RESEARCH DOSSIER
Multiple-Resource Interference in Concurrent Human Tasks
Abstract. Multiple-resource interference in concurrent human tasks describes the performance consequences of overlapping perceptual, representational, central-selection, and response demands. This critical integrative research monograph defines the construct, reconstructs the historical development of resource theories, and evaluates the four-dimensional multiple-resource model alongside central bottleneck, central capacity-sharing, EPIC, and threaded-cognition accounts. Original experiments, computational theories, meta-analyses, and applied driving studies are compared by design, mechanism, measurement, evidential limitations, and operational generalizability. The evidence supports the joint importance of sensory and response modality, representational code, temporal overlap, response selection, task priority, practice, and strategic scheduling. It also shows that preserving one routine performance measure can conceal costs in event detection or secondary task performance. The resulting synthesis specifies principles for experimentally defensible task decomposition, measurement, causal model comparison, and resource-aware human-system design. This is a critical integrative monograph, not a protocol-driven systematic review or an original meta-analysis.
Keywords: multiple-resource theory; dual-task interference; concurrent task performance; psychological refractory period; divided attention; response selection; modality pairing; central capacity sharing; task priority; time sharing; mental workload; threaded cognition; EPIC; driving distraction; human factors; cognitive ergonomics.
The Scientific Problem of Concurrent Human Tasks
Human activity regularly involves partially overlapping demands: a driver interprets traffic while controlling a vehicle and responding to a passenger; an aircrew member monitors several displays while entering a clearance; an emergency clinician tracks a patient while communicating with a team. Such activities become scientifically difficult when two individually manageable tasks compete during the same interval. Multiple-resource interference refers to the measurable impairment, delay, variability, or change of priority arising because concurrent task components demand overlapping perceptual, representational, central-selection, or motor-processing resources. The phenomenon is defined at the level of task interaction, with specific resource demands and outcomes that an experiment can manipulate and measure. It is therefore distinguishable from an individual's general tendency to become distracted or a retrospective judgment that an experience was difficult.
A complete explanation must account for both interference and successful time sharing. Some pairs of activities produce pronounced secondary-task costs despite apparently separate sensory channels; other pairs can be combined with limited loss after practice. The existence of several possible constraints explains why the number of tasks alone is a poor measure of concurrent workload. The central research questions are which operations overlap, what processing resources they require, how temporal priority is assigned, how task representations conflict, and under what conditions interference generalizes from laboratory paradigms to operational performance. The four-dimensional multiple-resource framework developed by Wickens, C. D (2002) and Wickens, C. D (2008) supplies a major starting point, while central-bottleneck, capacity-sharing, and computational coordination theories provide essential alternatives. The outcome of this investigation is an evidence-grounded account of their complementary and competing predictions.
Definitions, Units of Analysis, and Scope
A task is a goal-directed set of perceptual, cognitive, or motor operations organized around an outcome. A concurrent-task episode occurs when the demands of two or more tasks have a temporal overlap relevant to performance. Dual-task interference is inferred when joint performance changes relative to an appropriate single-task or lower-overlap comparison, with attention to practice, order, incentive, and speed–accuracy tradeoffs. An observed cost is a behavioral pattern; the attribution to a particular resource is a theoretical explanation requiring selective experimental evidence.
Resource is an operational construct describing constrained processing capability, channel, or stage, whose joint demand predicts competition under specified conditions. Modality refers to the sensory input or response medium, such as vision, audition, manual action, or speech. Processing code distinguishes characteristics of the information being transformed, particularly spatial and verbal operations. Stage concerns perceptual/cognitive processing versus response operations in the Wickens model. The degree of overlap depends on how a task is implemented; the same nominal goal can produce different profiles with a graphical rather than spoken interface. Wickens, C. D., Sandry, D. L., & Vidulich, M (1983) gives evidence that input, central processing, and output assignments affect competition.
Several neighboring research objects should remain analytically separate. Divided attention is an overarching cognitive process; task switching analyzes sequential alternation and reconfiguration; mental workload measures demand or experienced effort; vigilance concerns sustained monitoring over time. This dossier concentrates on moment-to-moment interference among concurrently required operations. It uses those related literatures as explanatory supports and does not absorb their entire subject matter. Monsell (2003)00028-7) and Kiesel, A., Steinhauser, M., Wendt, M., Falkenstein, M., Jos (2010) establish why switch costs and overlapping-task costs require different experimental contrasts.
Historical Foundations of Resource Theories
The intellectual history of dual-task research proceeds through successive questions about the limits of selection, the distribution of effort, and the organization of processing components. Early theories of attention made clear that different events can require selection or priority. Later capacity accounts treated human processing as economically constrained and sensitive to task demand, incentive, and strategic allocation. The resource-economy argument in Navon, D., & Gopher, D (1979) is historically important because it makes allocation a property of a system optimizing competing performances, not simply an intrinsic number of tasks that a person can carry out.
Engineering psychology moved that question into concrete task pairs. The empirical issue became whether performance suffers primarily when sensory or motor channels are shared or whether additional central competition remains after those channels are separated. Wickens, C. D., Sandry, D. L., & Vidulich, M (1983) investigated compatibility and competition across input, central operations, and output in experiments involving memory search, tracking, and simulated flight activities. This tradition established the importance of decomposing a task before predicting a joint cost. It also linked psychological theory to display placement, communication modalities, control assignments, and real operator performance.
The next important historical development was the sustained examination of response selection. Pashler, H (1994) reviewed dual-task evidence pointing toward central limitations in simple choice tasks. Later work challenged universal serialization through intensive practice and flexible scheduling, including Schumacher, E. H., Seymour, T. L., Glass, J. M., Fencsik, D (2001) and the EPIC architecture of Meyer, D. E., & Kieras, D. E (1997). Quantitative central capacity sharing Tombu, M., & Jolicœur, P (2003) and resource-mediated computational coordination Salvucci, D. D., & Taatgen, N. A (2008) subsequently offered formal alternatives. These lines of research retain distinct assumptions: some specify the content and overlap of resources; others specify how stages are sequenced, shared, or controlled.
The Four-Dimensional Multiple-Resource Framework
The contemporary version of multiple-resource theory predicts interference from the similarity of task demands along several dimensions rather than from a single global capacity. In Wickens, C. D (2002), the relevant dimensions are processing stages, sensory modalities, processing codes, and the distinction between focal and ambient visual channels. Two tasks whose demands overlap on more of these dimensions are expected, other things held reasonably comparable, to compete more strongly. The model has value because interface designers can often change the mapping of a secondary task onto channels without changing its larger objective.
This is a predictive framework, not a complete neurological inventory. Its dimensions are abstractions over functional performance regularities. A verbal auditory warning and an ongoing visual tracking task may avoid some conflicts of sensory acquisition; an auditory verbal command may still compete with linguistic response selection or speech. Likewise, two visual activities can share input modality while differing in code or spatial and focal organization. Wickens, C. D (2008) discusses a computational implementation and emphasizes the strongest usefulness of the model for overload-associated performance failures.
The resource framework should be read conditionally. A cross-modal pairing is not automatically safe, and the salience or timing of a cue can change resource demands. Task design also controls how long a resource is held, whether operations occur together, the precision of required responses, and the consequences of an error. A scientifically meaningful prediction therefore names the competing tasks, their mapping to model dimensions, the overlap window, and the endpoint to be tested.
Processing Stages and the Locus of Competition
A task can be analyzed into phases of stimulus registration, perceptual categorization, working representation, decision or response selection, and action execution. The decomposition matters because two tasks sharing a stimulus modality may interact differently from tasks sharing only a response stage. In the classic tracking and search research, manual response competition affected tracking in a way that depended on which operations overlapped, whereas the visual encoding of memory-search material created a separate locus of cost. Wickens, C. D., Sandry, D. L., & Vidulich, M (1983) provides the foundational experimental comparison.
Stage distinctions also prevent the misleading assumption that every delay indicates sensory overload. A participant may detect two signals successfully and still be unable to select and initiate both responses at the same pace. Conversely, an apparently slowed motor response may arise from an upstream decision delay. Pashler, H (1994) describes the behavioral signatures of such central limitations, and Meyer, D. E., & Kieras, D. E (1997) models executive scheduling in relation to concurrent perceptual-motor activity. These accounts generate tests in which a manipulation of perceptual difficulty is compared with one of response selection or motor complexity.
For operational prediction, stages must be assigned on the basis of actual behavior rather than interface labels. Reading a short number may be largely visual, but verbally classifying it may add phonological coding and response selection. Pressing a button may be mechanically simple while deciding which button to press is costly. Resource profiles are better treated as hypotheses to be verified by targeted manipulations of task components.
Sensory Modality and Cross-Modal Design
The separation of visual and auditory input can sometimes reduce competition between tasks, particularly when one activity already saturates visual search or scanning. This motivates auditory alerts during visually demanding work and visual feedback when speech perception is heavily occupied. The experimental background in Wickens, C. D., Sandry, D. L., & Vidulich, M (1983) shows that modality assignments influence dual-task interference, and the larger model in Wickens, C. D (2002) translates this finding into an engineering prediction.
The meaning of a cue and its required response remain central. A spoken warning may impose comprehension, working-memory, decision, and verbal response requirements. A visual icon may be quickly detected in peripheral vision while requiring intensive focal analysis once it is interpreted. A channel substitution can therefore redistribute the bottleneck rather than eliminate it. Source selection is likewise relevant: one auditory stream may mask another, and visual clutter can affect detection before central resources are considered. These physical and cognitive effects should be measured separately whenever possible.
An informative experiment crosses input modality with output modality and balances single-task difficulty. Such designs permit inference about whether an apparent cross-modal advantage results from input separation or from how stimuli and responses are paired. Hazeltine, E., & Ruthruff, E (2006) and Hazeltine, E., Ruthruff, E., & Remington, R. W (2006) specifically demonstrate why an input-only model is insufficient. The scientific lesson for design is that modality optimization must include the whole path from sensory acquisition to action.
Representational Codes and Semantic Conflict
Multiple-resource theories distinguish spatial representations, such as position, direction, or trajectory, from verbal representations, such as names and symbolic sequences. Tasks may interfere because they require overlapping representational transformations even when their input channels differ. Tracking a changing spatial target while remembering directions can be more demanding than a task pair that partitions representational demands, but the magnitude and direction of this effect depend on the actual response mappings and task priorities. The code dimension in Wickens, C. D (2002) formalizes this principle.
Representational competition introduces an important limitation to crude sensory explanations. Two tasks that begin in different modalities may converge on the same decision category, action rule, or verbal representation. A visual road sign and a spoken navigation command can both demand the driver's selection of a direction. Thus, the relevant unit of analysis may be a common internal operation, and its overlap may be invisible from outward sensory descriptions. Content-dependent interference in Hazeltine, E., Ruthruff, E., & Remington, R. W (2006) directly supports studying this convergence.
It follows that research should specify the representational code at a level that is experimentally discriminable. Descriptive labels such as cognitive, complex, or information-heavy do little analytic work unless they are translated into stimulus classifications, memory transformations, competing response rules, or spatial judgments. A successful mechanistic account predicts which manipulation changes the joint cost while leaving noncompeting elements approximately constant.
Focal and Ambient Vision
Vision does not function as a uniform processing channel for applied task analysis. Detailed recognition, text reading, and symbolic discrimination usually demand focal inspection, whereas broad spatial orientation, motion, and optic-flow information can support guidance using a different distribution of visual activity. The focal–ambient distinction in Wickens, C. D (2002) recognizes that some concurrent visual tasks differ in interference despite sharing the visual modality. This is especially relevant to vehicle guidance, spatial display design, and monitoring environments with motion information.
The distinction does not license a claim that ambient processing is costless or entirely independent. A pilot may use broad visual flow for orientation and still lose situational information because an inset display draws gaze, decision time, or working memory. A driver may preserve lane position while missing a rare braking cue. An observed stable tracking measure therefore cannot establish the absence of interference; a second endpoint can reveal that attentional costs have shifted to the monitoring requirement. Horrey, W. J., & Wickens, C. D (2006) demonstrates why driving research must distinguish reaction-time outcomes from tracking measures.
The applied prediction should specify the task's visual information, its retinal or display characteristics, gaze demands, and response endpoint. A focal–ambient resource hypothesis becomes testable when one changes display placement or type of visual representation and measures multiple concurrent performance dimensions, including missed events and safety margins.
Central Response Selection and Psychological Refractory Period
A prominent competing theory proposes that central response-selection operations cannot always proceed fully in parallel. The psychological refractory period (PRP) paradigm tests this claim by presenting two stimuli with varying stimulus-onset asynchrony. When the second event occurs shortly after the first, the response to the second event can be delayed even when its own sensory analysis has begun. The characteristic pattern has been used to infer postponement of a central stage. Pashler, H (1994) synthesizes the traditional evidence.
The PRP is valuable because changing temporal overlap while holding task identities constant can expose a bottleneck that simple comparisons of two independent sensory channels would miss. The usual task order matters: first-task priority can protect the first response while the second waits. Manipulations of perceptual duration and choice requirements can help localize the delay. Yet the inference is model-dependent; a similar reaction-time pattern can emerge from strongly constrained capacity sharing or adaptive scheduling. A reaction-time slope alone should never be treated as a direct image of a discrete mental processor.
Operational relevance is illustrated by Levy, J., Pashler, H., & Boer, E (2006), who applied overlapping response demands to simulated driving and examined braking alongside a separate choice task. Reduced stimulus-onset asynchrony slowed braking, while input and output modality arrangements exerted additional effects. The case demonstrates the joint importance of central timing and resource configuration. Even a trained, familiar motor response may suffer when another task requires overlapping selection operations.
Central Capacity Sharing as an Alternative Model
The central capacity-sharing approach models central processing as divisible rather than obligatorily monopolized by one task at a time. In the account developed by Tombu, M., & Jolicœur, P (2003), a limited central capacity is allocated between simultaneous tasks, and a set of formal equations predicts reaction-time patterns as the temporal relationship between tasks changes. Importantly, the model was constructed to reproduce several signatures traditionally interpreted as evidence for strict bottleneck serialization. The distinction between theories therefore depends on experiments designed around their divergent predictions, not on the mere presence of a PRP effect.
This approach emphasizes continuous allocation and the consequences of priority. A task may proceed more slowly because it receives a smaller share of an available process, and the first task can also incur costs when overlap increases. Tombu, M., & Jolicœur, P (2005) tested predictions of the model. Its analytic strength lies in specifying which outcomes would change under capacity redistribution and in forcing a contrast between qualitatively similar timing patterns generated by different mechanisms.
A general scientific challenge remains: capacity is rarely observed directly. Investigators infer it from response-time distributions, manipulations, and model fit. If models with different structural assumptions describe the same aggregate means, independent predictions involving response variability, task order, preparation, or individual differences become essential. The responsible synthesis treats the central-bottleneck and capacity-sharing accounts as empirically contestable models of central interaction, each useful within explicitly tested conditions.
Executive Process Interactive Control
The executive-process interactive control architecture, or EPIC, is a computational framework for explaining how executive routines schedule perceptual, cognitive, and motor operations when multiple tasks are present. Meyer, D. E., & Kieras, D. E (1997) introduced basic mechanisms that permit explicit simulation of concurrent performance and procedural control. Its importance lies in turning broad propositions about strategy into sequenced operations with measurable response-time predictions.
EPIC challenges the idea that all delays must be assigned to an immutable central structural bottleneck. Task-specific executive coordination, interleaving, and rules for initiating or postponing operations can contribute materially to observed costs. However, flexibility does not imply unlimited capacity: peripheral mechanisms, task requirements, and effective scheduling rules still constrain performance. A simulation that fits a task pair provides a candidate mechanism whose generality must be tested on other conditions rather than assumed.
The contrast with bottleneck accounts became influential after Schumacher, E. H., Seymour, T. L., Glass, J. M., Fencsik, D (2001) reported cases approaching near-perfect time sharing after practice in laboratory choice tasks. Subsequent work including Tombu, M., & Jolicœur, P (2004) challenged whether some comparisons adequately excluded methodological confounds. Together, these studies show how theory advances through carefully designed boundary conditions. The relevant question is which combinations of training, stimulus mapping, response demands, and scheduling rules allow efficient overlap, and whether these conditions persist when a rare, unexpected event interrupts routine operation.
Threaded Cognition and Resource Acquisition
Threaded cognition presents concurrent tasks as partially independent threads that compete for shared processing resources through a general coordination mechanism. Salvucci, D. D., & Taatgen, N. A (2008) integrated perceptual, motor, and central procedural constraints within a computational theory, demonstrating how resource acquisition and release can produce either interference or efficient overlap. The proposed architecture explains coordination without requiring a separate executive controller for every concurrent task episode.
A crucial contribution is the representation of temporal structure. Two activities can be individually demanding yet have complementary schedules: one uses perception while the other completes a motor action. Conversely, tasks that share a resource only briefly can experience a substantial delay when critical stages coincide. The model is therefore well suited to investigating why two apparently identical workloads behave differently under altered pacing or environmental interruptions.
Threaded cognition is a theoretical model, and the degree of empirical confirmation depends on which tested tasks and parameters are considered. The simulations in Salvucci, D. D., & Taatgen, N. A (2008) cover laboratory dual-choice tasks and applied situations, including driver distraction, but any transfer to new technologies needs new behavioral validation. Its enduring methodological value is the requirement to describe resource use as a sequence of acquisition, contention, processing, and release rather than as a static label attached to a whole task.
Input–Output Pairing and Content-Dependent Central Costs
Input and output modality pairings provide particularly strong evidence against simplistic models in which every central operation is interchangeable. Hazeltine, E., Ruthruff, E., & Remington, R. W (2006) contrasted visual-manual plus auditory-vocal task combinations with visual-vocal plus auditory-manual combinations. The latter configurations produced substantially greater dual-task costs across practice in the reported experiments, despite controls addressing peripheral resource overlap and the single-task speed of the constituent activities. The investigators interpreted the persistent difference as evidence that central interference depends partly on the content of processing operations.
A related PRP experiment reported by Hazeltine, E., & Ruthruff, E (2006) varied modality pairings while keeping the families of stimuli and responses comparable. It found relatively small single-task effects of pairing together with larger dual-task effects. This pattern challenges explanations based entirely on independent sensory and response modules or a content-neutral duration of response selection. The relationship between input and output mappings may alter the ease of selecting and transforming responses.
These findings create useful design constraints. An auditory signal followed by a spoken response may combine well with visual-manual activity in one experimental family, but generalizing that advantage requires matching response semantics, urgency, proficiency, and temporal overlap. Modality assignments cannot be evaluated in isolation from action mappings. The stronger hypothesis is that the interaction among input, internal code, and response demands predicts the specific location and size of joint performance loss.
Response Conflict, Compatibility, and Motor Coupling
Motor resources have a concrete role in concurrent-task interference. Two tasks demanding the same hand, gaze transition, or speech production can collide at the level of action execution even when perception and categorization proceed independently. A control operation can also conflict with the response code of a second task: a participant may need to make a spatial steering movement while selecting a button mapped to a competing spatial category. The modality and response competition work of Wickens, C. D., Sandry, D. L., & Vidulich, M (1983) makes these distinct loci explicit.
Experimental interpretation requires more than counting limbs or input devices. Automatic and well-practiced responses can still require response selection or postural stabilization; manual output can differ greatly in duration, precision, and interruptibility. When a joint condition impairs one activity, the experiment must establish whether the cause is mechanical incompatibility, higher-order selection, motor planning, or deliberate prioritization. A simple task pair can confound all of these without factorial control.
A useful design varies input pairing, response mapping, and motor demands independently, then records both tasks. If costs change when the same motor requirements are reassigned to another selection rule, the evidence points beyond a purely peripheral limitation. Such contrasts are consistent with the multi-component analysis developed by Hazeltine, E., & Ruthruff, E (2006) and the causal discipline needed to distinguish resource overlap from task difficulty.
Temporal Overlap, Scheduling, and Event Density
Concurrency is a temporal property of processing episodes, and nominal task simultaneity does not guarantee that the demands on a given resource coincide. A person may alternate between two tasks often enough to appear constantly busy while critical central stages occur in a serial sequence. Conversely, two sparse activities can produce a large cost when rare high-priority events happen together. The PRP methods summarized by Pashler, H (1994) turn overlap into an experimentally controlled factor through stimulus-onset asynchrony.
Event density matters because a system with sufficient average spare capacity may fail during brief periods of high temporal demand. A stable mean workload estimate cannot directly describe the tail risk of missed alerts during short overload episodes. The temporal schedule of stimulus sampling, motor corrections, confirmation messages, and unexpected interruptions therefore belongs in a realistic task model. Salvucci, D. D., & Taatgen, N. A (2008) is instructive here because it represents contention at the level of resource acquisition and timing.
A research design should examine distributions rather than only aggregate means. The proportion of very slow responses, missed urgent events, response reversals, and recovery time after an interruption can be operationally significant. Time-resolved modeling should be matched to the hazards of the environment. A visually smooth lane trajectory may conceal a dangerous delay to an infrequent brake signal, as demonstrated conceptually by the driving evidence in Levy, J., Pashler, H., & Boer, E (2006).
Task Priority, Strategic Allocation, and Speed–Accuracy Tradeoffs
Human performers actively allocate processing effort according to instructions, goals, incentives, and perceived consequences. An observer who sees unchanged primary-task accuracy during a secondary task cannot conclude that there was no interference; the operator may have protected the primary goal at the expense of the secondary task. Navon, D., & Gopher, D (1979) describes resource allocation as an economic problem, and Schumacher, E. H., Seymour, T. L., Glass, J. M., Fencsik, D (2001) reports that instructions about priority can modify dual-task performance.
This produces a central measurement problem. A dual-task cost is indexed relative to a chosen baseline, but the meaning of that comparison changes with which outcome the participant is encouraged to defend. Faster responding can be purchased at a cost in errors; a protected safety task can leave a lower-priority communication task with longer completion times. The appropriate unit of evaluation is often a joint performance tradeoff, expressed as a profile across both tasks, rather than a single score.
Experimental protocols should report task-priority instructions, feedback, compensation for speed and accuracy, the ordering of tasks, and evidence of compliance with the instructions. When two tasks are each measured in isolation, both single-task baselines are necessary, and performance operating characteristics should be explored under different priority conditions. This makes strategic adaptation observable and prevents it from being misclassified as a change in structural resources.
Practice, Automaticity, and Skill Acquisition
Practice can alter dual-task costs by reducing processing durations, stabilizing mappings, and improving how stages are scheduled. In Schumacher, E. H., Seymour, T. L., Glass, J. M., Fencsik, D (2001), some trained participants reached levels of joint choice-task performance described as virtually perfect time sharing, suggesting that the interference expected under unpracticed conditions can be strongly attenuated. This evidence is important because a theory predicting permanent large costs for every paired choice task would be too rigid.
The subsequent critique in Tombu, M., & Jolicœur, P (2004) argued that potential confounding features of earlier conditions could conceal residual interference; experiments designed to minimize those confounds found statistically detectable costs. The debate illustrates how different operational definitions of near-perfect time sharing, training histories, and comparison conditions change conclusions. Evidence for substantial practice-related improvement is compatible with a remaining constraint under more stringent measurement.
Transfer is the unresolved practical issue. Training two repetitive laboratory mappings may yield an efficient local solution without preparing a person to handle a new alert, rule reversal, equipment failure, or unexpected increase in task difficulty. Claims of automaticity therefore require separate evidence about retention, perturbation, transfer to new response mappings, and sensitivity to rare events. In safety-critical design, trained performance under routine conditions should be measured alongside performance under simulated abnormal conditions.
Operational Measurement of Dual-Task Interference
The most direct empirical design compares performance in joint and isolated task conditions, with careful control over practice, order, priority, and temporal demands. Depending on the endpoint, researchers measure accuracy, missed signals, response times, response-time distributions, tracking error, action variability, completion time, or recovery after interruption. A dual-task cost can be expressed as a change from a single-task baseline, but percentages and differences require consistent denominators and explicit reporting of whether higher or lower values indicate better performance. McIsaac, T. L., Lamberg, E. M., & Muratori, L. M (2015) documents major variation in dual-task terminology and measurement procedures.
A task pair must also be characterized for comparability. One can compare a choice reaction under a secondary tracking task, or examine a continuous task while intermittent decision events are injected. These paradigms sample different interference processes. A high average completion rate during continuous monitoring may conceal event-specific impairment, while average reaction-time slowing may capture a cost that has little practical relevance for a noncritical task. The measure should be chosen according to the mechanism and outcome under investigation.
Interpretation is strengthened when both tasks are measured concurrently and when primary and secondary instructions are varied systematically. Behavioral data can be complemented by eye tracking and physiological measures, but none alone identifies the location of interference. A stronger conclusion emerges when selective manipulations of modality, code, response mapping, or timing generate a distinctive predicted pattern across multiple endpoints.
Mental Workload and Subjective Effort
Mental workload concerns demand, experienced effort, and the capacity available to respond, but it is not numerically identical to dual-task interference. A task pair can feel demanding yet be performed successfully; an individual can experience high interference without reporting a correspondingly high effort score. The NASA Task Load Index developed by Hart & Staveland (1988)62386-9) provides a multidimensional subjective assessment of workload-related experience, including mental and physical demand, temporal demand, performance, effort, and frustration. It is most useful when interpreted alongside behavioral outcomes.
The relation between workload and multiple resources is explicitly considered in Wickens, C. D (2008). The model is especially relevant to predicted performance breakdown when tasks overlap in their resource requirements. Yet workload judgments reflect expectations, strategies, and individual experience, which can differ across participants who achieve similar performance. A researcher should therefore avoid treating a single workload rating as direct proof of one particular processing bottleneck.
An integrated assessment records objective dual-task costs, timing of critical episodes, perceived workload, and the strategy used to protect different tasks. In operational systems, the discrepancy between subjective confidence and observed missed events may itself be informative. The goal is not to collapse these indicators into a universal scalar but to understand which indicators predict the consequences of overload in the task environment.
Experimental Design and Causal Identification
A decisive dual-task experiment manipulates factors tied to rival explanations while controlling overall difficulty. A factorial design can cross visual versus auditory input, manual versus vocal response, congruent versus incongruent response codes, and short versus long overlap intervals. The target result is an interaction: a specific pattern of interference that is larger or smaller under one mapping than another, beyond a change in single-task performance. This approach is exemplified by Hazeltine, E., Ruthruff, E., & Remington, R. W (2006).
Repeated testing must address learning and order effects. Counterbalancing or randomization of condition sequences helps separate treatment differences from familiarity and fatigue. Each participant's proficiency, stimulus-response associations, and task priority should be documented. Reaction-time analysis should distinguish correct response times from error rates, and when possible use distributional rather than only mean comparisons. A mechanism should be favored when its specified predictions survive these controls and novel experimental conditions.
Operational experiments introduce further constraints. High-fidelity simulation can reproduce time pressure but may create incentives unlike those in live work. Field observations can document natural workload and rare events but invite confounding by weather, traffic, task choice, or operator selection. A disciplined evidence program combines controlled mechanistic studies, increasingly realistic simulations, and field measures, making the limits of each design explicit. Its aim is the identification of a testable cause of interference rather than the retrospective labeling of a task as complex.
Evidence from Simulated Driving
Driving is an unusually informative domain because it combines continuous spatial control, distributed visual monitoring, decision making, and occasional urgent response. In controlled simulated-driving studies, Strayer, D. L., & Johnston, W. A (2001) reported that some passive listening conditions did not produce the same disruption as engaging phone conversation. Their experiments associated conversation with slower reactions and missed simulated traffic signals, including with hands-free use. This evidence points toward cognitive processing demands beyond the mechanical act of holding a device.
The experiment in Levy, J., Pashler, H., & Boer, E (2006) made a different contribution by examining the time relationship between an overlapping choice response and braking. Braking reactions lengthened when the stimulus events were placed closer together, consistent with central interference, while variations in stimulus and response modalities also affected performance. These studies support a multifactor explanation that includes both resource configuration and temporal central operations.
Driving evidence also shows why a single aggregate measure can mislead. A driver may maintain adequate lane position through periods when detection of a rare traffic event becomes impaired. The operator's apparently stable routine control performance can be the product of strong task priority rather than spare processing capacity. Laboratory tasks permit isolation of mechanisms, but a road system introduces social interaction, unexpected hazards, vehicle technologies, and variation in risk. Evidence from simulation should be integrated with field designs rather than assumed to describe every collision mechanism.
Meta-Analytic and Epidemiological Evidence
Aggregated driving research provides an important check on conclusions derived from one laboratory protocol. Horrey, W. J., & Wickens, C. D (2006) combined 23 studies contributing 47 analysis entries and reported costs from phone conversation that were especially apparent in reaction-time measures, with smaller effects for some tracking measures. Caird, J. K., Willness, C. R., Steel, P., & Scialfa, C (2008) included 33 studies meeting its criteria, contributing 94 effect-size estimates and a total participant sample of approximately 2,000. These syntheses demonstrate that the observed size and type of interference depend on the selected endpoint, task demands, and study design.
Epidemiological evidence asks a different question. In a case-crossover study of 699 drivers involved in property-damage collisions, Redelmeier, D. A., & Tibshirani, R. J (1997) compared phone use close to collision times with control intervals within the same drivers. The authors reported an association between phone use and collision risk (relative risk 4.3, 95% confidence interval 3.0–6.5, for the reported comparison). Such an association supplies a field safety perspective, but it does not identify the precise combination of selection, modality, and motor mechanisms involved. Observational control, exposure measurement, and historical technology context matter.
A synthesis across methods should not pool incompatible outcomes into a single abstract claim about human multitasking. Reaction delay, tracking error, missed signals, and collision associations carry different meanings. The most robust conclusion is that concurrent communication can impair important components of driving performance and that the explanatory contribution of central processing demands persists even when a manual interface is removed.
Aviation, Supervisory Control, and High-Reliability Work
Aviation and complex supervisory environments require operators to monitor changing states, communicate, manipulate controls, and respond to irregular events. These tasks provide the application context in which multiple-resource theory was developed. Wickens, C. D., Sandry, D. L., & Vidulich, M (1983) examined simulated flight and task combinations involving spatial and verbal operations, supporting the practical claim that sensory and response assignment can change competition among concurrent activities. The design question is how to distribute demands so that essential information remains perceptible and actionable during periods of time pressure.
A potential benefit of modality redistribution is accompanied by a requirement for cross-channel coordination. A spoken alert can preserve visual availability, but an operator may need to interpret the message, consult a display, remember a conditional rule, and execute a manual response. The sequence can produce a central bottleneck after early sensory acquisition is complete. Resource theory therefore supports testing the entire operator-system interaction rather than simply adding more displays or more alarms.
Evaluation in such systems should prioritize event detection, task sequencing, decision correctness, response to abnormal conditions, and recovery from interruption. Workload measures alone cannot determine safety. Realistic crew and supervisory simulations also include communications from other people, task delegation, and variable authority, which can produce coordination costs beyond an individual cognitive model. The individual resource framework becomes one part of a wider human-systems assessment.
Healthcare, Mobility, and Other Applied Settings
Concurrent-task interference also appears in clinical workflows, mobility, rehabilitation, and other contexts in which a person must coordinate cognitive activity with monitoring or movement. The dual-task taxonomy discussed by McIsaac, T. L., Lamberg, E. M., & Muratori, L. M (2015) emphasizes that the literature uses heterogeneous tasks and definitions, particularly when research crosses laboratory and clinical settings. The meaning of a cost depends on whether the experiment tests walking while calculating, responding while maintaining balance, or communicating while completing a precision procedure.
The methodological priority in these settings is a clear description of both task components and an outcome meaningful to the situation. For mobility, investigators may be interested in gait variability, obstacle avoidance, or recovery of balance; in healthcare, a lapse in monitoring or an error in documentation may be more important than the total time to complete an auxiliary task. A change in performance may also reflect intentional adaptation, such as slowing walking to maintain stability, rather than an uncontrolled failure of attention.
Cross-population interpretation requires sensitivity to age, practice, motor limitations, familiarity with the environment, and baseline task proficiency. It is inappropriate to treat a cost measured with one task pair as a general index of a person's cognitive capacity. A well-specified multiple-resource analysis instead predicts which combination of task codes, motor outputs, timing, and response demands explains the observed performance pattern.
Digital Multitasking and Technology-Mediated Distraction
Digital environments intensify opportunities for overlapping demands through notifications, conversations, application transitions, and simultaneous media streams. Not every instance of digital multitasking constitutes concurrent resource interference: repeated switching between separate tasks may primarily involve reconfiguration and carryover costs. The task-switching review of Kiesel, A., Steinhauser, M., Wendt, M., Falkenstein, M., Jos (2010) and the account in Monsell (2003)00028-7) support separating the effects of frequent alternation from the effects of simultaneous operations.
Research on habitual media multitasking raises a different level of analysis. Ophir, E., Nass, C., & Wagner, A. D (2009) reported associations between patterns of media use and aspects of cognitive control in their participant sample. This cross-sectional line of research does not by itself establish that media exposure caused a stable cognitive trait, nor does it determine the real-time cost of a particular notification while a task is ongoing. A valid design must distinguish acute experimental interruption, chronic usage patterns, task preference, and self-selection.
For the immediate topic of this dossier, a digital notification is best modeled as a sequence of events: detection, classification, relevance judgment, possible response selection, and return to the original task. Some stages may overlap with the primary activity, whereas others force an interruption or task switch. The empirical target should be the cost of specified configurations, such as reading an alert while tracking a moving object, rather than the broad assumption that all screen use is equivalent.
Automation, Artificial Systems, and Cognitive Partnership
Automation can redistribute demands between a human and a technical system, but the resulting human task configuration must be experimentally evaluated. Parasuraman, R., Sheridan, T. B., & Wickens, C. D (2000) distinguishes automation of information acquisition, analysis, decision selection, and action implementation. Each type changes what the human must monitor, verify, decide, or execute. A system may remove repetitive manual work while increasing the burden of interpreting machine recommendations or intervening during unexpected conditions.
Within the Artificial Era, human–artificial coordination is a research domain in which competing resource demands can be explicitly measured. Consider an operator who visually controls a dynamic process while an artificial assistant issues spoken explanations, a recommendation, and a request for approval. Sensory separation may reduce visual search competition; understanding the recommendation may still require central selection and working memory. The psychologically relevant question is how the interface changes the operator's total task schedule, including error checking, trust calibration, and response urgency.
The concepts of artificial agency and cognitive partnership should be used here as theoretical descriptions of system roles and observable interactions. They do not establish a new resource theory on their own. Empirical tests can compare advisory timing, concise versus extended explanations, automated action versus human confirmation, and interruption policies while measuring joint outcomes. The central design principle is to allocate operations in ways that fit task demand, verified human capabilities, and the consequences of failure.
Individual Differences, Task Familiarity, and Generalization
Dual-task performance varies with experience, sensory accessibility, motor proficiency, motivation, and the properties of the task pair. A resource framework should allow for differences in performance without presuming that every observed difference reflects a permanent trait. Practice and instructions have demonstrable effects on coordination, as indicated by Schumacher, E. H., Seymour, T. L., Glass, J. M., Fencsik, D (2001). A trained professional may handle a familiar pair of operations efficiently while showing costs when an unusual response mapping is introduced.
Generalization is particularly sensitive to task baselines. Two people can show the same percentage cost while differing greatly in absolute reaction time, or have different apparent costs because one already performs the single task near a ceiling or floor. Measurements must therefore report absolute performance, relative changes, errors, and baseline variance. If a group comparison involves age, health status, or professional background, the design must also consider sensory and motor differences that can alter effective resource assignments.
The scientific purpose of individual-difference analysis is to predict performance under well-defined conditions. A resource theory becomes stronger when parameters estimated on one task configuration predict independent conditions or when an intervention changes outcomes in the direction specified beforehand. A theory that merely relabels every difference after it has occurred lacks such predictive force.
Replication, Model Comparison, and Evidential Boundaries
The dual-task literature contains experimental findings, theoretical models, computational simulations, narrative reviews, and quantitative evidence syntheses. These forms of evidence answer different questions. A controlled manipulation of modality pairing can identify a causal influence of that pairing within the tested design; a simulation can show that a proposed mechanism reproduces observations; a meta-analysis can characterize cross-study variability under its inclusion rules. None of these alone establishes a universal resource structure.
Model comparison is especially demanding because strict bottleneck, central capacity-sharing, and strategic scheduling models can reproduce overlapping sets of reaction-time phenomena. Tombu, M., & Jolicœur, P (2003) demonstrated how central sharing predicts standard PRP patterns, while Tombu, M., & Jolicœur, P (2004) challenged claims of virtually interference-free performance under certain training conditions. Hazeltine, E., Ruthruff, E., & Remington, R. W (2006) further shows that theories assuming content-neutral central operations can miss systematic modality-pairing effects. A productive comparison therefore requires preregistered contrasts where rival models predict different interaction patterns.
Independent replication should vary research teams, task materials, populations, and apparatus while preserving critical theoretical manipulations. Effect magnitudes should be presented with uncertainty intervals and linked to task priorities and safety-relevant outcomes. Publication bias, small samples, flexible exclusion criteria, and reuse of narrowly similar laboratory tasks can all exaggerate apparent generality. The strongest scientific claims remain those that survive changes in setting while retaining a specified mechanism and outcome.
An Integrated Method for Predicting Interference
An applied prediction can proceed through a series of explicit analytic decisions. First, define the primary task and the prospective concurrent task, including all observable endpoints. Second, decompose both into sensory acquisition, representation, central decisions, and response execution. Third, determine the timing and duration of each critical operation. Fourth, generate rival predictions from the multiple-resource dimensions and from central bottleneck or sharing models. Fifth, test whether priority instructions, practice, and response mappings alter the observed cost.
This approach follows the strengths of Wickens, C. D (2002), Pashler, H (1994), and Salvucci, D. D., & Taatgen, N. A (2008) while preserving their theoretical differences. A task pair that shares visual acquisition may be improved by shifting one input to audition, but the predicted benefit should be qualified when the auditory stream requires a verbal decision at the same moment as the primary task's critical selection. Two noncompeting channels can coexist with a shared central process; conversely, one visually rich continuous task can coexist with a sparse secondary signal if their critical operations are temporally separated.
Predictive performance should be evaluated on untrained conditions or independent holdout data, with sensitivity analyses for priority, workload, and participant characteristics. A successful model specifies the conditions under which it will fail and guides a verifiable redesign. The scientific objective is reliable operational prediction, not the automatic selection of a single favored theory across every task configuration.
Design Implications and Intervention Evaluation
Resource-aware design can change the input modality of information, response modality, compatibility of control mappings, temporal schedule of alerts, degree of automation, and priority structure of tasks. These are testable interventions rather than generic recommendations to multitask less. Evidence from Wickens, C. D., Sandry, D. L., & Vidulich, M (1983) and Wickens, C. D (2008) provides a rationale for reducing overlap in the dimensions most heavily demanded by a primary activity. The magnitude of benefit must be established for the actual operator, task, and environment.
An important distinction is between reducing the average number of interactions and protecting the most safety-critical seconds. An alert scheduler that suppresses nonessential messages during an imminent hazard might reduce interference even when the overall number of messages stays unchanged. Conversely, combining many alerts into one event may create a longer and more complex response decision at precisely the wrong moment. Task decomposition and event-time analysis are therefore necessary for evaluating proposed remedies.
The intervention should be judged using primary-task safety outcomes, secondary-task function, errors, response latency, subjective workload, recovery, and performance during abnormal episodes. Trials comparing design variants should specify which outcome receives priority and whether users can strategically adapt. The existence of a theoretically plausible redesign is a reason to test it, not proof that operational safety has already improved.
Contemporary Tests of Resource Sharing and Cognitive Capacity
Recent experiments continue to test whether interference follows the interaction of task content and limited processing capacity rather than a uniform limit triggered by any second task. Wu, Y., & Wang, Q (2024) examined tone discrimination alongside lexical classification across stimulus-onset asynchronies, including reversed task order. Their two experiments reported continuing dual-task interference when processing was manipulated between more automatic and more controlled conditions. The authors interpreted the results as consistent with resource sharing, because both tasks showed evidence of reduced processing efficiency. That inference should be evaluated against the precise operational definition of automaticity, the task-specific stimulus mapping, and the competing predictions of strategic scheduling. A single experimental family advances a mechanistic debate without settling every form of the bottleneck question.
An experiment with more differentiated task content was reported by Fleming, J. T., Njoroge, J. M., Noyce, A. L., Perrachione, T (2024). Participants retained temporal or spatial information in working memory while an intervening auditory task also required temporal or spatial discrimination. The pattern of recall and perceptual errors was sensitive to overlapping sensory modality and information domain, with particularly marked difficulty when both operations were auditory-temporal. Pupillary changes and neural measurements furnished additional indicators of demand. This evidence directly supports the thesis that modality and representational content interact rather than functioning as wholly independent labels. It also demonstrates why behavioral accuracy near a ceiling may conceal more subtle resource competition observable in neural or physiological measures.
The working-memory experiments of Pougeon, J., Camos, V., Belletier, C., & Barrouillet, P (2024) examine a related but distinguishable question: how storage and processing can share limited resources when each is calibrated to the participant's own capability. Their experiments combine letter retention with parity judgments, using relative performance to estimate the degree of reciprocal cost. This type of design is valuable because it supplies a quantitative interpretation of sharing rather than simply classifying a task pair as interfering. Interpretation nonetheless depends on assumptions about the relation between invested capacity and performance and on the specificity of the storage-processing tasks being studied.
These three lines of research extend the original engineering-psychology framework in complementary directions. One isolates automaticity and central timing, another jointly manipulates modality and information domain while recording multiple physiological outcomes, and the third quantifies the tradeoff between memory storage and cognitive processing. None justifies assigning a universal fixed resource budget to all people or tasks. Together, they motivate explicitly multidimensional models with identifiable, falsifiable predictions.
Neural Representation, Practice, and the Interpretation of Effort
An emerging research program tests whether interference partly depends on neural representational overlap. Mueckstein, M., Görgen, K., Heinzel, S., Granacher, U., Rapp (2025) examined modality-based dual-task interference and practice-related change using behavioral and neural indicators. The reported findings link improved multitasking performance to separation of competing task representations in sensory brain regions. This provides a candidate mechanism for some practice effects: training can alter the configuration of representations used by concurrent tasks rather than simply accelerate all responses by an equal amount. The evidence belongs to the specific mappings and training procedures investigated, and the strongest generalization awaits further independent replications.
Neural differentiation alone cannot establish that every interference cost is caused by representational crosstalk. Task performance depends on sensory encoding, response mapping, central selection, priority policies, and timing, so imaging or electrophysiological differences should be interpreted in relation to behavioral manipulations and explicit computational predictions. The contemporary evidence of Fleming, J. T., Njoroge, J. M., Noyce, A. L., Perrachione, T (2024) is instructive because the same experiment measured behavioral errors, pupil dilation, event-related potentials, and alpha-band oscillatory activity. Agreement between measures strengthens an explanation when the measurements are expected to covary for a specific mechanistic reason; disagreement can reveal ceiling effects, compensatory effort, or distinct processing stages.
The measurement question also appears in research on listening. Kuchinsky, S. E., Gallun, F. J., & Lee, A. K. C (2024) analyzed the dual-task paradigm as a measure of listening effort and highlighted that estimates depend on the identity of the secondary task, processing resources, and experimental design. Their discussion clarifies why a secondary-task cost cannot be taken as a universal unit of mental effort across different primary activities. A verbal secondary task may be especially sensitive to linguistic competition, whereas a spatial motor task may index a partly different constraint. When such indicators are applied in auditory accessibility research, task calibration and the selected theory of resource allocation become essential parts of construct validation.
These findings place the original resource models in a more measurable contemporary context. They connect task content, the plasticity of neural representation, and performance tradeoffs while preserving the distinction between an observed cost and a proposed physiological explanation. Their practical value is greatest when investigators combine a priori hypotheses, independent behavioral endpoints, and sufficiently specific manipulations to identify the process under study.
Unresolved Questions and Research Agenda
Several unresolved problems define the next phase of the field. The first is how to identify the processing content responsible for central conflict. Modality pairing effects reported by Hazeltine, E., Ruthruff, E., & Remington, R. W (2006) indicate that a simple count of simultaneous central operations may omit crucial representational structure. Future experiments should independently manipulate semantic mappings, response codes, and temporal overlap, while estimating distributional response outcomes.
The second problem is predictive transportability. Models developed with short laboratory choice tasks must account for rare-event monitoring, extended operations, communication with other people, and failure recovery. The third concerns adaptation: a model should predict how training, task priority, and situational urgency alter resource sharing. The fourth is measurement: workload ratings, eye movements, physiological indices, and overt performance should be integrated without treating any one signal as direct proof of a particular cognitive resource.
The emergence of adaptive artificial assistants makes this agenda operationally urgent. Such assistants can time alerts, transform modalities, summarize information, and execute actions, but they may also create new confirmation, supervision, and coordination demands. An experimentally credible account will compare human outcomes under multiple intervention policies, evaluate rare consequential events, and investigate whether benefits generalize to unfamiliar tasks. These questions can be answered through preregistered factorial experiments, model comparison, high-fidelity simulations, and longitudinal field evaluation.
Conclusion
Multiple-resource interference is a structured interaction among concurrent task demands. Its observable forms include delayed responses, missed signals, changes in tracking or decision accuracy, shifts in task priority, and recovery costs. The history of engineering psychology demonstrates that the pattern cannot be explained solely by the number of simultaneous tasks. Resource dimensions, the content of central operations, temporal overlap, response mapping, and flexible coordination jointly influence performance.
The four-dimensional model developed by Wickens provides a practical way to predict task competition from stage, sensory modality, code, and visual-channel overlap. Central bottleneck and capacity-sharing models reveal important limits of selection under close temporal overlap, while EPIC and threaded cognition describe executive scheduling and resource acquisition. Experimental findings on modality pairing and the practice debate establish that the models must be compared by their distinctive predictions rather than grouped under a single explanation. Studies of driving show that an apparently preserved routine control task can coexist with serious costs to rare-event detection.
A durable scientific synthesis therefore joins clear task decomposition, experimentally tested mechanisms, multiple performance endpoints, and independent validation in realistic systems. The central achievement of this research program is a method for anticipating when task combinations will impair human performance and for redesigning the joint activity around measured, context-dependent processing requirements.
References
- Caird, J. K., Willness, C. R., Steel, P., & Scialfa, C. (2008). A meta-analysis of the effects of cell phones on driver performance. Accident Analysis & Prevention, 40(4), 1282–1293. https://doi.org/10.1016/j.aap.2008.01.009
- Fleming, J. T., Njoroge, J. M., Noyce, A. L., Perrachione, T. K., & Shinn-Cunningham, B. G. (2024). Sensory modality and information domain contribute jointly to dual-task interference between working memory and perceptual processing. Imaging Neuroscience, 2, Article imag_a_00130. https://doi.org/10.1162/imag_a_00130
- Hart, S. G., & Staveland, L. E. (1988). Development of NASA-TLX (Task Load Index): Results of empirical and theoretical research. In P. A. Hancock & N. Meshkati (Eds.), Human Mental Workload (pp. 139–183). North-Holland. https://doi.org/10.1016/S0166-4115(08)62386-9
- Hazeltine, E., & Ruthruff, E. (2006). Modality pairing effects and the response selection bottleneck. Psychological Research, 70(6), 504–513. https://doi.org/10.1007/s00426-005-0017-3
- Hazeltine, E., Ruthruff, E., & Remington, R. W. (2006). The role of input and output modality pairings in dual-task performance: Evidence for content-dependent central interference. Cognitive Psychology, 52(4), 291–345. https://doi.org/10.1016/j.cogpsych.2005.11.001
- Horrey, W. J., & Wickens, C. D. (2006). Examining the impact of cell phone conversations on driving using meta-analytic techniques. Human Factors, 48(1), 196–205. https://doi.org/10.1518/001872006776412135
- Kiesel, A., Steinhauser, M., Wendt, M., Falkenstein, M., Jost, K., Philipp, A. M., & Koch, I. (2010). Control and interference in task switching—A review. Psychological Bulletin, 136(5), 849–874. https://doi.org/10.1037/a0019842
- Kuchinsky, S. E., Gallun, F. J., & Lee, A. K. C. (2024). Note on the dual-task paradigm and its use to measure listening effort. Trends in Hearing, 28, Article 23312165241292215. https://doi.org/10.1177/23312165241292215
- Levy, J., Pashler, H., & Boer, E. (2006). Central interference in driving: Is there any stopping the psychological refractory period? Psychological Science, 17(3), 228–235. https://doi.org/10.1111/j.1467-9280.2006.01690.x
- McIsaac, T. L., Lamberg, E. M., & Muratori, L. M. (2015). Building a framework for a dual task taxonomy. BioMed Research International, 2015, Article 591475. https://doi.org/10.1155/2015/591475
- Meyer, D. E., & Kieras, D. E. (1997). A computational theory of executive cognitive processes and multiple-task performance: Part 1. Basic mechanisms. Psychological Review, 104(1), 3–65. https://doi.org/10.1037/0033-295X.104.1.3
- Monsell, S. (2003). Task switching. Trends in Cognitive Sciences, 7(3), 134–140. https://doi.org/10.1016/S1364-6613(03)00028-7
- Mueckstein, M., Görgen, K., Heinzel, S., Granacher, U., Rapp, M. A., & Stelzel, C. (2025). Multitasking practice eliminates modality-based interference by separating task representations in sensory brain regions. Journal of Neuroscience, 45(4), Article e0755242024. https://doi.org/10.1523/JNEUROSCI.0755-24.2024
- Navon, D., & Gopher, D. (1979). On the economy of the human-processing system. Psychological Review, 86(3), 214–255. https://eric.ed.gov/?id=EJ217039
- Ophir, E., Nass, C., & Wagner, A. D. (2009). Cognitive control in media multitaskers. Proceedings of the National Academy of Sciences, 106(37), 15583–15587. https://doi.org/10.1073/pnas.0903620106
- Parasuraman, R., Sheridan, T. B., & Wickens, C. D. (2000). A model for types and levels of human interaction with automation. IEEE Transactions on Systems, Man, and Cybernetics—Part A: Systems and Humans, 30(3), 286–297. https://doi.org/10.1109/3468.844354
- Pashler, H. (1994). Dual-task interference in simple tasks: Data and theory. Psychological Bulletin, 116(2), 220–244. https://doi.org/10.1037/0033-2909.116.2.220
- Pougeon, J., Camos, V., Belletier, C., & Barrouillet, P. (2024). Quantifying resource sharing in working memory. Psychonomic Bulletin & Review, 31(5), 2305–2312. https://doi.org/10.3758/s13423-024-02494-4
- Redelmeier, D. A., & Tibshirani, R. J. (1997). Association between cellular-telephone calls and motor vehicle collisions. New England Journal of Medicine, 336(7), 453–458. https://doi.org/10.1056/NEJM199702133360701
- Salvucci, D. D., & Taatgen, N. A. (2008). Threaded cognition: An integrated theory of concurrent multitasking. Psychological Review, 115(1), 101–130. https://doi.org/10.1037/0033-295X.115.1.101
- Schumacher, E. H., Seymour, T. L., Glass, J. M., Fencsik, D. E., Lauber, E. J., Kieras, D. E., & Meyer, D. E. (2001). Virtually perfect time sharing in dual-task performance: Uncorking the central cognitive bottleneck. Psychological Science, 12(2), 101–108. https://doi.org/10.1111/1467-9280.00318
- Strayer, D. L., & Johnston, W. A. (2001). Driven to distraction: Dual-task studies of simulated driving and conversing on a cellular telephone. Psychological Science, 12(6), 462–466. https://doi.org/10.1111/1467-9280.00386
- Tombu, M., & Jolicœur, P. (2003). A central capacity sharing model of dual-task performance. Journal of Experimental Psychology: Human Perception and Performance, 29(1), 3–18. https://doi.org/10.1037/0096-1523.29.1.3
- Tombu, M., & Jolicœur, P. (2004). Virtually no evidence for virtually perfect time-sharing. Journal of Experimental Psychology: Human Perception and Performance, 30(5), 795–810. https://doi.org/10.1037/0096-1523.30.5.795
- Tombu, M., & Jolicœur, P. (2005). Testing the predictions of the central capacity sharing model. Journal of Experimental Psychology: Human Perception and Performance, 31(4), 790–802. https://doi.org/10.1037/0096-1523.31.4.790
- Wickens, C. D. (2002). Multiple resources and performance prediction. Theoretical Issues in Ergonomics Science, 3(2), 159–177. https://doi.org/10.1080/14639220210123806
- Wickens, C. D. (2008). Multiple resources and mental workload. Human Factors, 50(3), 449–455. https://doi.org/10.1518/001872008X288394
- Wickens, C. D., Sandry, D. L., & Vidulich, M. (1983). Compatibility and resource competition between modalities of input, central processing, and output. Human Factors, 25(2), 227–248. https://doi.org/10.1177/001872088302500209
- Wu, Y., & Wang, Q. (2024). Dual-task interference: Bottleneck constraint or capacity sharing? Evidence from automatic and controlled processes. Attention, Perception, & Psychophysics, 86(3), 815–827. https://doi.org/10.3758/s13414-024-02854-1