HISTORY AND MULTIPLE RESOURCEMODEL DEVELOPMENT

Multitasking is prevalent in our society Issuessuch as the dangers of using cell phones whiledriving call for understanding the extent to whichsuch dual-task performance will lead to decreasesin time-sharing ability Multiple resource theory isone approach to this understanding The conceptof multiple resources in attention was spawnedin my mind from two seeds First Kahnemanrsquos(1973) influential book on attention inspired meas a graduate student with concise theory-basedwriting and a model in which human performanceis supported by a general pool of mental ldquoeffortrdquoor undifferentiated resources (although this modelwas actually proposed earlier in a short chapter byMoray 1967 see also Kalsbeek amp Sykes 1967)The concept of graduated effort stood in markedcontrast to the then existing all-or-none single-channel bottleneck view of attention (Broadbent1971 Welford 1967) Kahnemanrsquos model em-phasized the demand of task for these limitedresources the lack of availability of resources for

concurrent tasks and the suffering of performanceof the latter as a consequence However in thefinal chapter Kahneman makes note of the othersources of ldquostructural interferencerdquo which couldnot be accounted for by a pure resource demandor ldquoundifferentiated capacityrdquo model

Second there was by this time a growing bodyof multitasking studies some in the experimentalliterature (eg Bahrick Noble amp Fitts 1954Bahrick amp Shelly 1958 Briggs Peters amp Fisher1972) and during the 1960s These contributed tothe creation of the study of ldquodivided attentionrdquo inperformance as a discipline Two such studiesexplicitly cast their results within a framework ofmultiple resources postulating that all tasks didnot compete for a single undifferentiated ldquopoolrdquo ofdemand-sensitive resources (Kantowitz amp Knight1976 Wickens 1976)

Shortly after this and stimulated by the paral-lel work of North and Gopher (1976 North 1977)I embarked on a series of studies to examine thecosts and benefits of the newly emerging technol-ogy of voice recognition and synthesis particu-larly as applied within the multitask environment

Multiple Resources and Mental Workload

Christopher D Wickens University of Illinois Champaign Illinois

Objective The objective is to lay out the rationale for multiple resource theory andthe particular 4-D multiple resource model as well as to show how the model is use-ful both as a design tool and as a means of predicting multitask workload overloadBackground I describe the discoveries and developments regarding multiple resourcetheory that have emerged over the past 50 years that contribute to performance andworkload prediction Method The article presents a history of the multiple resourceconcept a computational version of the multiple resource model applied to multitaskdriving simulation data and the relation of multiple resources to workload ResultsResearch revealed the importance of the four dimensions in accounting for task inter-ference and the association of resources with brain structure Multiple resource mod-els yielded high correlations between model predictions and data Lower correlationsalso identified the existence of additional resources Conclusion The model was shownto be partially relevant to the concept of mental workload with greatest relevance toperformance breakdowns related to dual-task overload Future challenges are identi-fied Application The most important application of the multiple resource model isto recommend design changes when conditions of multitask resource overload exist

Address correspondence to Christopher D Wickens Alion Science Corporation Micro Analysis and Design 4949 Pearl EastCircle Suite 300 Boulder CO 80301 cwickensalionsciencecom HUMAN FACTORS Vol 50 No 3 June 2008 pp 449ndash455DOI 101518001872008X288394 Copyright copy 2008 Human Factors and Ergonomics Society All rights reserved

GOLDEN ANNIVERSARY SPECIAL ISSUE

450 June 2008 ndash Human Factors

of the aircraft cockpit In interpreting the resultsof studies in this area along with the collectiveimplications of the growing body of multitaskstudies referred to earlier I created for a chapterin Attention amp Performance VIII a sort of meta-analysis In this analysis I tried to account for thevariance in time-sharing efficiency revealed acrossover 50 different studies by two characteristics(a) the extent to which time-shared tasks used thesame versus different processing structures and(b) the extent to which ldquodifficulty insensitivityrdquowas expressed when the two tasks used differentstructures Difficulty insensitivity occurs when anincrease in the difficulty of one task fails to de-grade the performance of a concurrent one

Out of these two analyses emerged a fairly co-herent picture that ldquodefinedrdquo separate resourcesin terms of a set of three dichotomies of infor-mation processing now quite familiar to manyreaders and expressed (because of their threedimensions) in the ldquocuberdquo shown in Figure 1

bull The stages of processing dimension indicates thatperceptual and cognitive (eg working memory)tasks use different resources from those underlyingthe selection and execution of action (Isreal Ches-ney Wickens amp Donchin 1980)

bull The codes of processing dimension indicates thatspatial activity uses different resources than doesverballinguistic activity a dichotomy expressed inperception working memory (Baddeley 1986) andaction (eg speech vs manual control Liu amp Wick-ens 1992 Wickens amp Liu 1988)

bull The modalities dimension (nested within perceptionand not manifest within cognition or response) in-dicates that auditory perception uses different re-sources than does visual perception

Thus to the extent that two tasks use differentlevels along each of the three dimensions time-sharing will be better Note that this assertion doesnot imply that perfect time-sharing will emergewhenever different resources are used for twotasks For example time-sharing an auditory andvisual task will still compete for common percep-tual resources (and may also compete for commoncode-defined resources if say both are linguisticinvolving speech perception and reading)

bull To these three dimensions was later added a fourthvisual channels distinguishing between focal andambient vision (Leibowitz amp Post 1982 Previc1998) a nested dimension within visual resourcesFocal vision primarily (but not exclusively) fovealsupports object recognition and in particular highacuity perception such as that involved in readingtext and recognizing symbols Ambient vision dis-tributed across the entire visual field and (unlikefocal vision) preserving its competency in peripheralvision is responsible for perception of orientation andmovement for tasks such as those supporting walk-ing upright in targeted directions or lane keeping onthe highway (Horrey Wickens amp Consalus 2006)

Concurrent with the identification of the firstthree dimensions Navon and Gopher (1979) de-veloped an elegant mathematical theory of multi-ple resources that predicts the consequences to

Figure 1 The 4-D multiple resource model

MULTIPLE RESOURCES AND MENTAL WORKLOAD 451

concurrent task performance of the resource allo-cation policy between the tasks their demand forlimited resources (task difficulty) and their de-mand for common versus distinct resources (egresource overlap) After these authors integratedtheir theory with my postulation of the identity ofthe three (later four) specific resource dimensionsthis one particular version of the multiple resourcemodel emerged (Wickens 1980 1984)

The rationale for defining these four dimensionsis based strongly on the confluence and joint sat-isfaction of two criteria First the four dimensionsdefining the model should have neurophysiolog-ical plausibility in the sense that the dichotomieshave parallels in brain anatomy In the 3-D + 1model they certainly do (a) Perceptual-cognitiveactivity is associated with neural activity posteriorto the central sulcis of the brain whereas motorand action-oriented activity is anterior (b) Awell-established line of research associates the pro-cessing of spatial and verbal material respectivelywith the right and left cerebral hemispheres ofmost individuals Recently work by Just and col-leagues (Just et al 2001 Just Carpenter amp Miyaki2003) has directly associated activity in these areasas assessed by functional magnetic resonanceimaging (fMRI) analysis with dual-task process-ing and resource demand (c) Auditory and visualprocesses are distinctly associated with auditoryand visual cortices respectively (d) Focal and am-bient vision are supported by ventral and dorsalvisual pathways respectively (Previc 1998)

The second criterion emerged from my humanfactors orientation I felt it important that thedimensions of the model coincide with relativelystraightforward decisions that a designer couldmake in configuring a task or work space to sup-port multitask activities Should one use keyboardor voice Spoken words tones or text Graphs ordigits Can one ask people to control while en-gaged in visual search or memory rehearsal Thesetwo criteria neurophysiological plausibility anddesign decisions appeared to be fairly well sat-isfied in the proposed cube model Furthermorewith a few qualifications described at the end ofthis article the model has appeared to stand the testof time in its ability to account for three decadesof dual-task research and to support design deci-sions (see Wickens 2002 for a recent summary)

Alternative multiple resource models have beenproposed Two have been grounded most promi-nently in the hemispheric laterality framework

described earlier Polson and Friedman (1988)focused exclusively on this dimension whereasBoles (Boles 2002 Boles Bursk Phillips ampPerdelwitz 2007 Boles amp Law 1998) using fac-tor analysis has further differentiated auditory andvisual processing within each hemisphere intosubprocesses such as spatial positional spatialquantitative auditory linguistic and auditoryemotional resources In this manner 14 separateperceptual resources emerge (and 17 overall) Thisstructure also has been evaluated in dual-task par-adigms revealing that task pairs with a greaterdegree of resource overlap suffer greater dual-task decrements With the proliferation of moreresources it becomes more difficult to preciselyassociate each with brain locations (and thereforegain full neurophysiological plausibility Wickens2007) but fMRI technology offers promise thatsuch association might be forthcoming

Both EPIC (Kieras 2007 Meyer amp Kieras1997) and a model of threaded cognition (Salvucciamp Taatgen 2008) invoke multiple resource con-structs within perceptual modalities to account fordual-task interference patterns Finally single-channel bottleneck theory (Pashler 1998) rep-resents a version of multiple resource theory(although not cast in those terms) based primar-ily on the stage dimension Here response selec-tion depends on a very limited supply of resourcesforcing essentially on sequential processing withhigh time demand tasks whereas perceptual-cognitive resources are limited but more availableAs noted by Wickens and Hollands (2000) suchmultiple resource assumptions are quite consistentwith data from the double-stimulation paradigm(Pashler 1989 1998)

A COMPUTATIONAL MODEL

We have recently developed a relatively simplecomputational model of multiple resources fol-lowing a more elaborate version described and val-idated in Sarno and Wickens (1995) This simplemodel (Horrey amp Wickens 2003 Wickens 20022005 Wickens Dixon amp Ambinder 2006) pre-dicts total interference between a time-shared pairof tasks to be the sum of two components a de-mand component (resource demand) and a multi-ple resource conflict component (degree to whichoverlapping resources are required) Each of thesecomponents can range in ordinal values from 0 to4 providing a simple intuitive and bounded scale

452 June 2008 ndash Human Factors

For the demand component each task is specifiedas being automated (D = 0) easy (D = 1) or dif-ficult (D = 2) independently of which resourcesmay be demanded (eg perception vs responseauditory vs visual) Hence the total task demandcan range from 0 (two automated tasks) to 4 (twodifficult tasks)

For the resource conflict component the twotasks are compared in the extent to which theyshare demands on common levels of each of thefour dimensions of the 3-D + 1 multiple resourcemodel 0 1 2 3 or 4 From the sum of these twocomponents a total interference component canbe produced ranging from 0 to 8 This is a simpli-fied version of the kind of computation that goesinto predicting multitask workload in softwarepackages such as IMPRINTcopy (Laughery LeBiereamp Archer 2006) or MIDAS (Booher amp Minninger2003 Gore amp Jarvis 2005) Of course the simpleprediction of the total interference between tasksdoes not inform the modeler as to which task suf-fers when there is interference and hence addeduse must be made of a third resource allocationcomponent of the multiple resource model whichcan express for example the extent to which driv-ing versus cell phone conversation suffers whenthe two compete for multiple resources This issueis discussed later

Although the validation of a more elaborateversion of this computational model is describedin Sarno and Wickens (1995) and in Wickens et al(2006) in studies demonstrating the importanceof invoking multiple rather than single resourcemodels a third validation has been described(Horrey amp Wickens 2003) The experimentaldata reported in Horrey and Wickens (2004) werethose collected in a high-fidelity driving simula-tor as drivers drove on roadways of three differ-ing demands (varying traffic and curvature) whileengaging in concurrent tasks delivered in differ-ent modalities (auditory visual head up and headdown) A slightly modified (but conceptuallyequivalent) version of the multiple resource for-mula was imposed on the task interference datacreated by these nine conditions of the 3 times 3 de-sign for the three tasks of lane keeping respond-ing to unexpected hazards and performing thein-vehicle task For each task the dual-task in-terference score was assessed by subtracting the single-task (baseline) performance measure Theanalysis revealed great success in predicting bothhazard response (98 of variance accounted for

by the model) and in-vehicle task performance(92) However the model did not predict differ-ences in lane-keeping performance well at all fortwo reasons (a) separate ambient and focal visionchannels which were not included in that versionof the model Thus for example it is the ambientvision of the upper visual field that allows thedriver to glance downward and still keep the carheaded forward in the lane center (Horrey et al2006) (b) Drivers protected lane keeping frominterference (treated it as the primary task) henceprotecting it from variance in interference causedby structural differences in task resource competi-tion This clearly demonstrates the resource allo-cation effect

Before I conclude this section on multiple re-source model prediction it is important to empha-size that the primary value of such a model ispredicting relative differences in multitasking be-tween different conditions or interfaces Such amodel is not designed to make absolute predic-tions of performance ndash that is generate a predictedmultitask performance level that can be desig-nated as acceptable or unacceptable This issueas related to workload will be addressed later

MULTIPLE RESOURCES AND WORKLOAD

Multiple resource theory and mental workloadare two related concepts that are often confusedThey overlap but are distinct To distinguish themremember that the multiple resource model archi-tecture consists of three components related todemand resource overlap and allocation policyThe concept of mental workload relates moststrongly to the first of these characterizing thedemand imposed by tasks on the humanrsquos limitedmental resources whether considered as singleor multiple (Moray 1979)

Importantly this demand can be conceptuallyassociated with one of two ldquoregionsrdquo of taskdemand level (Wickens amp Hollands 2000) Thefirst is one in which the demand is less than thecapacity of resources available (ie there is ldquoresid-ual capacityrdquo unused in task performance) Thisis the ideal state because it means that the workerwill have some resources available should unex-pected circumstances be imposed The secondregion is one in which the demand exceeds thecapacity and as with any economic system (Navonamp Gopher 1979) performance will break downSometimes the distinction between these regions

MULTIPLE RESOURCES AND MENTAL WORKLOAD 453

is referred to as a ldquored linerdquo of workload (Grier2008)

Task demand varying along this continuumcan result from either single-task demand (egdriving progressively faster on a winding road im-poses progressively more resource demand untillane keeping begins to fail) or dual-task demandIn characterizing single-task demand the ldquomulti-plerdquo aspect of multiple resource theory has littlerelevance Furthermore whether one is doing oneor two tasks in the ldquoresidual capacityrdquo region mul-tiple resource theory is not fully germane Only inthe region where overload is imposed by multi-ple tasks does multiple resource theory make animportant contribution to mental workload by pre-dicting how much performance will fail once over-load has been reached (ie the size of dual-taskdecrements) Here multiple resource theory can ofcourse also provide guidance as to how redesigncan restore performance to the residual capacityregion

This distinction between regions is critical be-cause most measures of workload are designed asmuch for the residual capacity region as for theoverload region if not more so (Gawron 2008Tsang amp Vidulich 2006) Secondary tasks are ex-plicitly designed to probe ldquoresidual capacityrdquo notused for a primary task physiological and sub-jective measures operate across the entire rangeof both regions (and do not generally distinguishwhich resources within the multiple resourceframework are used) Thus the greatest value ofsuch measures is to ensure that task demands canremain within the residual capacity region Impor-tantly some models such as IMPRINTcopy (Laugheryet al 2006) or MIDAS (Gore amp Jarvis 2005) dohave a mental workload component with taskcompetition based on multiple resource theory andwith workload channels defined to correspond tothe dimensions in Figure 1 However this com-ponent is really intended to predict performancedecrements in the overload region as a result ofmultitask requirements which as we saw repre-sents only one portion of the workload conceptSuch models have been employed in important de-sign decisions regarding workload and crew size(Booher amp Minninger 2003)

FUTURE CHALLENGES FOR MULTIPLE RESOURCES

Although the multiple resource model has beenable to account for a good deal of variance in

time-sharing efficiency when heterogeneoustasks are imposed (Horrey amp Wickens 2003 Liuamp Wickens 1992 Sarno amp Wickens 1995 Wick-ens amp Colcombe 2007 Wickens Goh HellebergHorrey amp Talleur 2003 Wickens Sandry ampVidulich 1983) both the theory and the modelremain challenged by a number of issues Theseinclude the following

bull Adding another level to the modalities dimensionrelated to tactile input (Boles et al 2007)

bull The fact that other mechanisms unrelated to re-sources also account for variance in dual-task per-formance In particular these relate to confusionbetween task elements (eg the problems of time-sharing a perception and working memory task bothusing digits Hirst amp Kalmar 1987) and cooperationbetween elements (eg dual-task processing sup-ported by different dimensions of a single objectDuncan 1979 Wickens amp Hollands 2000)

bull The inability to characterize resource demand on asingle scale on which the ldquored linerdquo distinguishingthe reserve capacity from the overload region can beplaced (Hart amp Wickens 1990 Grier 2008)

bull Understanding what drives the allocation policy Inthe laboratory this often can be driven by primaryand secondary task instructions (Tsang 2006) but inthe real world phenomena such as unwanted diver-sion of attention to interruptions (Trafton amp Monk2008) ldquocognitive tunnelingrdquo (Wickens amp Alexan-der in press) and auditory preemption (Horrey amp Wickens 2004 Wickens amp Colcombe 2007Wickens Dixon amp Seppelt 2005) often operate inways that are clearly at odds with optimal allocationas witnessed by the safety compromise of drivingwhile using a cell phone (Strayer amp Drews 2007)Such findings are consistent with recent develop-ments that allocation policy is a function closely re-lated to that of the central executive a construct wellestablished to account for differences in time-sharingefficiency (Engle 2002 Monsell 2003 Wickens ampMcCarley 2008) as well as to identifiable prefrontalbrain regions (Kramer amp Parasuraman 2007)

Such challenges await the allocation of researchresources to the interested human factors investi-gator

ACKNOWLEDGMENTS

I thank Dr Bill Horrey and Dr David Boles fortheir comments on this manuscript

REFERENCES

Baddeley AD (1986) Working memory Oxford UK ClarendonBahrick H P Noble M amp Fitts P M (1954) Extra task performance

as a measure of learning a primary task Journal of ExperimentalPsychology 48 298ndash302

450 June 2008 ndash Human Factors

of the aircraft cockpit In interpreting the resultsof studies in this area along with the collectiveimplications of the growing body of multitaskstudies referred to earlier I created for a chapterin Attention amp Performance VIII a sort of meta-analysis In this analysis I tried to account for thevariance in time-sharing efficiency revealed acrossover 50 different studies by two characteristics(a) the extent to which time-shared tasks used thesame versus different processing structures and(b) the extent to which ldquodifficulty insensitivityrdquowas expressed when the two tasks used differentstructures Difficulty insensitivity occurs when anincrease in the difficulty of one task fails to de-grade the performance of a concurrent one

Out of these two analyses emerged a fairly co-herent picture that ldquodefinedrdquo separate resourcesin terms of a set of three dichotomies of infor-mation processing now quite familiar to manyreaders and expressed (because of their threedimensions) in the ldquocuberdquo shown in Figure 1

bull The stages of processing dimension indicates thatperceptual and cognitive (eg working memory)tasks use different resources from those underlyingthe selection and execution of action (Isreal Ches-ney Wickens amp Donchin 1980)

bull The codes of processing dimension indicates thatspatial activity uses different resources than doesverballinguistic activity a dichotomy expressed inperception working memory (Baddeley 1986) andaction (eg speech vs manual control Liu amp Wick-ens 1992 Wickens amp Liu 1988)

bull The modalities dimension (nested within perceptionand not manifest within cognition or response) in-dicates that auditory perception uses different re-sources than does visual perception

Thus to the extent that two tasks use differentlevels along each of the three dimensions time-sharing will be better Note that this assertion doesnot imply that perfect time-sharing will emergewhenever different resources are used for twotasks For example time-sharing an auditory andvisual task will still compete for common percep-tual resources (and may also compete for commoncode-defined resources if say both are linguisticinvolving speech perception and reading)

bull To these three dimensions was later added a fourthvisual channels distinguishing between focal andambient vision (Leibowitz amp Post 1982 Previc1998) a nested dimension within visual resourcesFocal vision primarily (but not exclusively) fovealsupports object recognition and in particular highacuity perception such as that involved in readingtext and recognizing symbols Ambient vision dis-tributed across the entire visual field and (unlikefocal vision) preserving its competency in peripheralvision is responsible for perception of orientation andmovement for tasks such as those supporting walk-ing upright in targeted directions or lane keeping onthe highway (Horrey Wickens amp Consalus 2006)

Concurrent with the identification of the firstthree dimensions Navon and Gopher (1979) de-veloped an elegant mathematical theory of multi-ple resources that predicts the consequences to

Figure 1 The 4-D multiple resource model

MULTIPLE RESOURCES AND MENTAL WORKLOAD 451

concurrent task performance of the resource allo-cation policy between the tasks their demand forlimited resources (task difficulty) and their de-mand for common versus distinct resources (egresource overlap) After these authors integratedtheir theory with my postulation of the identity ofthe three (later four) specific resource dimensionsthis one particular version of the multiple resourcemodel emerged (Wickens 1980 1984)

The rationale for defining these four dimensionsis based strongly on the confluence and joint sat-isfaction of two criteria First the four dimensionsdefining the model should have neurophysiolog-ical plausibility in the sense that the dichotomieshave parallels in brain anatomy In the 3-D + 1model they certainly do (a) Perceptual-cognitiveactivity is associated with neural activity posteriorto the central sulcis of the brain whereas motorand action-oriented activity is anterior (b) Awell-established line of research associates the pro-cessing of spatial and verbal material respectivelywith the right and left cerebral hemispheres ofmost individuals Recently work by Just and col-leagues (Just et al 2001 Just Carpenter amp Miyaki2003) has directly associated activity in these areasas assessed by functional magnetic resonanceimaging (fMRI) analysis with dual-task process-ing and resource demand (c) Auditory and visualprocesses are distinctly associated with auditoryand visual cortices respectively (d) Focal and am-bient vision are supported by ventral and dorsalvisual pathways respectively (Previc 1998)

The second criterion emerged from my humanfactors orientation I felt it important that thedimensions of the model coincide with relativelystraightforward decisions that a designer couldmake in configuring a task or work space to sup-port multitask activities Should one use keyboardor voice Spoken words tones or text Graphs ordigits Can one ask people to control while en-gaged in visual search or memory rehearsal Thesetwo criteria neurophysiological plausibility anddesign decisions appeared to be fairly well sat-isfied in the proposed cube model Furthermorewith a few qualifications described at the end ofthis article the model has appeared to stand the testof time in its ability to account for three decadesof dual-task research and to support design deci-sions (see Wickens 2002 for a recent summary)

Alternative multiple resource models have beenproposed Two have been grounded most promi-nently in the hemispheric laterality framework

described earlier Polson and Friedman (1988)focused exclusively on this dimension whereasBoles (Boles 2002 Boles Bursk Phillips ampPerdelwitz 2007 Boles amp Law 1998) using fac-tor analysis has further differentiated auditory andvisual processing within each hemisphere intosubprocesses such as spatial positional spatialquantitative auditory linguistic and auditoryemotional resources In this manner 14 separateperceptual resources emerge (and 17 overall) Thisstructure also has been evaluated in dual-task par-adigms revealing that task pairs with a greaterdegree of resource overlap suffer greater dual-task decrements With the proliferation of moreresources it becomes more difficult to preciselyassociate each with brain locations (and thereforegain full neurophysiological plausibility Wickens2007) but fMRI technology offers promise thatsuch association might be forthcoming

Both EPIC (Kieras 2007 Meyer amp Kieras1997) and a model of threaded cognition (Salvucciamp Taatgen 2008) invoke multiple resource con-structs within perceptual modalities to account fordual-task interference patterns Finally single-channel bottleneck theory (Pashler 1998) rep-resents a version of multiple resource theory(although not cast in those terms) based primar-ily on the stage dimension Here response selec-tion depends on a very limited supply of resourcesforcing essentially on sequential processing withhigh time demand tasks whereas perceptual-cognitive resources are limited but more availableAs noted by Wickens and Hollands (2000) suchmultiple resource assumptions are quite consistentwith data from the double-stimulation paradigm(Pashler 1989 1998)

A COMPUTATIONAL MODEL

We have recently developed a relatively simplecomputational model of multiple resources fol-lowing a more elaborate version described and val-idated in Sarno and Wickens (1995) This simplemodel (Horrey amp Wickens 2003 Wickens 20022005 Wickens Dixon amp Ambinder 2006) pre-dicts total interference between a time-shared pairof tasks to be the sum of two components a de-mand component (resource demand) and a multi-ple resource conflict component (degree to whichoverlapping resources are required) Each of thesecomponents can range in ordinal values from 0 to4 providing a simple intuitive and bounded scale

452 June 2008 ndash Human Factors

For the demand component each task is specifiedas being automated (D = 0) easy (D = 1) or dif-ficult (D = 2) independently of which resourcesmay be demanded (eg perception vs responseauditory vs visual) Hence the total task demandcan range from 0 (two automated tasks) to 4 (twodifficult tasks)

For the resource conflict component the twotasks are compared in the extent to which theyshare demands on common levels of each of thefour dimensions of the 3-D + 1 multiple resourcemodel 0 1 2 3 or 4 From the sum of these twocomponents a total interference component canbe produced ranging from 0 to 8 This is a simpli-fied version of the kind of computation that goesinto predicting multitask workload in softwarepackages such as IMPRINTcopy (Laughery LeBiereamp Archer 2006) or MIDAS (Booher amp Minninger2003 Gore amp Jarvis 2005) Of course the simpleprediction of the total interference between tasksdoes not inform the modeler as to which task suf-fers when there is interference and hence addeduse must be made of a third resource allocationcomponent of the multiple resource model whichcan express for example the extent to which driv-ing versus cell phone conversation suffers whenthe two compete for multiple resources This issueis discussed later

Although the validation of a more elaborateversion of this computational model is describedin Sarno and Wickens (1995) and in Wickens et al(2006) in studies demonstrating the importanceof invoking multiple rather than single resourcemodels a third validation has been described(Horrey amp Wickens 2003) The experimentaldata reported in Horrey and Wickens (2004) werethose collected in a high-fidelity driving simula-tor as drivers drove on roadways of three differ-ing demands (varying traffic and curvature) whileengaging in concurrent tasks delivered in differ-ent modalities (auditory visual head up and headdown) A slightly modified (but conceptuallyequivalent) version of the multiple resource for-mula was imposed on the task interference datacreated by these nine conditions of the 3 times 3 de-sign for the three tasks of lane keeping respond-ing to unexpected hazards and performing thein-vehicle task For each task the dual-task in-terference score was assessed by subtracting the single-task (baseline) performance measure Theanalysis revealed great success in predicting bothhazard response (98 of variance accounted for

by the model) and in-vehicle task performance(92) However the model did not predict differ-ences in lane-keeping performance well at all fortwo reasons (a) separate ambient and focal visionchannels which were not included in that versionof the model Thus for example it is the ambientvision of the upper visual field that allows thedriver to glance downward and still keep the carheaded forward in the lane center (Horrey et al2006) (b) Drivers protected lane keeping frominterference (treated it as the primary task) henceprotecting it from variance in interference causedby structural differences in task resource competi-tion This clearly demonstrates the resource allo-cation effect

Before I conclude this section on multiple re-source model prediction it is important to empha-size that the primary value of such a model ispredicting relative differences in multitasking be-tween different conditions or interfaces Such amodel is not designed to make absolute predic-tions of performance ndash that is generate a predictedmultitask performance level that can be desig-nated as acceptable or unacceptable This issueas related to workload will be addressed later

MULTIPLE RESOURCES AND WORKLOAD

Multiple resource theory and mental workloadare two related concepts that are often confusedThey overlap but are distinct To distinguish themremember that the multiple resource model archi-tecture consists of three components related todemand resource overlap and allocation policyThe concept of mental workload relates moststrongly to the first of these characterizing thedemand imposed by tasks on the humanrsquos limitedmental resources whether considered as singleor multiple (Moray 1979)

Importantly this demand can be conceptuallyassociated with one of two ldquoregionsrdquo of taskdemand level (Wickens amp Hollands 2000) Thefirst is one in which the demand is less than thecapacity of resources available (ie there is ldquoresid-ual capacityrdquo unused in task performance) Thisis the ideal state because it means that the workerwill have some resources available should unex-pected circumstances be imposed The secondregion is one in which the demand exceeds thecapacity and as with any economic system (Navonamp Gopher 1979) performance will break downSometimes the distinction between these regions

MULTIPLE RESOURCES AND MENTAL WORKLOAD 453

is referred to as a ldquored linerdquo of workload (Grier2008)

Task demand varying along this continuumcan result from either single-task demand (egdriving progressively faster on a winding road im-poses progressively more resource demand untillane keeping begins to fail) or dual-task demandIn characterizing single-task demand the ldquomulti-plerdquo aspect of multiple resource theory has littlerelevance Furthermore whether one is doing oneor two tasks in the ldquoresidual capacityrdquo region mul-tiple resource theory is not fully germane Only inthe region where overload is imposed by multi-ple tasks does multiple resource theory make animportant contribution to mental workload by pre-dicting how much performance will fail once over-load has been reached (ie the size of dual-taskdecrements) Here multiple resource theory can ofcourse also provide guidance as to how redesigncan restore performance to the residual capacityregion

This distinction between regions is critical be-cause most measures of workload are designed asmuch for the residual capacity region as for theoverload region if not more so (Gawron 2008Tsang amp Vidulich 2006) Secondary tasks are ex-plicitly designed to probe ldquoresidual capacityrdquo notused for a primary task physiological and sub-jective measures operate across the entire rangeof both regions (and do not generally distinguishwhich resources within the multiple resourceframework are used) Thus the greatest value ofsuch measures is to ensure that task demands canremain within the residual capacity region Impor-tantly some models such as IMPRINTcopy (Laugheryet al 2006) or MIDAS (Gore amp Jarvis 2005) dohave a mental workload component with taskcompetition based on multiple resource theory andwith workload channels defined to correspond tothe dimensions in Figure 1 However this com-ponent is really intended to predict performancedecrements in the overload region as a result ofmultitask requirements which as we saw repre-sents only one portion of the workload conceptSuch models have been employed in important de-sign decisions regarding workload and crew size(Booher amp Minninger 2003)

FUTURE CHALLENGES FOR MULTIPLE RESOURCES

Although the multiple resource model has beenable to account for a good deal of variance in

time-sharing efficiency when heterogeneoustasks are imposed (Horrey amp Wickens 2003 Liuamp Wickens 1992 Sarno amp Wickens 1995 Wick-ens amp Colcombe 2007 Wickens Goh HellebergHorrey amp Talleur 2003 Wickens Sandry ampVidulich 1983) both the theory and the modelremain challenged by a number of issues Theseinclude the following

bull Adding another level to the modalities dimensionrelated to tactile input (Boles et al 2007)

bull The fact that other mechanisms unrelated to re-sources also account for variance in dual-task per-formance In particular these relate to confusionbetween task elements (eg the problems of time-sharing a perception and working memory task bothusing digits Hirst amp Kalmar 1987) and cooperationbetween elements (eg dual-task processing sup-ported by different dimensions of a single objectDuncan 1979 Wickens amp Hollands 2000)

bull The inability to characterize resource demand on asingle scale on which the ldquored linerdquo distinguishingthe reserve capacity from the overload region can beplaced (Hart amp Wickens 1990 Grier 2008)

bull Understanding what drives the allocation policy Inthe laboratory this often can be driven by primaryand secondary task instructions (Tsang 2006) but inthe real world phenomena such as unwanted diver-sion of attention to interruptions (Trafton amp Monk2008) ldquocognitive tunnelingrdquo (Wickens amp Alexan-der in press) and auditory preemption (Horrey amp Wickens 2004 Wickens amp Colcombe 2007Wickens Dixon amp Seppelt 2005) often operate inways that are clearly at odds with optimal allocationas witnessed by the safety compromise of drivingwhile using a cell phone (Strayer amp Drews 2007)Such findings are consistent with recent develop-ments that allocation policy is a function closely re-lated to that of the central executive a construct wellestablished to account for differences in time-sharingefficiency (Engle 2002 Monsell 2003 Wickens ampMcCarley 2008) as well as to identifiable prefrontalbrain regions (Kramer amp Parasuraman 2007)

Such challenges await the allocation of researchresources to the interested human factors investi-gator

ACKNOWLEDGMENTS

I thank Dr Bill Horrey and Dr David Boles fortheir comments on this manuscript

REFERENCES

Baddeley AD (1986) Working memory Oxford UK ClarendonBahrick H P Noble M amp Fitts P M (1954) Extra task performance

as a measure of learning a primary task Journal of ExperimentalPsychology 48 298ndash302

MULTIPLE RESOURCES AND MENTAL WORKLOAD 451

concurrent task performance of the resource allo-cation policy between the tasks their demand forlimited resources (task difficulty) and their de-mand for common versus distinct resources (egresource overlap) After these authors integratedtheir theory with my postulation of the identity ofthe three (later four) specific resource dimensionsthis one particular version of the multiple resourcemodel emerged (Wickens 1980 1984)

The rationale for defining these four dimensionsis based strongly on the confluence and joint sat-isfaction of two criteria First the four dimensionsdefining the model should have neurophysiolog-ical plausibility in the sense that the dichotomieshave parallels in brain anatomy In the 3-D + 1model they certainly do (a) Perceptual-cognitiveactivity is associated with neural activity posteriorto the central sulcis of the brain whereas motorand action-oriented activity is anterior (b) Awell-established line of research associates the pro-cessing of spatial and verbal material respectivelywith the right and left cerebral hemispheres ofmost individuals Recently work by Just and col-leagues (Just et al 2001 Just Carpenter amp Miyaki2003) has directly associated activity in these areasas assessed by functional magnetic resonanceimaging (fMRI) analysis with dual-task process-ing and resource demand (c) Auditory and visualprocesses are distinctly associated with auditoryand visual cortices respectively (d) Focal and am-bient vision are supported by ventral and dorsalvisual pathways respectively (Previc 1998)

The second criterion emerged from my humanfactors orientation I felt it important that thedimensions of the model coincide with relativelystraightforward decisions that a designer couldmake in configuring a task or work space to sup-port multitask activities Should one use keyboardor voice Spoken words tones or text Graphs ordigits Can one ask people to control while en-gaged in visual search or memory rehearsal Thesetwo criteria neurophysiological plausibility anddesign decisions appeared to be fairly well sat-isfied in the proposed cube model Furthermorewith a few qualifications described at the end ofthis article the model has appeared to stand the testof time in its ability to account for three decadesof dual-task research and to support design deci-sions (see Wickens 2002 for a recent summary)

Alternative multiple resource models have beenproposed Two have been grounded most promi-nently in the hemispheric laterality framework

described earlier Polson and Friedman (1988)focused exclusively on this dimension whereasBoles (Boles 2002 Boles Bursk Phillips ampPerdelwitz 2007 Boles amp Law 1998) using fac-tor analysis has further differentiated auditory andvisual processing within each hemisphere intosubprocesses such as spatial positional spatialquantitative auditory linguistic and auditoryemotional resources In this manner 14 separateperceptual resources emerge (and 17 overall) Thisstructure also has been evaluated in dual-task par-adigms revealing that task pairs with a greaterdegree of resource overlap suffer greater dual-task decrements With the proliferation of moreresources it becomes more difficult to preciselyassociate each with brain locations (and thereforegain full neurophysiological plausibility Wickens2007) but fMRI technology offers promise thatsuch association might be forthcoming

Both EPIC (Kieras 2007 Meyer amp Kieras1997) and a model of threaded cognition (Salvucciamp Taatgen 2008) invoke multiple resource con-structs within perceptual modalities to account fordual-task interference patterns Finally single-channel bottleneck theory (Pashler 1998) rep-resents a version of multiple resource theory(although not cast in those terms) based primar-ily on the stage dimension Here response selec-tion depends on a very limited supply of resourcesforcing essentially on sequential processing withhigh time demand tasks whereas perceptual-cognitive resources are limited but more availableAs noted by Wickens and Hollands (2000) suchmultiple resource assumptions are quite consistentwith data from the double-stimulation paradigm(Pashler 1989 1998)

A COMPUTATIONAL MODEL

We have recently developed a relatively simplecomputational model of multiple resources fol-lowing a more elaborate version described and val-idated in Sarno and Wickens (1995) This simplemodel (Horrey amp Wickens 2003 Wickens 20022005 Wickens Dixon amp Ambinder 2006) pre-dicts total interference between a time-shared pairof tasks to be the sum of two components a de-mand component (resource demand) and a multi-ple resource conflict component (degree to whichoverlapping resources are required) Each of thesecomponents can range in ordinal values from 0 to4 providing a simple intuitive and bounded scale

452 June 2008 ndash Human Factors

For the demand component each task is specifiedas being automated (D = 0) easy (D = 1) or dif-ficult (D = 2) independently of which resourcesmay be demanded (eg perception vs responseauditory vs visual) Hence the total task demandcan range from 0 (two automated tasks) to 4 (twodifficult tasks)

For the resource conflict component the twotasks are compared in the extent to which theyshare demands on common levels of each of thefour dimensions of the 3-D + 1 multiple resourcemodel 0 1 2 3 or 4 From the sum of these twocomponents a total interference component canbe produced ranging from 0 to 8 This is a simpli-fied version of the kind of computation that goesinto predicting multitask workload in softwarepackages such as IMPRINTcopy (Laughery LeBiereamp Archer 2006) or MIDAS (Booher amp Minninger2003 Gore amp Jarvis 2005) Of course the simpleprediction of the total interference between tasksdoes not inform the modeler as to which task suf-fers when there is interference and hence addeduse must be made of a third resource allocationcomponent of the multiple resource model whichcan express for example the extent to which driv-ing versus cell phone conversation suffers whenthe two compete for multiple resources This issueis discussed later

Although the validation of a more elaborateversion of this computational model is describedin Sarno and Wickens (1995) and in Wickens et al(2006) in studies demonstrating the importanceof invoking multiple rather than single resourcemodels a third validation has been described(Horrey amp Wickens 2003) The experimentaldata reported in Horrey and Wickens (2004) werethose collected in a high-fidelity driving simula-tor as drivers drove on roadways of three differ-ing demands (varying traffic and curvature) whileengaging in concurrent tasks delivered in differ-ent modalities (auditory visual head up and headdown) A slightly modified (but conceptuallyequivalent) version of the multiple resource for-mula was imposed on the task interference datacreated by these nine conditions of the 3 times 3 de-sign for the three tasks of lane keeping respond-ing to unexpected hazards and performing thein-vehicle task For each task the dual-task in-terference score was assessed by subtracting the single-task (baseline) performance measure Theanalysis revealed great success in predicting bothhazard response (98 of variance accounted for

by the model) and in-vehicle task performance(92) However the model did not predict differ-ences in lane-keeping performance well at all fortwo reasons (a) separate ambient and focal visionchannels which were not included in that versionof the model Thus for example it is the ambientvision of the upper visual field that allows thedriver to glance downward and still keep the carheaded forward in the lane center (Horrey et al2006) (b) Drivers protected lane keeping frominterference (treated it as the primary task) henceprotecting it from variance in interference causedby structural differences in task resource competi-tion This clearly demonstrates the resource allo-cation effect

Before I conclude this section on multiple re-source model prediction it is important to empha-size that the primary value of such a model ispredicting relative differences in multitasking be-tween different conditions or interfaces Such amodel is not designed to make absolute predic-tions of performance ndash that is generate a predictedmultitask performance level that can be desig-nated as acceptable or unacceptable This issueas related to workload will be addressed later

MULTIPLE RESOURCES AND WORKLOAD

Multiple resource theory and mental workloadare two related concepts that are often confusedThey overlap but are distinct To distinguish themremember that the multiple resource model archi-tecture consists of three components related todemand resource overlap and allocation policyThe concept of mental workload relates moststrongly to the first of these characterizing thedemand imposed by tasks on the humanrsquos limitedmental resources whether considered as singleor multiple (Moray 1979)

Importantly this demand can be conceptuallyassociated with one of two ldquoregionsrdquo of taskdemand level (Wickens amp Hollands 2000) Thefirst is one in which the demand is less than thecapacity of resources available (ie there is ldquoresid-ual capacityrdquo unused in task performance) Thisis the ideal state because it means that the workerwill have some resources available should unex-pected circumstances be imposed The secondregion is one in which the demand exceeds thecapacity and as with any economic system (Navonamp Gopher 1979) performance will break downSometimes the distinction between these regions

MULTIPLE RESOURCES AND MENTAL WORKLOAD 453

is referred to as a ldquored linerdquo of workload (Grier2008)

Task demand varying along this continuumcan result from either single-task demand (egdriving progressively faster on a winding road im-poses progressively more resource demand untillane keeping begins to fail) or dual-task demandIn characterizing single-task demand the ldquomulti-plerdquo aspect of multiple resource theory has littlerelevance Furthermore whether one is doing oneor two tasks in the ldquoresidual capacityrdquo region mul-tiple resource theory is not fully germane Only inthe region where overload is imposed by multi-ple tasks does multiple resource theory make animportant contribution to mental workload by pre-dicting how much performance will fail once over-load has been reached (ie the size of dual-taskdecrements) Here multiple resource theory can ofcourse also provide guidance as to how redesigncan restore performance to the residual capacityregion

This distinction between regions is critical be-cause most measures of workload are designed asmuch for the residual capacity region as for theoverload region if not more so (Gawron 2008Tsang amp Vidulich 2006) Secondary tasks are ex-plicitly designed to probe ldquoresidual capacityrdquo notused for a primary task physiological and sub-jective measures operate across the entire rangeof both regions (and do not generally distinguishwhich resources within the multiple resourceframework are used) Thus the greatest value ofsuch measures is to ensure that task demands canremain within the residual capacity region Impor-tantly some models such as IMPRINTcopy (Laugheryet al 2006) or MIDAS (Gore amp Jarvis 2005) dohave a mental workload component with taskcompetition based on multiple resource theory andwith workload channels defined to correspond tothe dimensions in Figure 1 However this com-ponent is really intended to predict performancedecrements in the overload region as a result ofmultitask requirements which as we saw repre-sents only one portion of the workload conceptSuch models have been employed in important de-sign decisions regarding workload and crew size(Booher amp Minninger 2003)

FUTURE CHALLENGES FOR MULTIPLE RESOURCES

Although the multiple resource model has beenable to account for a good deal of variance in

time-sharing efficiency when heterogeneoustasks are imposed (Horrey amp Wickens 2003 Liuamp Wickens 1992 Sarno amp Wickens 1995 Wick-ens amp Colcombe 2007 Wickens Goh HellebergHorrey amp Talleur 2003 Wickens Sandry ampVidulich 1983) both the theory and the modelremain challenged by a number of issues Theseinclude the following

bull Adding another level to the modalities dimensionrelated to tactile input (Boles et al 2007)

bull The fact that other mechanisms unrelated to re-sources also account for variance in dual-task per-formance In particular these relate to confusionbetween task elements (eg the problems of time-sharing a perception and working memory task bothusing digits Hirst amp Kalmar 1987) and cooperationbetween elements (eg dual-task processing sup-ported by different dimensions of a single objectDuncan 1979 Wickens amp Hollands 2000)

bull The inability to characterize resource demand on asingle scale on which the ldquored linerdquo distinguishingthe reserve capacity from the overload region can beplaced (Hart amp Wickens 1990 Grier 2008)

bull Understanding what drives the allocation policy Inthe laboratory this often can be driven by primaryand secondary task instructions (Tsang 2006) but inthe real world phenomena such as unwanted diver-sion of attention to interruptions (Trafton amp Monk2008) ldquocognitive tunnelingrdquo (Wickens amp Alexan-der in press) and auditory preemption (Horrey amp Wickens 2004 Wickens amp Colcombe 2007Wickens Dixon amp Seppelt 2005) often operate inways that are clearly at odds with optimal allocationas witnessed by the safety compromise of drivingwhile using a cell phone (Strayer amp Drews 2007)Such findings are consistent with recent develop-ments that allocation policy is a function closely re-lated to that of the central executive a construct wellestablished to account for differences in time-sharingefficiency (Engle 2002 Monsell 2003 Wickens ampMcCarley 2008) as well as to identifiable prefrontalbrain regions (Kramer amp Parasuraman 2007)

Such challenges await the allocation of researchresources to the interested human factors investi-gator

ACKNOWLEDGMENTS

I thank Dr Bill Horrey and Dr David Boles fortheir comments on this manuscript

REFERENCES

Baddeley AD (1986) Working memory Oxford UK ClarendonBahrick H P Noble M amp Fitts P M (1954) Extra task performance

as a measure of learning a primary task Journal of ExperimentalPsychology 48 298ndash302

452 June 2008 ndash Human Factors

For the demand component each task is specifiedas being automated (D = 0) easy (D = 1) or dif-ficult (D = 2) independently of which resourcesmay be demanded (eg perception vs responseauditory vs visual) Hence the total task demandcan range from 0 (two automated tasks) to 4 (twodifficult tasks)

For the resource conflict component the twotasks are compared in the extent to which theyshare demands on common levels of each of thefour dimensions of the 3-D + 1 multiple resourcemodel 0 1 2 3 or 4 From the sum of these twocomponents a total interference component canbe produced ranging from 0 to 8 This is a simpli-fied version of the kind of computation that goesinto predicting multitask workload in softwarepackages such as IMPRINTcopy (Laughery LeBiereamp Archer 2006) or MIDAS (Booher amp Minninger2003 Gore amp Jarvis 2005) Of course the simpleprediction of the total interference between tasksdoes not inform the modeler as to which task suf-fers when there is interference and hence addeduse must be made of a third resource allocationcomponent of the multiple resource model whichcan express for example the extent to which driv-ing versus cell phone conversation suffers whenthe two compete for multiple resources This issueis discussed later

Although the validation of a more elaborateversion of this computational model is describedin Sarno and Wickens (1995) and in Wickens et al(2006) in studies demonstrating the importanceof invoking multiple rather than single resourcemodels a third validation has been described(Horrey amp Wickens 2003) The experimentaldata reported in Horrey and Wickens (2004) werethose collected in a high-fidelity driving simula-tor as drivers drove on roadways of three differ-ing demands (varying traffic and curvature) whileengaging in concurrent tasks delivered in differ-ent modalities (auditory visual head up and headdown) A slightly modified (but conceptuallyequivalent) version of the multiple resource for-mula was imposed on the task interference datacreated by these nine conditions of the 3 times 3 de-sign for the three tasks of lane keeping respond-ing to unexpected hazards and performing thein-vehicle task For each task the dual-task in-terference score was assessed by subtracting the single-task (baseline) performance measure Theanalysis revealed great success in predicting bothhazard response (98 of variance accounted for

by the model) and in-vehicle task performance(92) However the model did not predict differ-ences in lane-keeping performance well at all fortwo reasons (a) separate ambient and focal visionchannels which were not included in that versionof the model Thus for example it is the ambientvision of the upper visual field that allows thedriver to glance downward and still keep the carheaded forward in the lane center (Horrey et al2006) (b) Drivers protected lane keeping frominterference (treated it as the primary task) henceprotecting it from variance in interference causedby structural differences in task resource competi-tion This clearly demonstrates the resource allo-cation effect

Before I conclude this section on multiple re-source model prediction it is important to empha-size that the primary value of such a model ispredicting relative differences in multitasking be-tween different conditions or interfaces Such amodel is not designed to make absolute predic-tions of performance ndash that is generate a predictedmultitask performance level that can be desig-nated as acceptable or unacceptable This issueas related to workload will be addressed later

MULTIPLE RESOURCES AND WORKLOAD

Multiple resource theory and mental workloadare two related concepts that are often confusedThey overlap but are distinct To distinguish themremember that the multiple resource model archi-tecture consists of three components related todemand resource overlap and allocation policyThe concept of mental workload relates moststrongly to the first of these characterizing thedemand imposed by tasks on the humanrsquos limitedmental resources whether considered as singleor multiple (Moray 1979)

Importantly this demand can be conceptuallyassociated with one of two ldquoregionsrdquo of taskdemand level (Wickens amp Hollands 2000) Thefirst is one in which the demand is less than thecapacity of resources available (ie there is ldquoresid-ual capacityrdquo unused in task performance) Thisis the ideal state because it means that the workerwill have some resources available should unex-pected circumstances be imposed The secondregion is one in which the demand exceeds thecapacity and as with any economic system (Navonamp Gopher 1979) performance will break downSometimes the distinction between these regions

MULTIPLE RESOURCES AND MENTAL WORKLOAD 453

is referred to as a ldquored linerdquo of workload (Grier2008)

Task demand varying along this continuumcan result from either single-task demand (egdriving progressively faster on a winding road im-poses progressively more resource demand untillane keeping begins to fail) or dual-task demandIn characterizing single-task demand the ldquomulti-plerdquo aspect of multiple resource theory has littlerelevance Furthermore whether one is doing oneor two tasks in the ldquoresidual capacityrdquo region mul-tiple resource theory is not fully germane Only inthe region where overload is imposed by multi-ple tasks does multiple resource theory make animportant contribution to mental workload by pre-dicting how much performance will fail once over-load has been reached (ie the size of dual-taskdecrements) Here multiple resource theory can ofcourse also provide guidance as to how redesigncan restore performance to the residual capacityregion

This distinction between regions is critical be-cause most measures of workload are designed asmuch for the residual capacity region as for theoverload region if not more so (Gawron 2008Tsang amp Vidulich 2006) Secondary tasks are ex-plicitly designed to probe ldquoresidual capacityrdquo notused for a primary task physiological and sub-jective measures operate across the entire rangeof both regions (and do not generally distinguishwhich resources within the multiple resourceframework are used) Thus the greatest value ofsuch measures is to ensure that task demands canremain within the residual capacity region Impor-tantly some models such as IMPRINTcopy (Laugheryet al 2006) or MIDAS (Gore amp Jarvis 2005) dohave a mental workload component with taskcompetition based on multiple resource theory andwith workload channels defined to correspond tothe dimensions in Figure 1 However this com-ponent is really intended to predict performancedecrements in the overload region as a result ofmultitask requirements which as we saw repre-sents only one portion of the workload conceptSuch models have been employed in important de-sign decisions regarding workload and crew size(Booher amp Minninger 2003)

FUTURE CHALLENGES FOR MULTIPLE RESOURCES

Although the multiple resource model has beenable to account for a good deal of variance in

time-sharing efficiency when heterogeneoustasks are imposed (Horrey amp Wickens 2003 Liuamp Wickens 1992 Sarno amp Wickens 1995 Wick-ens amp Colcombe 2007 Wickens Goh HellebergHorrey amp Talleur 2003 Wickens Sandry ampVidulich 1983) both the theory and the modelremain challenged by a number of issues Theseinclude the following

bull Adding another level to the modalities dimensionrelated to tactile input (Boles et al 2007)

bull The fact that other mechanisms unrelated to re-sources also account for variance in dual-task per-formance In particular these relate to confusionbetween task elements (eg the problems of time-sharing a perception and working memory task bothusing digits Hirst amp Kalmar 1987) and cooperationbetween elements (eg dual-task processing sup-ported by different dimensions of a single objectDuncan 1979 Wickens amp Hollands 2000)

bull The inability to characterize resource demand on asingle scale on which the ldquored linerdquo distinguishingthe reserve capacity from the overload region can beplaced (Hart amp Wickens 1990 Grier 2008)

bull Understanding what drives the allocation policy Inthe laboratory this often can be driven by primaryand secondary task instructions (Tsang 2006) but inthe real world phenomena such as unwanted diver-sion of attention to interruptions (Trafton amp Monk2008) ldquocognitive tunnelingrdquo (Wickens amp Alexan-der in press) and auditory preemption (Horrey amp Wickens 2004 Wickens amp Colcombe 2007Wickens Dixon amp Seppelt 2005) often operate inways that are clearly at odds with optimal allocationas witnessed by the safety compromise of drivingwhile using a cell phone (Strayer amp Drews 2007)Such findings are consistent with recent develop-ments that allocation policy is a function closely re-lated to that of the central executive a construct wellestablished to account for differences in time-sharingefficiency (Engle 2002 Monsell 2003 Wickens ampMcCarley 2008) as well as to identifiable prefrontalbrain regions (Kramer amp Parasuraman 2007)

Such challenges await the allocation of researchresources to the interested human factors investi-gator

ACKNOWLEDGMENTS

I thank Dr Bill Horrey and Dr David Boles fortheir comments on this manuscript

REFERENCES

Baddeley AD (1986) Working memory Oxford UK ClarendonBahrick H P Noble M amp Fitts P M (1954) Extra task performance

as a measure of learning a primary task Journal of ExperimentalPsychology 48 298ndash302

MULTIPLE RESOURCES AND MENTAL WORKLOAD 453

is referred to as a ldquored linerdquo of workload (Grier2008)

Task demand varying along this continuumcan result from either single-task demand (egdriving progressively faster on a winding road im-poses progressively more resource demand untillane keeping begins to fail) or dual-task demandIn characterizing single-task demand the ldquomulti-plerdquo aspect of multiple resource theory has littlerelevance Furthermore whether one is doing oneor two tasks in the ldquoresidual capacityrdquo region mul-tiple resource theory is not fully germane Only inthe region where overload is imposed by multi-ple tasks does multiple resource theory make animportant contribution to mental workload by pre-dicting how much performance will fail once over-load has been reached (ie the size of dual-taskdecrements) Here multiple resource theory can ofcourse also provide guidance as to how redesigncan restore performance to the residual capacityregion

This distinction between regions is critical be-cause most measures of workload are designed asmuch for the residual capacity region as for theoverload region if not more so (Gawron 2008Tsang amp Vidulich 2006) Secondary tasks are ex-plicitly designed to probe ldquoresidual capacityrdquo notused for a primary task physiological and sub-jective measures operate across the entire rangeof both regions (and do not generally distinguishwhich resources within the multiple resourceframework are used) Thus the greatest value ofsuch measures is to ensure that task demands canremain within the residual capacity region Impor-tantly some models such as IMPRINTcopy (Laugheryet al 2006) or MIDAS (Gore amp Jarvis 2005) dohave a mental workload component with taskcompetition based on multiple resource theory andwith workload channels defined to correspond tothe dimensions in Figure 1 However this com-ponent is really intended to predict performancedecrements in the overload region as a result ofmultitask requirements which as we saw repre-sents only one portion of the workload conceptSuch models have been employed in important de-sign decisions regarding workload and crew size(Booher amp Minninger 2003)

FUTURE CHALLENGES FOR MULTIPLE RESOURCES

Although the multiple resource model has beenable to account for a good deal of variance in

time-sharing efficiency when heterogeneoustasks are imposed (Horrey amp Wickens 2003 Liuamp Wickens 1992 Sarno amp Wickens 1995 Wick-ens amp Colcombe 2007 Wickens Goh HellebergHorrey amp Talleur 2003 Wickens Sandry ampVidulich 1983) both the theory and the modelremain challenged by a number of issues Theseinclude the following

bull Adding another level to the modalities dimensionrelated to tactile input (Boles et al 2007)

bull The fact that other mechanisms unrelated to re-sources also account for variance in dual-task per-formance In particular these relate to confusionbetween task elements (eg the problems of time-sharing a perception and working memory task bothusing digits Hirst amp Kalmar 1987) and cooperationbetween elements (eg dual-task processing sup-ported by different dimensions of a single objectDuncan 1979 Wickens amp Hollands 2000)

bull The inability to characterize resource demand on asingle scale on which the ldquored linerdquo distinguishingthe reserve capacity from the overload region can beplaced (Hart amp Wickens 1990 Grier 2008)

bull Understanding what drives the allocation policy Inthe laboratory this often can be driven by primaryand secondary task instructions (Tsang 2006) but inthe real world phenomena such as unwanted diver-sion of attention to interruptions (Trafton amp Monk2008) ldquocognitive tunnelingrdquo (Wickens amp Alexan-der in press) and auditory preemption (Horrey amp Wickens 2004 Wickens amp Colcombe 2007Wickens Dixon amp Seppelt 2005) often operate inways that are clearly at odds with optimal allocationas witnessed by the safety compromise of drivingwhile using a cell phone (Strayer amp Drews 2007)Such findings are consistent with recent develop-ments that allocation policy is a function closely re-lated to that of the central executive a construct wellestablished to account for differences in time-sharingefficiency (Engle 2002 Monsell 2003 Wickens ampMcCarley 2008) as well as to identifiable prefrontalbrain regions (Kramer amp Parasuraman 2007)

Such challenges await the allocation of researchresources to the interested human factors investi-gator

ACKNOWLEDGMENTS

I thank Dr Bill Horrey and Dr David Boles fortheir comments on this manuscript

REFERENCES

Baddeley AD (1986) Working memory Oxford UK ClarendonBahrick H P Noble M amp Fitts P M (1954) Extra task performance

as a measure of learning a primary task Journal of ExperimentalPsychology 48 298ndash302

454 June 2008 ndash Human Factors

Bahrick H P amp Shelly C (1958) Time sharing as an index of automiza-tion Journal of Experimental Psychology 56 288ndash293

Boles D B (2002) Lateralized spatial processes and their lexical im-plications Neuropsychologia 40 2125ndash2135

Boles D B Bursk J H Phillips J B amp Perdelwitz J R (2007) Pre-dicting dual-task performance with the multiple resource question-naire Human Factors 49 32ndash45

Boles D B amp Law M B (1998) A simultaneous task comparison ofdifferentiated and undifferentiated hemispheric resource theoriesJournal of Experimental Psychology Human Perception and Per-formance 24 204ndash215

Booher H amp Minninger J (2003) Human systems integration in Armysystems acquisition In H R Booher (Ed) Handbook of humansystems integration (pp 663ndash698) Hoboken NJ Wiley

Briggs G Peters G amp Fisher R (1972) On the locus of the dividedattention effects Perception amp Psychophysics 11 315ndash320

Broadbent D (1971) Decision and stress London Academic PressDuncan J (1979) Divided attention The whole is more than the sum

of its parts Journal of Experimental Psychology Human Perceptionand Performance 5 216ndash228

Engle R W (2002) Working memory capacity as executive attentionCurrent Directions in Psychological Science 11 19ndash23

Gawron V J (2008) Human performance workload and situationawareness measures handbook 2nd ed Boca Raton FL CRC Press

Gore B F amp Jarvis P A (2005) New integrated modeling capabili-ties MIDASrsquo recent behavioral enhancements (SAW-2005-01-2701) Warrendale PA Society of Automotive Engineers

Grier R (2008) The redline of workload Theory research and designA panel To be presented at the 52nd Annual Meeting of the HumanFactors and Ergonomics Society September 22ndash26 in New York

Hart S G amp Wickens C D (1990) Workload assessment and pre-diction In H R Booher (Ed) MANPRINT An approach to systemsintegration (pp 257ndash296) New York Van Nostrand Reinhold

Hirst W amp Kalmar D (1987) Characterizing attentional resourcesJournal of Experimental Psychology General 116 68ndash81

Horrey W J amp Wickens C D (2003) Multiple resource modeling oftask interference in vehicle control hazard awareness and in-vehicletask performance In Proceedings of the Second International Driv-ing Symposium on Human Factors in Driver Assessment Trainingand Vehicle Design (pp 7ndash12) Iowa City Human Factors ResearchProgram University of Iowa

Horrey W J amp Wickens C D (2004) Driving and side task perfor-mance The effects of display clutter separation and modalityHuman Factors 46 611ndash624

Horrey W J Wickens C D amp Consalus K P (2006) Modeling driv-ersrsquo visual attention allocation while interacting with in-vehicletechnologies Journal of Experimental Psychology Applied 12(2)67ndash86

Isreal J Chesney G Wickens C D amp Donchin E (1980) P300 andtracking difficulty Evidence for a multiple capacity view of atten-tion Psychophysiology 17 259ndash273

Just M A Carpenter P A Keller T A Emery L Zajac H ampThulborn K R (2001) Interdependence of nonoverlapping corticalsystems in dual cognitive tasks Neuroimage 14 417ndash426

Just M A Carpenter P A amp Miyaki A (2003) Neuroindices of cog-nitive workload Theoretical Issues in Ergonomic Sciences 4 56ndash88

Kahneman D (1973) Attention and effort Englewood Cliffs NJPrentice Hall

Kalsbeek J amp Sykes R (1967) Objective measurement of mental loadActa Psychologica 27 253ndash261

Kantowitz B H amp Knight J (1976) Testing tapping time sharingAuditory secondary task Acta Psychologica 40 343ndash362

Kieras D (2007) Control of cognition In W Gray (Ed) Integratedmodels of cognitive systems (pp 327ndash355) New York Oxford Uni-versity Press

Kramer A F amp Parasuraman R (2007) Neuroergonomics Ap-plications of neuroscience to human factors In J T Cacioppo L G Tassinary amp G G Berntson (Eds) Handbook of psy-chophysiology (3rd ed pp 704ndash722) New York CambridgeUniversity Press

Laughery K R LeBiere C amp Archer S (2006) Modeling humanperformance in complex systems In G Salvendy (Ed) Handbookof human factors and ergonomics (3rd ed pp 965ndash996) HobokenNJ Wiley

Leibowitz H amp Post R (1982) The two modes of processing conceptand some implications In J Beck (Ed) Organization and repre-sentation in perception (pp 343ndash363) Hillsdale NJ Erlbaum

Liu Y amp Wickens C D (1992) Visual scanning with or without spa-tial uncertainty and divided and selective attention Acta Psycho-logica 79 131ndash153

Meyer D E amp Kieras D E (1997) A computational theory of exec-utive cognitive processes and multiple-task performance Part 2Accounts of psychological refractory-period phenomena Psycho-logical Review 104 749ndash791

Monsell S (2003) Task switching Trends in Cognitive Science 7134ndash140

Moray N (1967) Where is attention limited A survey and a modelActa Psychologica 27 84ndash92

Moray N (1979) Mental workload Its theory and measurement NewYork Plenum

Navon D amp Gopher D (1979) On the economy of the human pro-cessing systems Psychological Review 86 254ndash255

North R (1977) Task functional demands as factors in dual task per-formance In Proceedings of the 21st Annual Meeting of the HumanFactors Society (pp 367ndash371) Santa Monica CA Human FactorsSociety

North R amp Gopher D (1976) Measures of attention as predictors offlight performance Human Factors 18 1ndash14

Pashler H (1989) Dissociations and contingencies between speed andaccuracy Evidence for a two-component theory of divided atten-tion in simple tasks Cognitive Psychology 21 469ndash514

Pashler H (1998) The psychology of attention Cambridge MA MITPress

Polson M C amp Friedman A (1988) Task-sharing within and betweenhemispheres A multiple-resources approach Human Factors 30633ndash643

Previc F H (1998) The neuropsychology of 3-D space PsychologicalBulletin 124 123ndash164

Salvucci D amp Taatgen N A (2008) Threaded cognition An integratedtheory of concurrent multitasking Psychological Review 115101ndash130

Sarno K J amp Wickens C D (1995) The role of multiple resources inpredicting time-sharing efficiency An evaluation of three workloadmodels in a multiple task setting International Journal of AviationPsychology 5(1) 107ndash130

Strayer D amp Drews F (2007) Multi-tasking in the automobile In D W A Kramer D Wiegmann amp A Kirlik (Eds) Attention Fromtheory to practice (pp 121ndash133) New York Oxford UniversityPress

Trafton J G amp Monk C A (2008) Task interruptions In D ABoehm-Davis (Ed) Reviews of human factors and ergonomics(Vol 3 pp 111ndash126) Santa Monica CA Human Factors andErgonomics Society

Tsang P (2006) Regarding time-sharing with convergent operationsActa Psychologica 121 137ndash175

Tsang P amp Vidulich M A (2006) Mental workload and situationawareness In G Salvendy (Ed) Handbook of human factors ampergonomics (pp 243ndash268) Hoboken NJ Wiley

Welford A T (1967) Single channel operation in the brain ActaPsychologica 27 5ndash21

Wickens C D (1976) The effects of divided attention on informationprocessing in tracking Journal of Experimental PsychologyHuman Perception and Performance 2 1ndash13

Wickens C D (1980) The structure of attentional resources In R Nickerson (Ed) Attention and performance VIII (pp 239ndash257)Hillsdale NJ Erlbaum

Wickens C D (1984) Processing resources in attention In R Para-suraman amp R Davies (Eds) Varieties of attention (pp 63ndash101)New York Academic Press

Wickens C D (2002) Multiple resources and performance predictionTheoretical Issues in Ergonomics Science 3(2) 159ndash177

Wickens C D (2005) Multiple resource time sharing models In N Stanton A Hedge K Brookhuis E Salas amp H Hendrick(Eds) Handbook of human factors and ergonomics methods (pp401ndash407) Boca Raton FL CRC Press

Wickens C D (2007) How many resources and how to identify themCommentary on Boles et al and Vidulich and Tsang HumanFactors 49 53ndash56

Wickens C D amp Alexander A (in press) Attentional tunneling andtask management in synthetic vision displays International Journalof Aviation Psychology

Wickens C D amp Colcombe A (2007) Dual-task performance con-sequences of imperfect alerting associated with a cockpit displayof traffic information Human Factors 49 839ndash850

MULTIPLE RESOURCES AND MENTAL WORKLOAD 455

Wickens C D Dixon S R amp Ambinder M S (2006) Workloadand automation reliability in unmanned air vehicles In N J CookeH Pringle H Pedersen amp O Connor (Eds) Advances in humanperformance and cognitive engineering research Vol 7 Humanfactors of remotely operated vehicles (pp 209ndash222) AmsterdamElsevier

Wickens C D Dixon S R amp Seppelt B (2005) Auditory preemp-tion versus multiple resources Who wins in interruption manage-ment In Proceedings of the Human Factors and ErgonomicsSociety 49th Annual Meetingmdash2005 (pp 463ndash467) Santa MonicaCA Human Factors and Ergonomics Society

Wickens C D Goh J Helleberg J Horrey W amp Talleur D A(2003) Attentional models of multi-task pilot performance usingadvanced display technology Human Factors 45 360ndash380

Wickens C D amp Hollands J G (2000) Engineering psychology andhuman performance (3rd ed) Upper Saddle River NJ PrenticeHall

Wickens C D amp Liu Y (1988) Codes and modalities in multiple re-sources Asuccess and a qualification Human Factors 30 599ndash616

Wickens C D amp McCarley J (2008) Applied attention theory Boca-Raton FL Taylor amp Francis

Wickens C D Sandry D amp Vidulich M (1983) Compatibility andresource competition between modalities of input output and cen-tral processing Human Factors 25 227ndash248

Christopher D Wickens is a senior scientist at AlionScience Corporation Micro Analysis amp Design Opera-tions Boulder Colorado and professor emeritus at theUniversity of Illinois at Urbana-Champaign He re-ceived his PhD in psychology from the University ofMichigan in 1974

Date received September 7 2007Date accepted February 26 2008