Dual-Task Tests

Abstract

Dual-task tests are a family of task performance and analysis methods that measure what happens to performance when a person attempts two tasks at once. By comparing performance on each task alone with performance on the two combined, they expose the limits of divided attention and furnish a quantitative index, the dual-task cost. Two classical accounts explain that cost: a structural bottleneck, which forces one central operation to wait for another, and a graded capacity, shared between tasks until it runs out. The paradigm underlies the psychological refractory period in the laboratory and, applied, the walking-while-talking tests that predict falls and the driving studies that showed conversation, not manual handling, impairs control. Modern work treats multitasking cost as a product of cognitive structure, flexibility and plasticity rather than a single fixed limit.

Keywords: dual-task cost, psychological refractory period, bottleneck, attentional capacity, cognitive-motor interference

A dual-task test is any procedure that requires a person to perform two tasks concurrently and compares the result against performance on each task alone. The logic is simple and powerful: if two activities draw on the same limited processing resource, doing them together will degrade one or both, and the size of that degradation measures how much they compete (Pashler, 1994). Because almost every mental operation can be paired with another, the dual-task method has become one of psychology's most general tools, used to map the architecture of attention in the laboratory and, in the clinic, to probe the hidden cost of walking while thinking. The Medical Subject Headings file it under task performance and analysis, the broad category of methods for measuring how people carry out defined tasks. This article sets out what the tests measure, the two theories that explain the cost they reveal, how that cost is quantified, and the clinical and applied work the paradigm now drives.

What Dual-Task Tests Are

Every dual-task test has the same three-part structure (Figure 1). A person first performs a primary task alone, establishing a single-task baseline; then a secondary task alone, establishing its baseline; then both together. Comparing the concurrent performance with the two baselines isolates the effect of combining them, holding the tasks themselves constant. If performance is unchanged, the two tasks are drawing on independent resources and can proceed in parallel. If one or both decline, they are competing for something shared, and the paradigm's purpose is to find out what (Pashler, 1994).

The method is a probe of divided attention, the ability to distribute processing across more than one input or action at a time, and it is the complement of the selective-attention tasks that ask a person to ignore a distractor rather than to handle it. What makes it diagnostic is the choice of task pairing. Pairing a demanding task with a trivial one, or two tasks that use entirely separate input and output channels, reveals little; pairing two tasks that both demand central processing reveals the structural limit (McIsaac et al., 2015). The secondary task can also be used as a load: a researcher who wants to know whether an activity requires central resources can simply add a standard secondary task and see whether it suffers, a technique working memory research uses to fractionate the system into its components (Baddeley et al., 1991).

Figure 1

The Bottleneck and Single-Channel Theory

The oldest explanation of dual-task cost is structural. Reaction-time experiments in the 1930s found that when a second stimulus follows a first too quickly, the response to it is delayed, as though the system had entered a refractory phase after the first (Telford, 1931). A. T. Welford gave this the form it held for decades: a single-channel theory in which a central mechanism can carry out only one response selection, the operation of choosing what to do about a stimulus, at a time (Welford, 1952). Perception and motor execution may run in parallel, but the central decision stage is a bottleneck through which tasks must pass one after another.

This produces the signature phenomenon of the laboratory dual-task test, the psychological refractory period (PRP): when two choice-reaction tasks are presented with a short stimulus onset asynchrony (SOA, the gap between the two stimuli), the response to the second is slowed, and the slowing grows as the SOA shrinks. Harold Pashler formalized the logic that makes the PRP a measuring instrument rather than a curiosity. If the bottleneck lies at response selection, then lengthening the perceptual stage of the second task, which happens before the bottleneck, should be absorbed into the waiting time at short SOAs and leave the second response time unchanged; lengthening a stage at or after the bottleneck should not. This is the locus-of-slack method, and it localizes exactly which stage is the shared one (Pashler, 1994). In the first demonstration below, varying the SOA and the duration of the second task's perceptual stage shows the bottleneck create, and then absorb, the slack.

Capacity, Sharing, and the Graded View

The bottleneck is not the only account, and for many task pairings it is not the best one. Daniel Kahneman proposed the major alternative: attention is a single pool of undifferentiated attentional capacity that can be allocated, in graded amounts, to concurrent activities (Kahneman, 1973). On this view there is no hard switch that serializes processing; instead each task draws as much capacity as it needs, and interference appears only when their combined demand exceeds the supply. Effort is the mobilization of this capacity, and the pool can expand somewhat with arousal, which is why a person can sometimes do two things at once that they could not do when tired.

The capacity model predicts a graded trade-off rather than an all-or-none delay. As a person shifts allocation from one task toward the other, performance on the first should fall as the second rises, tracing a smooth performance operating characteristic. The capacity-sharing model unified the two traditions by proposing that the central stage is not strictly serial but shares a limited capacity between tasks, so that giving more to one necessarily leaves less for the other and lengthens its processing in proportion (Tombu & Jolicœur, 2003). Which picture holds depends on the tasks and on practice: heavily overlearned skills that have reached automaticity draw little capacity and interfere little, while novel or effortful tasks compete fiercely. The second demonstration allocates a fixed capacity between two tasks and traces the trade-off curve the graded model predicts.

Measuring Dual-Task Cost

Whatever the underlying mechanism, the test yields a number: the dual-task cost (DTC), the change in performance from single- to dual-task conditions. The standard metric is the proportional dual-task cost, which expresses the change as a percentage of the single-task baseline so that tasks measured in different units, seconds, metres per second, items recalled, can be compared on one scale (Plummer & Eskes, 2015):

DTC (%) = (dual-task − single-task) / single-task × 100.

A negative value means performance worsened under dual-task conditions, the usual result; a value near zero means the tasks did not compete. Because the cost is computed separately for each task, a single trial yields two numbers, and their pattern is itself informative. Plummer and Eskes set out a four-quadrant interference model from these paired costs: both tasks may decline (mutual interference), one may be protected at the other's expense (a trade-off), one may decline while the other is unaffected, or, rarely, one may even improve. Reporting both costs, rather than collapsing them into one, is what distinguishes a diagnostic dual-task test from a mere demonstration that multitasking is hard (Plummer & Eskes, 2015). The third demonstration computes both proportional costs from values the reader sets and places the result in the correct interference quadrant.

From Laboratory to Clinic

The dual-task test left the laboratory because the cost it measures predicts things that matter. The most developed clinical application is in gait. A clinical team noticed that older people who stopped walking when they began to talk were far more likely to fall in the following months, turning an informal observation into a validated screening test (Lundin-Olsson et al., 1997). The underlying phenomenon, cognitive-motor interference, is now well quantified: a meta-analysis of walking-while-thinking studies confirmed that adding a cognitive secondary task reliably slows gait and increases its variability, with larger effects in older and neurologically impaired groups (Al-Yahya et al., 2011). The dual-task paradigm also serves as a sensitive assay of the central executive: in Alzheimer's disease, the ability to coordinate two tasks declines even when performance on each alone is relatively preserved, making dual-task cost an early marker of the disorder's impact on working memory's central executive (Baddeley et al., 1991).

The applied strand that reached the public concerns driving. In a series of dual-task studies, David Strayer and colleagues showed that conversing on a mobile phone impairs simulated driving, producing a form of inattention blindness to objects in the visual scene, and, crucially, that the impairment comes from the conversation and not from the manual handling of the phone, so that a hands-free set offers little protection (Strayer & Johnston, 2001). Against this uniform picture of impairment stands a striking exception. A small minority of people, perhaps one in forty, show almost no dual-task cost even on demanding combinations; Strayer's group named them supertaskers and argued that their existence bears on whether the bottleneck is a fixed structural feature of the brain or a limit that varies across individuals (Watson & Strayer, 2010).

Worked Example

Consider an older adult assessed with a standard walking-while-talking test, the kind the gait literature uses (Plummer & Eskes, 2015). The two single-task baselines are measured first. Walking alone, the person's gait speed is 1.20 m/s. Performing serial-seven subtractions while seated, they produce 20 correct answers per minute. Then the two are combined, walking while subtracting aloud. Under the dual task, gait speed falls to 0.90 m/s and correct answers fall to 17 per minute.

Apply the proportional dual-task cost separately to each task. For gait, DTC = (0.90 − 1.20) / 1.20 × 100 = (−0.30 / 1.20) × 100 = −25.0%: the person walks a quarter slower when also calculating. For the cognitive task, DTC = (17 − 20) / 20 × 100 = (−3 / 20) × 100 = −15.0%: they also lose about a sixth of their arithmetic output. Because the costs are expressed as percentages of each baseline, the 0.30 m/s drop in speed and the 3-per-minute drop in answers, quantities in utterly different units, become directly comparable: the motor task suffered more than the cognitive one.

The sign of both costs is what classifies the result. Both are negative, so both tasks declined when combined: this is mutual interference, the quadrant of Plummer and Eskes's model in which neither task is protected and the two genuinely compete for shared capacity. Had the person held gait speed constant at 1.20 m/s while their arithmetic collapsed, the pattern would instead be a trade-off, signalling that they prioritized walking, perhaps because they sensed instability. The numbers alone do not say which task should be protected; that is a clinical judgment. But by converting raw performance into a pair of signed proportional costs, the dual-task test turns a vague impression that someone “struggles to multitask” into two comparable quantities and a classifiable interference pattern, which is exactly what makes it useful for tracking change over the course of a rehabilitation.

Discussion

The long contest between bottleneck and capacity accounts has not ended in a clean victory for either, and the modern consensus is that it should not. For very simple choice-reaction tasks at short SOAs, the data fit a discrete central bottleneck at response selection remarkably well, and the locus-of-slack logic continues to localize that stage (Pashler, 1994). For richer task combinations, and for the graded trade-offs seen when people deliberately shift priority, a shared-capacity model fits better, and the capacity-sharing formulation shows the two are limiting cases of one scheme rather than rival theories (Tombu & Jolicœur, 2003). The practical upshot is that the dual-task cost is not a single quantity with a single cause; what it measures depends on the tasks paired and the stage at which they collide, which is precisely why a taxonomy of task pairings is needed to interpret it (McIsaac et al., 2015).

A second lesson is that dual-task cost is not fixed within a person. Practice reduces it, sometimes to near zero, as a task is overlearned into automaticity and stops demanding central resources; the existence of supertaskers shows that even without special practice the limit varies across individuals (Watson & Strayer, 2010). This individual and trainable variation is what makes the paradigm clinically valuable: a dual-task cost that shrinks over a course of rehabilitation is evidence that central coordination is recovering, not merely that one task has improved in isolation.

Current Directions

Contemporary research has moved in two directions. The first is theoretical integration. Rather than asking whether the limit is a bottleneck or a capacity, recent reviews frame human multitasking as the joint product of three things: the cognitive structure that determines which operations must share a stage, the flexibility with which people reconfigure task sets, and the plasticity through which practice reshapes both. On this view the dual-task and task-switching traditions, long pursued separately, are studying two faces of the same control system (Koch et al., 2018).

The second direction is the refinement of the clinical gait application into a diagnostic instrument. A theoretical framework now links specific components of the central executive to specific subphases of gait, posture, initiation, steady walking and turning, so that the pattern of dual-task interference, not just its presence, carries information about where control is failing (Bayot et al., 2018). Taxonomies of secondary cognitive tasks have been built to standardize which load is applied, after meta-analysis showed that different cognitive tasks, mental tracking, verbal fluency, reaction time, interfere with gait to very different degrees (Wollesen et al., 2019). Dual-task gait is increasingly evaluated as an early marker of cognitive decline, with reviews assessing which parameters best predict later impairment (Ramírez & Gutiérrez, 2021), and the most recent meta-analyses continue to sharpen the age norms, confirming that the locomotor-cognitive cost is reliably larger in older than in younger adults (Mustafovska et al., 2025). The through-line is a steady movement from a laboratory measure of attentional architecture toward a validated clinical sign.

Key Researchers

Alan Baddeley

(b. 1934). Used the dual-task method to isolate the central executive of working memory and showed that dual-task coordination declines selectively in Alzheimer's disease, making dual-task cost a diagnostic probe. Faculty · Wikipedia · Google Scholar

Daniel Kahneman

(1934–2024). Proposed the capacity (resource) model of attention in Attention and Effort, the graded-interference alternative to the single-channel bottleneck. Wikipedia

Iring Koch

(contemporary). Co-authored the integrative modern review that reconciled the dual-task and task-switching traditions, reframing multitasking cost as a product of cognitive structure, flexibility and plasticity. ORCID

Harold Pashler

(contemporary). Established the central-bottleneck account of the psychological refractory period through the locus-of-slack logic and wrote the authoritative review defining how dual-task interference is measured. Faculty · Wikipedia · Google Scholar

David L. Strayer

(contemporary). Brought the dual-task paradigm to driving research, showing that conversation rather than phone handling impairs control, and identified the rare supertaskers whose dual-task cost is near zero. Faculty · Google Scholar

Glossary

Attentional Capacity.

In Kahneman's model, a single limited pool of undifferentiated processing resource that can be allocated in graded amounts to concurrent tasks; interference arises when their combined demand exceeds the supply.

Automaticity.

The state a skill reaches after extensive practice (overlearning) in which it demands little or no central capacity and so interferes little with a concurrent task.

Capacity Sharing.

A model in which the central stage is not strictly serial but divides a limited capacity between tasks, so that allocating more to one lengthens the processing of the other in proportion; a bridge between the bottleneck and resource accounts.

Central Bottleneck.

A single central stage, usually identified with response selection, that can process only one task at a time, forcing concurrent tasks to queue; the structural explanation of dual-task cost.

Cognitive-Motor Interference.

The degradation of a motor task (typically gait) when a cognitive task is performed at the same time, and vice versa; the basis of the walking-while-talking fall-risk tests.

Divided Attention.

The capacity to distribute processing across more than one input or action simultaneously; the ability a dual-task test is designed to measure.

Dual-Task Cost.

The change in performance on a task from single- to dual-task conditions, usually expressed as a proportional percentage of the single-task baseline; the central measure the paradigm yields.

Locus of Slack.

A logical method for localizing the bottleneck: prolonging a stage of the second task before the bottleneck is absorbed into waiting time and leaves the second response time unchanged, whereas prolonging a stage at or after it is not.

Multitasking.

The attempt to perform two or more tasks concurrently or in rapid alternation; dual-task tests are the laboratory tool for measuring its cost.

Primary Task.

The task of chief interest in a dual-task test, whose performance is measured against its single-task baseline when a secondary task is added.

Psychological Refractory Period.

The slowing of the response to a second stimulus when it follows a first at a short interval; the signature phenomenon of the laboratory dual-task test and the main evidence for a central bottleneck.

Response Selection.

The central operation of choosing which response to make to a stimulus; in single-channel theory the stage that constitutes the bottleneck and can serve only one task at a time.

Secondary Task.

The additional task performed alongside the primary one, often used as a controlled load to test whether an activity draws on central resources.

Single-Channel Theory.

Welford's account in which a central mechanism can carry out only one response-selection operation at a time, so that concurrent tasks must pass through it serially.

Stimulus Onset Asynchrony.

The time between the onset of the first and second stimuli in a psychological-refractory-period task; the shorter the SOA, the larger the delay to the second response.

Supertasker.

A member of the rare minority who show almost no dual-task cost even on demanding task combinations; their existence bears on whether the processing limit is fixed or variable across people.

Frequently Asked Questions

What is a dual-task test?

It is a procedure in which a person performs two tasks at once and the result is compared with performance on each task alone. The drop from the single-task baselines, the dual-task cost, measures how much the two tasks compete for shared processing (Pashler, 1994).

What does the dual-task cost reveal?

It quantifies interference. Expressed as a proportional percentage of the single-task baseline, it lets tasks measured in different units be compared, and because a cost is computed for each task, its pattern shows whether both declined, one was protected, or the two traded off (Plummer & Eskes, 2015).

What is the psychological refractory period?

It is the slowing of the response to a second stimulus when it follows a first too closely. As the gap (the stimulus onset asynchrony) shrinks, the second response is delayed, which is the classic laboratory signature of a central bottleneck (Telford, 1931; Welford, 1952).

Is dual-task cost caused by a bottleneck or by limited capacity?

Both accounts have support. A discrete bottleneck at response selection fits simple choice-reaction tasks, while a graded shared capacity fits richer or continuous combinations; a capacity-sharing model shows the two are limiting cases of one scheme rather than rivals (Tombu & Jolicœur, 2003).

Why does walking while talking matter clinically?

Older adults who stop walking when they begin to talk are at much higher risk of falling, so the walking-while-talking dual-task test is a validated fall-risk screen. The cognitive-motor interference it reveals also tracks neurological decline (Lundin-Olsson et al., 1997; Al-Yahya et al., 2011).

Does using a hands-free phone make driving safe?

Largely no. Dual-task driving studies show the impairment comes from the conversation, not from the manual handling of the phone, so a hands-free set removes little of the risk (Strayer & Johnston, 2001).

Can anyone multitask without a cost?

A rare minority, perhaps one in forty, show almost no dual-task cost even on demanding combinations. These supertaskers suggest the processing limit varies across individuals rather than being a fixed feature of every brain (Watson & Strayer, 2010).

How is the dual-task paradigm used today?

Theoretically it is being integrated with task-switching research under a single account of cognitive structure, flexibility and plasticity; clinically, dual-task gait is being refined into an early marker of cognitive decline with standardized secondary tasks and age norms (Koch et al., 2018; Mustafovska et al., 2025).

References