a function of the level of de-mand on the other. This may be an artifact of the specific levels of de-mand tested (i.e., there may be ranges of task load in which thepattern is reversed). However, the pattern may derive from differentstrategies of adaptation to structural vs. temporal demands. Partici-pants adapted to higher temporal demand by favoring a strategy ofspeed, which was relatively effective (in terms of maintaining accura-cy) at lower levels of demand but was less effective at the mostextreme level. The Dynamic Adaptability Theory does not easilyaccommodate such strategic shifts in speed and accuracy. However,the broader energetic resource perspective can account for thesepatterns.

4.2. Data-limited and resource-limited processes

The interactive effects of the task dimensions on speed and accu-racy may be explained in terms of the relationship of task demandsto the availability of processing resources. Norman and Bobrow(1975) identified regions of task demands that they defined asresource- or data-limited. Levels of demand that are sensitive to theallocation of resources (i.e., ranges of task load in which increased re-source allocation results in improved performance) are defined as aresource-limited range. Tasks in which allocation of more resourcesdoes not affect performance (or performance declines in spite of in-creased resource allocation) are defined as being in the data-limitedrange, because performance is determined by the quality of the infor-mation to be processed rather than the allocation of more processingcapacity.

482 J.L. Szalma, G.W.L. Teo / Acta Psychologica 139 (2012) 471–485

Norman and Bobrow (1975) argued that speed–accuracy tradeoffsreflect a resource-limited range, because improvement in accuracyoccurs as a result of more resources being allocated to meet taskdemands, which slows processing time (or, alternatively, resourceallocation is decreased permitting speed to increase but impairing ac-curacy). Cases in which speed and accuracy co-vary positively wereinterpreted as indicative of a data-limited process, because eitherboth accuracy and speed improve without the additional allocationof resources (based on the assumption that greater resource alloca-tion results in longer processing time) or because performance de-clines on both measures regardless of any additional effort allocation.

The results of the present study indicate that the task was in thedata-limited range across levels of source complexity (responsetime increased and accuracy declined), and for event rate the taskwas in the resource-limited range until the highest level of temporaldemand, at which point the task entered the data-limited range.However, the effects of one dimension on response time and accuracydepended on the level of the other dimension, complicating an inter-pretation of each dimension in terms of data-limited vs. resource-limited processing. The pattern of effects is clearer when one con-siders the throughput measure (i.e., bits of information accuratelyprocessed per unit time; Thorne, 2006).

The analysis of the throughput measure revealed independent ef-fects of changes in structural and temporal demand. Increased eventrate increased bits per minute processed up to 16 events/min, afterwhich throughput declined (see Fig. 8b). This suggests that thelower event rates were in the resource-limited range, as investmentof more capacity improved throughput. However, at the highestlevel of event rate a decline in throughput was observed, indicatingthat the task had reached the data-limited range. In contrast, theincrease in source complexity (display uncertainty; see Fig. 8a) wasrelated to a decline in throughput, indicating a data-limited rangeacross the levels of source complexity tested. This suggests that theslower response time as a function of display was not the result ofgreater resource investment, but rather the data-limited constraintthat scanning takes longer as the angle of displays to be inspected in-creases. This was maximized in the 8-display condition, in partbecause of the aforementioned spatial separation of the two grids tobe monitored. That display uncertainty was in the data-limitedrange was supported by comparison of the performance and theself-report data, which indicated that effort increased as a functionof the number of displays to be monitored, but that this allocationof effort did not improve performance.

4.3. Performance–workload relationships

The range of stability in performer response can manifest inthe pattern of performance–workload relationships (Oron-Gilad etal., 2008; Szalma, in press). In the present study global workloadincreased as a function of increases in structural and temporal de-mands, but the effects of the two task dimensions were independentof one another. In contrast, performance changed as a function ofinteractive effects. However, performance–workload relationshipscan be evaluated in these circumstances because for each dimensionchange in workload was equivalent across levels of the other dimen-sion (e.g., at each level of complexity there was an increase inworkload as a function of event rate). For instance, for event rate aperformance–workload association was observed for accuracy at thehigher levels of demand and a workload insensitivity at lower levelsof demand (i.e., performance was relatively stable but workload in-creased). In contrast, a performance–workload dissociation patternwas obtained for response time, indicating improved performanceat the cost of greater perceived workload. Consideration of through-put provides a clearer picture: Performance–workload dissociationoccurred from 8 to 16 events/min, after which association was ob-served. Note that performance–workload associations, performance

insensitivity with increasing workload, and workload insensitivitywith decreasing performance are most likely in the data-limitedrange (greater investment of resources does not improve performance),and dissociations and insensitivities in which workload decreases orperformance improves are in the resource-limited range (c.f., Norman& Bobrow, 1975).

With respect to source complexity, performance accuracy–workloadassociation was observed at higher levels of demand, with perfor-mance insensitivities (with increased workload) at the lower levels.The latter result can occur when the individual maintains performanceat the cost of greater resource investment. For response time perfor-mance–workload association was observed for display across levels oftemporal demand. Further, evaluation of throughput, combining accu-racy and response time, indicated performance–workload associationacross the range of source complexity, supporting the contentionthat the structural dimension in this study was entirely in the data-limited range.

4.4. Implications for theories of sustained attention

Although the present study did not employ a typical vigilance task,the results have implications for two theories of sustained attention,the mindlessness theory (Robertson, Manly, Andrade, Baddeley, &Yiend, 1997) and the resource perspective (Parasuraman, Warm, &Dember, 1987). According to the mindlessness view, decrements insustained attention result from increased automaticity in responseover time, manifesting as ‘mindless’ responding. However, severalstudies (e.g., Grier et al., 2003; Helton & Warm, 2008) have providedevidence that there is little automaticity in vigilance, and that sus-tained attention is associated with effortful, compensatory activity.

The results of the present study are consistent with this latter per-spective because the lowest correct detection scores were associatedwith the condition with the highest cognitive demand but also thehighest perceived workload and the lowest levels of worry (taskunrelated thoughts). This is clearly inconsistent with the mindless-ness view, which would predict low workload and high levels oftask unrelated thoughts. To be sure, the mindlessness perspectiveapplies to vigilance contexts in which response inhibition is requiredfor relatively simple cognitive tasks, and the task demands are pri-marily attentional. In the present study, however, the task requiredboth maintaining attention and more complex cognitive processingto discriminate signal from non-signal. Hence, the data from thisexperiment may be limited in its capacity to test the arguments ofmindlessness theory. To do so would require manipulating the cogni-tive demands of the task.

4.5. Implications for the Dynamic Adaptability Theory

One implication of the present results for the Dynamic Adaptabil-ity Theory is that it is unlikely that the thresholds of adaptive instabil-ity are in the symmetrical hemispherical form shown in Fig. 3, butthat they are instead irregular in shape. A second implication is thatthe zone of psychological adaptation, which has always been inter-preted in terms of ‘performance,’ may need to be divided into tworegions, one relating to speed and the other to accuracy. This is be-cause the form of performance change as a function of informationstructure and information rate depended on the performance mea-sure used. Further, the model needs to account for ranges of informa-tion demand in which performance improves or speed–accuracytradeoffs occur. An alternative to creating two regions of psychologi-cal adaptability is to use derived measures such as throughput toevaluate the stability of performance. However, regardless of howthe outcome measures are represented, a complication for modifica-tion of the model is that the pattern of adaptive response may bedifferent for the two task dimensions, as was the case in this study.

483J.L. Szalma, G.W.L. Teo / Acta Psychologica 139 (2012) 471–485

Although the evidence from this study suggests that some aspectsof the theory may need to be reconsidered (i.e., the nested structureof adaptive failure and the shape of the adaptive functions; the stabil-ity of adaptation; and the relationship between the two task dimen-sions), it is important to note that Hancock and Warm (1989) didnot assert that the symmetry and common forms of the adaptivefunctions shown in Figs. 2 and 3 must exist, that the distances be-tween thresholds must be equivalent, or that the level of adaptationis literally flat (indeed, they noted the potential value of chaos and ca-tastrophe theory representations in modeling local maxima and min-ima in adaptive response). Rather, they viewed the model presentedin Figs. 2 and 3 as an initial step to be followed by modificationsbased on empirical and theoretical developments. The present studyprovides evidence to support such modification, particularly in thepatterns of performance, workload, and stress as a function of infor-mation structure and rate.

In contrast to performance, the pattern of results for perceivedworkload seemed to generally conform to the model, in that as de-mands increased (on both dimensions) perceived workload in-creased. This pattern was also observed for distress, but perceivedtask-engagement showed a relative insensitivity to task demand,even increasing somewhat as a function of event rate at the lowestlevel of structural demand. For the third stress dimension, worry,stress actually decreased as a function of display demand. Across sub-jective measures results suggest that participants responded tohigher demands by maintaining or increasing effort, but at greatersubjective cost. Cases in which performance remains stable or evenincreases would correspond to the region AC–Aψ in Fig. 2. When per-formance is also impaired, the pattern manifests in the region Aψ–AP.

4.5.1. The (dynamic) stability of adaptationThe small but significant linear trends observed in this study at the

lowest levels of information demand suggest that the plateau shownin Fig. 2 may not represent absolute stability, but rather consist ofsmall changes in adaptive response across the range of input stress,with local maxima (perhaps even improvement in performance)and minima and regions of transient or ‘local’ plateaus (e.g., correctdetections in the 4-display condition; see Fig. 5a) driving the specificadaptive state of a person at a particular level of task demand. The re-sults also suggest that even in these plateau regions there may begradual declines in adaptation (c.f., Hockey, 1997, 2003). For instance,even at the lowest levels of demand (1 display, 8–12 events/min)there was a decline in performance accuracy and an increase in work-load (see Fig. 9a), indicating small but statistically reliable changes inadaptive response.

Recall that the dynamic adaptability theory identifies regions ofhypo and hyper stress. In their original conception, Hancockand Warm (1989) identified vigilance as an instance of the former.However, a large body of work has shown that vigilance imposeshigh workload and is stressful (Warm, Dember, & Hancock, 1996),which indicates that in many instances vigilance tasks (and perhapssignal detection in general) reside in the hyper stress region.Although it may be possible that there are tasks in which vigilanceand signal detection are in the hypostress regions (e.g., a low eventrate, single display) the present results indicated low workload/stressand high performance under these conditions. However, this studyemployed rather short task durations (12 min) and a cognitivelydemanding discrimination, and it is not clear this pattern of resultswould extend to longer task durations or to easier signal/non-signaldiscriminations. Future theoretical and empirical efforts shoulddistinguish between the demands for cognitive processing stimuliversus to requirement to maintain attention. That is, potential differ-ences in the relationships of task dimensions to the discriminationversus the sustained attention requirements of the task should beevaluated. These different forms of demand may exhibit distinctranges of ‘hypo’ and ‘hyper’ stress.

Note that Hancock andWarm (1989) conceived of hyper and hypostress as task demands, i.e., as deterministic task loads, rather than theadaptive response of the individual. Thus, the high workload andstress associated with vigilance reflects compensatory response bythe performer to deal either with overload (as in the case of an ex-tremely high event rate; e.g., see Helton & Warm, 2008) or underload(e.g., the finding that even low event rates or single displays are asso-ciated with perceived workload scores in the middle of the scale;Galinsky, Dember, & Warm, 1989; cited in Warm et al., 1996; Grubbet al., 1995). Although the task load and the adaptive response areconceptually distinct, a problem for Dynamic Adaptability Theoryand stress research generally is to empirically distinguish these con-structs. Such distinctions may avoid circularity in defining hypo/hyper stress (i.e., defining a strong compensatory response as hyperstress and a weak response – e.g., low perceived workload – ashypo stress), but a unit of measure of the environmental demandsis necessary to achieve clear empirical distinction of input and adap-tive response. Indeed, it may be that ranges of hypo or hyper stressare specifiable only in the context of cognitive appraisals by the per-former (Lazarus, 1999; Matthews et al., 2002). The transactionalapproach to stress (Lazarus, 1999; Matthews, 2001) argues that theunit of analysis should be the transaction between the input andresponse, but to date this has not been articulated with sufficientprecision to identify the relations at the ‘microcognitive’ (Crandall,Klein, & Hoffman, 2006) level of analysis.

4.5.2. Task dimensionsAccording to DAT the two information dimensions should have in-

teractive effects. However, in this study interactions were observedfor performance but not for subjective state. These results raise thepossibility that whether the information dimensions interact to influ-ence adaptive response may depend on the specific range of informa-tion demand and the form of adaptive response. Future researchshould examine the relationships between these dimensions usingthese manipulations with other kinds of information processingtasks and using manipulations other than display uncertainty andevent rate.

A broader issue that Hancock andWarm (1989) themselves raisedis the quantification of information structure and rate. In this studythese dimensions were operationalized as display uncertainty andevent rate, respectively. Hancock and Caird (1993) manipulatedthese dimensions in terms of distance to target goal and time avail-able for action. The relationship between the two dimensions maydepend in part on how they are quantified. For instance, one can ma-nipulate information structure by varying spatial uncertainty, thenumber of steps required to complete a task, or the working memorycapacity required. Information rate may be manipulated by varyingstimulus duration or temporal uncertainty (for reviews see Davies &Parasuraman, 1982; Warm & Jerison, 1984).

However, quantifying information in complex tasks with multiplecomponents of structure and rate may be difficult. It is theoreticallypossible to combine multiple forms of information structure or infor-mation rate, and to combine these two dimensions using the vectorapproach advocated by Hancock and Warm (1989), but it is unclearhow the multiple variables should be combined (i.e., additivelyor multiplicatively). Efforts to develop a vector model would be facil-itated by the development of a common ‘adaptation scale’ for com-parison of different forms of response (e.g., correct detections vs.perceived workload). Nevertheless, accurate modeling of the effectsof multiple stressors is a problem for all theories of stress and perfor-mance (Hancock& Szalma, 2008), and the quantification of informationprocessing is a challenge for most cognitive theories in psychology(McBride & Schmorrow, 2005). Development of an approach to com-bining these elements (e.g., a vector representation) would be asubstantial theoretical advance for theories of stress and performance.

484 J.L. Szalma, G.W.L. Teo / Acta Psychologica 139 (2012) 471–485

5. Summary and conclusions

The present study tested elements of the dynamic adaptabilitytheory of Hancock andWarm (1989). Results were broadly consistentwith the model, but as Hancock and Warm (1989) themselves noted,structural modification of the model may be necessary. These includerecognition that the widths of the zones of adaptability may varyacross contexts (even approaching zero in some cases), and that thezone of psychological adaptability needs to be modified to account fordifferences across performance measures (i.e., accuracy, responsetime) and to also account for increased performance quality. However,the greatest challenges to this theory, and perhaps to all cognitiveand energetic theories of performance, are the quantification ofattentional resources and of information processing (c.f., McBride& Schmorrow, 2005). The utility of the Dynamic Adaptability Theorywill be constrained by whether valid measures of informationstructure and rate can be derived, and whether the precision of theresource metaphor can be improved (Hancock & Szalma, 2007;Szalma & Hancock, 2007). This may prove to be particularly difficult inperformance settings in which there are multiple components consti-tuting each dimension. Whether the vector representation advocatedby Hancock and Warm (1989) can adequately address this issue isa matter for further empirical research and theoretical refinement.

References

Benjamins, J. S., Hooge, I. T. C., van Elst, J. C., Wertheim, A. H., & Verstraten, F. A. J.(2009). Search time critically depends on irrelevant subset size in visual search.Vision Research, 49, 398–406.

Cameron, E. L., Tai, J. C., Eckstein, M. P., & Carrasco, M. (2004). Signal detection theoryapplied to three visual search tasks—Identification, yes/no detection and localiza-tion. Spatial Vision, 17, 295–325.

Conway, G., Szalma, J. L., & Hancock, P. A. (2007). A quantitative meta-analytic exami-nation of whole-body vibration effects on human performance. Ergonomics, 50,228–245.

Crandall, B., Klein, G., & Hoffman, R. R. (2006). Working minds: A practitioner's guide tocognitive task analysis. Cambridge, MA: MIT Press.

Davies, D. R., & Parasuraman, R. (1982). The psychology of vigilance. London:Academic Press.Galinsky, T. L., Dember, W. N., & Warm, J. S. (1989, March). Effects of event rate on sub-

jective workload in vigilance performance. Paper presented at the Southern Societyfor Philosophy and Psychology.

Grier, R. R., Warm, J. S., Dember, W. N., Matthews, G., Galinsky, T. L., Szalma, J. L., et al.(2003). The vigilance decrement reflects limitations in effortful attention, notmindlessness. Human Factors, 45, 349–359.

Grubb, P. L., Warm, J. S., Dember, W. N., & Berch, D. B. (1995). Effects of multiple signaldiscrimination on vigilance performance and perceived workload. Proceedings ofthe Human Factors and Ergonomics Society, 39, 1360–1364.

Hancock, P. A. (1996). Effects of control order, augmented feedback, input device and prac-tice on tracking performance and perceived workload. Ergonomics, 39, 1146–1162.

Hancock, P. A., & Caird, J. K. (1993). Experimental evaluation of a model of mentalworkload. Human Factors, 35, 413–429.

Hancock, P. A., Ross, J. M., & Szalma, J. L. (2007). A meta-analysis of performanceresponse under thermal stressors. Human Factors, 49, 851–877.

Hancock, P. A., & Szalma, J. L. (2007). Stress and neuroergonomics. In R. Parasuraman, &M. Rizzo (Eds.), Neuroergonomics: The brain at work (pp. 195–206). Oxford: OxfordUniversity Press.

Hancock, P. A., & Szalma, J. L. (2008). Stress and performance. In P. A. Hancock, & J. L.Szalma (Eds.), Performance under stress (pp. 1–18). Hampshire, UK: Ashgate.

Hancock, P. A., &Warm, J. S. (1989). A dynamic model of stress and sustained attention.Human Factors, 31, 519–537.

Harris, W. C., Hancock, P. A., & Harris, S. C. (2005). Information processing changesfollowing extended stress. Military Psychology, 17, 15–128.

Hart, S. G., & Staveland, L. E. (1988). Development of NASA-TLX (Task Load Index):Results of empirical and theoretical research. In P. A. Hancock, & M. Najmedin(Eds.), Human mental workload (pp. 139–183). Oxford, England: North-Holland.

Hebb, D. O. (1955). Drives and the C.N.S. (conceptual nervous system). PsychologicalReview, 62, 243–254.

Helton, W. S., Matthews, G., & Warm, J. S. (2010). Stress state mediation between envi-ronmental variables and performance: The case of noise and vigilance. Acta Psycho-logica, 130, 204–213.

Helton, W. S., & Warm, J. S. (2008). Signal salience and the mindlessness theory ofvigilance. Acta Psychologica, 129, 18–25.

Hockey, R. (1984). Varieties of attentional state: The effects of environment. In R.Parasuraman, & D. R. Davies (Eds.), Varieties of attention (pp. 449–483). NewYork: Academic Press.

Hockey, G. R. J. (1997). Compensatory control in the regulation of human performanceunder stress and high workload: A cognitive–energetical framework. BiologicalPsychology, 45, 73–93.

Hockey, G. R. J. (2003). Operator functional state as a framework for the assessment ofperformance. In A. W. K. Gaillard, G. R. J. Hockey, & O. Burov (Eds.), Operator func-tional state: The assessment and prediction of human performance degradation incomplex tasks (pp. 8–23). Amsterdam: IOS Press.

Hockey, G. R. J., Gaillard, A. W. K., & Coles, M. G. H. (Eds.). (1986). Energetic aspects ofhuman information processing. The Netherlands: Nijhoff.

Hockey, R., & Hamilton, P. (1983). The cognitive patterning of stress states. In G. R. J.Hockey (Ed.), Stress and fatigue in human performance (pp. 331–362). Chichester:Wiley.

Jerison, H. J., & Pickett, R. M. (1964). Vigilance: The importance of the elicited observingprocesses in vigilance. Science, 143, 970–971.

Lamiell, J. T. (2003). Beyond individual and group differences: Human individuality, scien-tific psychology, and William Stern's Critical Personalism. Thousand Oaks, CA: Sage.

Lanzetta, T. M., Dember, W. N., Warm, J. S., & Berch, D. B. (1987). Effects of task type andstimulus homogeneity on the event rate function in sustained attention. HumanFactors, 29, 625–633.

Lazarus, R. S. (1999). Stress and emotion: A new synthesis. New York: Springer.Lazarus, R. S., & Folkman, S. (1984). Stress, appraisal, and coping. New York: Springer

Verlag.Loeb, M., & Binford, J. R. (1968). Variation in performance on auditory and visual

monitoring tasks as a function of signal and stimulus frequencies. Perception & Psy-chophysics, 4, 361–367.

Matthews, G. (2001). Levels of transaction: A cognitive science framework for operatorstress. In P. A. Hancock, & P. A. Desmond (Eds.), Stress, workload, and fatigue(pp. 5–33). Mahwah, NJ: Erlbaum.

Matthews, G., Campbell, S. E., Falconer, S., Joyner, L. A., Huggins, J., Gilliland, K., et al.(2002). Fundamental dimensions of subjective state in performance settings:Task engagement, distress, and worry. Emotion, 2, 315–340.

Matthews, G., Joyner, L., Gilliland, K., Campbell, S., Falconer, S., & Huggins, J. (1999).Validation of a comprehensive stress state questionnaire: Towards a state ‘bigthree’? In I. J. Deary, I. Mervielde, F. De Fruyt, & F. Ostendorf (Eds.), PersonalityPsychology in Europe, 7. (pp. 335–350)Tilburg, the Netherlands: Tilburg UniversityPress.

Maxwell, S. E., & Delaney, H. D. (2004). Designing experiments and analyzing data: Amodel comparison perspective (2nd ed). Mahwah, NJ: Erlbaum.

McBride, D. K., & Schmorrow, D. (Eds.). (2005). Quantifying human information processing.Lanham, MD: Lexington Books.

Norman, D., & Bobrow, D. (1975). On data-limited and resource-limited processing.Journal of Cognitive Psychology, 7, 44–60.

Oron-Gilad, T., Szalma, J. L., Stafford, S. C., & Hancock, P. A. (2008). The workload and per-formance relationship in the real world: A study of police officers in a field shootingexercise. International Journal of Occupational Safety and Ergonomics, 14, 119–131.

Parasuraman, R. (1979). Memory load and event rate control sensitivity decrements insustained attention. Science, 205, 924–927.

Parasuraman, R., & Hancock, P. A. (2001). Adaptive control of mental workload. In P. A.Hancock, & P. A. Desmond (Eds.), Stress, workload, and fatigue (pp. 305–320).Mahwah, NJ: Erlbaum.

Parasuraman, R., Warm, J. S., & Dember, W. N. (1987). Vigilance: Taxonomy and utility.In J. S. Warm, L. S. Mark, & R. L. Houston (Eds.), Ergonomics and human factors:Recent research (pp. 11–32). New York: Springer-Verlag.

Pavlovskaya, M., Ring, H., Groswasser, Z., Keren, O., & Hochstein, S. (2001). Visualsearch in peripheral vision: Learning effects and set-size dependence. SpatialVision, 14, 151–173.

Pedhazur, E. J. (1997). Multiple regression in behavioral research (3rd ed). HarcourtBrace: Orlando, FL.

Reid, G. B., & Nygren, T. E. (1988). The subjective workload assessment technique: A scal-ing procedure for the measuring of mental workload. In P. A. Hancock, & N. Meshkati(Eds.), Human mental workload (pp. 185–218). Amsterdam: North-Holland.

Robertson, I. H., Manly, T., Andrade, J., Baddeley, B. T., & Yiend, J. (1997). “Oops!”:Performance correlates of everyday attentional failures in traumatic brain injuredand normal subjects. Neuropsychologia, 35, 747–758.

Selye, H. (1976). The stress of life (Revised edition). New York: McGraw-Hill.Szalma, J. L. (2008). Individual differences in stress reaction. In P. A. Hancock, & J. L.

Szalma (Eds.), Performance under stress (pp. 323–357). Aldershot, Hampshire,UK: Ashgate.

Szalma, J.L. (in press). Individual differences in stress, fatigue, and performance. In: P.A.Desmond, G. Matthews, and P.A. Hancock (Eds.), The handbook of operator fatigue.Aldershot, Hampshire, UK: Ashgate.

Szalma, J. L., & Hancock, P. A. (2007). Task loading and stress in human–computerinteraction: Theoretical frameworks and mitigation strategies. In J. A. Jacko, & A.Sears (Eds.), Handbook for human–computer interaction in interactive systems(pp. 115–132). (2nd Ed.). New York: Erlbaum Taylor & Francis Group.

Szalma, J. L., & Hancock, P. A. (2011). Noise effects on human performance: A meta-analytic synthesis. Psychological Bulletin, 137, 682–707.

Taub, H. A., & Osborne, H. (1968). Effects of signal and stimulus rates on vigilance per-formance. The Journal of Applied Psychology, 52, 133–138.

Thorne, D. R. (2006). Throughput: A simple performance index with desirable charac-teristics. Behavior Research Methods, 38, 569–573.

Treisman, A. M., & Gelade, G. (1980). A feature-integration theory of attention. Cogni-tive Psychology, 12, 97–136.

Warm, J. S., Dember, W. N., & Hancock, P. A. (1996). Vigilance and workload in auto-mated systems. In R. Parasuraman, & M. Mouloua (Eds.), Automation and humanperformance: Theory and applications (pp. 183–200). Hillsdale, NJ: Erlbaum.

Warm, J. S., Howe, S. R., Fishbein, H. D., Dember, W. N., & Sprague, R. L. (1984). Cogni-tive demand and the vigilance decrement. In A. Mital (Ed.), Trends in ergonomic-s/human factors I (pp. 15–20). North Holland: Elsevier.

485J.L. Szalma, G.W.L. Teo / Acta Psychologica 139 (2012) 471–485

Warm, J. S., & Jerison, H. J. (1984). The psychophysics of vigilance. In J. S. Warm (Ed.),Sustained attention in human performance (pp. 15–59). Chichester, United King-dom: Wiley.

Wickens, C. D., & Hollands, J. G. (2000). Engineering psychology and human performance(3rd ed). Upper Saddle River, NJ: Prentice-Hall.

Wolfe, J. M. (1994). Guided search 2.0: A revised model of visual search. PsychonomicBulletin & Review, 1, 202–238.

Wolfe, J. M., Horowitz, T. S., & Kenner, N. M. (2005). Rare items often missed in visualsearches. Nature, 435, 439–440.

Wolfe, J. M., Horowitz, T. S., Van Wert, M. J., Kenner, N. M., Place, S. S., & Kibbi, N.(2007). Low target prevalence is a stubborn source of errors in visual searchtasks. Journal of Experimental Psychology. General, 136, 623–638.

Yeh, Y., & Wickens, C. D. (1988). Dissociation of performance and subjective measuresof workload. Human Factors, 30, 111–120.