Divided attention is what we call multitasking when we study it carefully: trying to drive while talking, to listen while typing, to do two things that each, alone, would be easy. This page traces the field's long argument over why that fails — is there a single bottleneck, one shared pool of effort, or several separate resources? — and what practice, executive control, and aging do to the cost. Three interactive tasks let you measure your own dual-task limits: a processing bottleneck, the cost of holding a load while you respond, and the price of switching between rules.
Divided attention is the allocation of limited mental resources to two or more tasks that must be carried out at the same time (Kahneman, 1973). The defining scientific fact is the dual-task cost: performing two tasks concurrently is almost always slower and more error-prone than performing either one alone, even when each is simple. The central question is why. Does a single stage of processing form a hard bottleneck that can serve only one task at a time, or is there a graded pool of capacity that the tasks must share, or are there several specialized resources that interfere only when two tasks draw on the same one? The sections below follow that debate from the first studies of the psychological refractory period through Kahneman's capacity theory and Wickens' multiple resources to the modern view that combines a central bottleneck with flexible executive control, and then to what practice, the brain, driving, and aging reveal about the limit.
What Divided Attention Is
Divided attention is the deployment of attention across more than one task or stream of information at the same time, in contrast to selective attention, which prioritizes one stream and suppresses the rest. The everyday word for it is multitasking, but the laboratory treats it more precisely as concurrent performance under measurement: two defined tasks, each with its own stimulus and response, performed together so that the cost of combination can be read off against single-task baselines. That cost is the central datum. When people add a second task, their responses to the first typically slow, their accuracy on one or both falls, or they trade speed on one for accuracy on the other. The size of the cost is not fixed; it grows with the difficulty of each task, with the similarity between the two, and with how little either has been practiced, and it shrinks toward zero in special cases. Explaining that pattern, why two easy tasks can be nearly impossible together while two others combine almost freely, is the work of the theories that follow.
The Dual-Task Method
The experimental engine of the field is the dual-task paradigm. In its simplest form a participant performs Task A alone, Task B alone, and then both together, and the interference is the drop from single- to dual-task performance. A more incisive version is the psychological refractory period (PRP) procedure, in which two stimuli are presented in quick succession, each demanding its own speeded response, and the gap between them, the stimulus-onset asynchrony (SOA), is varied (Welford, 1952; Pashler, 1994). The reliable result is that the response to the second stimulus is slowed when the SOA is short, and the shorter the gap the greater the slowing, as if the second task must wait its turn. A complementary tradition uses continuous dual tasks, such as tracking a moving target while monitoring for tones, and reads off the trade-off between them. Across these methods one fact recurs: combining tasks costs something, and the size and shape of that cost is the evidence every theory must explain. Figure 1 lays out the three accounts that have organized the debate.
The Central Bottleneck and the Refractory Period
The oldest and most precise account is the central bottleneck. Its evidence is the psychological refractory period: when a second stimulus arrives soon after the first, the second response is delayed, and the delay tracks the SOA almost one-for-one, as though a stage of processing for Task 2 cannot begin until the same stage for Task 1 has finished (Telford, 1931; Welford, 1952). Pashler's careful program of experiments localized that stage: perceptual analysis and the final motor execution of the two tasks can proceed in parallel, but the central act of selecting a response, mapping a stimulus onto a decision, is strictly serial and forms the bottleneck (Pashler, 1994). On this view the cost of multitasking is not a vague dilution of effort but a queue: when two tasks need the response-selection stage at once, one must wait. The model predicts the PRP curve quantitatively and explains why the second task slows while the first is largely spared. The demonstration below is an original two-response task that recreates the bottleneck; the gap between the two signals is short on some trials and long on others, and your reaction time to the second signal reveals the queue.
Try It
The Bottleneck
Each trial has two tasks. First a coloured disc appears — press its colour at once. Then an arrow appears — press its direction. The arrow sometimes follows the disc almost immediately and sometimes after a pause. Answer both as fast as you can. There are 12 trials.
Capacity, Effort, and Multiple Resources
A different tradition rejects the idea of a single all-or-none gate in favor of a graded resource. Kahneman proposed that attention is a limited pool of effort or capacity that can be allocated flexibly to whatever the current activity demands, with arousal raising the supply and task difficulty raising the draw; dual-task interference arises when the combined demand exceeds the pool (Kahneman, 1973). This explained graded trade-offs that a fixed bottleneck handles awkwardly, but a single undifferentiated pool predicts that any two demanding tasks should interfere equally, which is false: tasks that use different senses or different response systems often combine far better than tasks that share them. Wickens answered this with multiple-resource theory, which holds that processing draws on several separate pools defined by stage (perception and cognition versus response), by modality (visual versus auditory), and by code (verbal versus spatial), so that two tasks interfere in proportion to how much they overlap on these dimensions (Wickens, 2002; Wickens, 2008). Navon and Gopher had earlier put the economic logic of such shared and separate resources on a formal footing (Navon & Gopher, 1979). The framework predicts that a visual-manual task and an auditory-vocal task can be combined with little cost, and it underwrites workload prediction in cockpit and interface design. The demonstration below measures a dual-task cost directly: you respond to a simple signal alone, and then again while holding a memory load, so you can feel the price of sharing capacity.
Try It
One Task or Two
Press the direction of each arrow as fast as you can. In the first block that is all you do. In the second block three digits flash before each arrow — hold them in mind, answer the arrow, then say whether a probe digit was in the set. 12 arrows in all.
Executive Control and Task-Switching
Doing two things at once is not only a matter of capacity; it requires a controller to schedule the tasks, hold each task's rules in mind, and reconfigure the system when the demand shifts. Much of what feels like multitasking in daily life is in fact rapid switching between tasks, and switching has a measurable price. When people alternate between two simple classifications, responses on a trial that requires a different rule from the last are reliably slower and more error-prone than responses on a trial that repeats the rule, a difference called the switch cost (Rogers & Monsell, 1995). Part of the cost reflects the time to reconfigure the task set, and part reflects interference from the previous set that lingers even when there is ample time to prepare, a residue that points to a passive, carry-over component as well as an active control component (Monsell, 2003). Task-switching thus exposes the executive machinery of divided attention, the same machinery that coordinates genuinely concurrent tasks, and it shows why interleaving jobs is rarely free even when no two responses literally collide. The demonstration below is an original switching task; a cue tells you which rule applies on each trial, and the cost of a switch shows up in your reaction time.
Try It
Switching Tasks
A box above the digit names the current rule. ODD / EVEN means judge whether the digit is odd or even; LOW / HIGH means judge whether it is below or above 5. Always press left for odd or low and right for even or high. The rule keeps changing. There are 14 trials.
Practice and Automaticity
Dual-task costs are not immutable. With extensive practice, one task can become automatic, running with little demand on the central stages, and the cost of pairing it with another can shrink dramatically or vanish. Schneider and Shiffrin drew the governing distinction between controlled processing, which is effortful, capacity-limited, and serial, and automatic processing, which is fast, effortless, and able to run in parallel because it bypasses the controlled bottleneck (Schneider & Shiffrin, 1977; Shiffrin & Schneider, 1977). The most striking evidence that combination can become nearly free came from Spelke, Hirst, and Neisser, who trained two students for months to read stories while writing down dictated words; with practice the students could do both with comprehension, a feat that seems to defy a strict single-bottleneck account (Spelke, Hirst, & Neisser, 1976). Whether such cases reflect true parallelism or extremely efficient switching is debated, but the practical lesson is firm: the cost of divided attention depends on how practiced its components are, which is why an experienced driver can converse on a familiar road yet falls silent at a complex junction.
Divided Attention in the Brain
The behavioral bottleneck has a neural signature. Using time-resolved imaging, Dux and colleagues showed that a region of the lateral prefrontal cortex processes the two tasks of a PRP procedure serially, its response to the second task postponed in step with the behavioral delay, which identifies a central, frontally mediated stage as the locus of the queue (Dux, Ivanoff, Asplund, & Marois, 2006). More broadly, performing two tasks together recruits lateral prefrontal and parietal cortex beyond what either task demands alone, reflecting the added work of coordination, while the anterior cingulate monitors for conflict between competing task sets (Marois & Ivanoff, 2005). The convergence of behavior and imaging supports a hybrid picture: peripheral perceptual and motor processes run in parallel, but a capacity-limited central operation, implemented in frontoparietal control regions, gates the flow, which is why the cost of multitasking is so resistant to good intentions.
Driving and the Costs of Multitasking
The most consequential test bed for divided attention is the road. Strayer and Johnston showed in controlled simulation that talking on a phone impairs driving, slowing responses to signals and increasing missed events, and crucially that the impairment is no less for hands-free than for handheld phones, because the interference is central rather than manual (Strayer & Johnston, 2001). The cost is not a failure to see but a kind of inattention: drivers on a call look at objects yet fail to register and remember them, a pattern Strayer and Drews characterized as conversation consuming the attention that driving needs (Strayer & Drews, 2007). The everyday belief in costless multitasking fares no better in the laboratory: Ophir, Nass, and Wagner found that people who habitually juggle many media streams were actually worse at filtering distraction and switching tasks, not better, undercutting the idea that practice at multitasking builds a general multitasking skill (Ophir, Nass, & Wagner, 2009). The applied moral is consistent with the theory: when two tasks compete for the central stage, sincerity and effort do not buy parallelism.
Aging and Divided Attention
Divided attention is among the abilities most sensitive to healthy aging. Older adults typically show larger dual-task and switch costs than younger adults, and a meta-analytic synthesis by Verhaeghen and Cerella found that while some apparent age effects on executive tasks reduce to a general slowing, the costs of coordinating and switching between tasks carry an additional age-related penalty over and above that slowing (Verhaeghen & Cerella, 2002). The deficit is not uniform, and it is partly trainable: Kramer, Larish, and Strayer showed that older adults can improve their dual-task performance with practice at allocating attention between tasks, narrowing though not erasing the gap with younger adults (Kramer, Larish, & Strayer, 1995). The aging evidence thus does double duty, confirming that the executive coordination of divided attention is a distinct, frontally dependent capacity, and showing that it retains some plasticity across the lifespan.
Comparing the Accounts
Table 1 sets the major accounts of the dual-task limit side by side on the questions that divide them.
Table 1
Theories of Divided Attention Compared
| Account | Nature of the limit | Predicts no cost when… | Signature evidence |
|---|---|---|---|
| Central bottleneck (Welford; Pashler) | A single serial stage: response selection | Tasks never need the central stage at the same time | The PRP: Task 2 slows as the gap to Task 1 shortens |
| Single capacity (Kahneman) | One graded pool of effort, flexibly allocated | Combined demand stays below the pool's supply | Graded trade-offs that rise with task difficulty and arousal |
| Multiple resources (Wickens) | Several pools by stage, modality, and code | Tasks draw on non-overlapping resources | Visual–manual and auditory–vocal tasks combine with little cost |
| Executive control (Monsell) | Cost of configuring and switching task sets | The same task set repeats with no reconfiguration | Switch costs: slower responses when the rule changes |
Note. The accounts are not mutually exclusive; the modern consensus combines a central bottleneck with graded, partly separable resources and an executive controller (Pashler, 1994; Wickens, 2008).
Criticisms and Open Questions
The central dispute is whether the bottleneck is a fixed structural feature or a strategic one. Strict bottleneck theorists hold that response selection is inherently serial, but others argue that the apparent bottleneck is a flexible allocation policy: when people are encouraged to share, and the tasks are compatible and well practiced, dual-task costs can fall close to zero, which a hard structural gate should not allow (Navon & Gopher, 1979). The extreme cases of trained near-parallelism sharpen the problem, since a single immovable stage cannot easily accommodate two people reading and writing at once (Spelke, Hirst, & Neisser, 1976); defenders reply that such performances may be very fast switching rather than true simultaneity, a distinction that behavior alone struggles to settle. A second open question is how cleanly the proposed multiple resources can be individuated, since the dimensions of stage, modality, and code are useful for prediction but hard to pin to fully separate mechanisms. The honest summary is that no single account suffices: a central, capacity-limited operation is real and frontally implemented, yet its severity is modulated by resource overlap, practice, and strategy, and the boundary between a structural limit and a controllable one remains the live question.
Worked Example
Consider a driver merging onto a busy motorway while taking a hands-free call. The visual-manual task of steering and the auditory-vocal task of conversing draw on largely separate resources, so on an open road they combine with little visible cost, exactly as multiple-resource theory predicts (Wickens, 2002). As the merge tightens, the central demand spikes: judging gaps, selecting a lane, and timing the acceleration all need the response-selection stage at once, and now the call competes for that single stage, so the driver's replies lag, a hazard is registered late, and the conversation may stall entirely, the behavioral mark of the bottleneck (Pashler, 1994; Strayer & Johnston, 2001). If the driver has done this merge ten thousand times, much of it runs automatically and the cost shrinks; if the junction is unfamiliar, controlled processing reclaims the stage and the talking stops (Schneider & Shiffrin, 1977). The single episode contains all three accounts at once, which is why each captures part of the truth: separate resources when the tasks do not overlap, a hard bottleneck when they do, and practice deciding which regime applies.
Why It Matters
Divided attention matters because modern life constantly demands it and human capacity quietly refuses. The stakes are most visible on the road, where the evidence that even hands-free conversation degrades driving has reshaped how we think about distraction and law (Strayer & Johnston, 2001). The same limit governs cockpits, control rooms, operating theatres, and classrooms, anywhere a person must track several streams and a missed signal is costly, and multiple-resource theory turns the science into design guidance by predicting which task pairings will collide and which will coexist (Wickens, 2008). The deeper lesson cuts against intuition: the feeling of effortless multitasking is largely an illusion produced by fast switching and by tasks that happen not to overlap, and the people most confident in their multitasking are not measurably better at it (Ophir, Nass, & Wagner, 2009). Understanding where the limit lies, at a central stage that can be shared but not duplicated, is the first step to designing tasks, tools, and habits that work with human capacity instead of pretending it is unlimited, and the demonstrations above are small, direct measurements of that limit in your own attention.
Key Researchers
Daniel Kahneman (1934–2024). Nobel laureate whose Attention and Effort recast attention as a limited, flexibly allocated pool of capacity, providing the capacity-theory alternative to the fixed bottleneck.
A. T. Welford (1914–1995). Formulated the single-channel theory of the psychological refractory period, arguing that a central decision mechanism can handle only one response selection at a time.
Harold Pashler. Distinguished Professor of Psychology at the University of California, San Diego; his locus-of-slack experiments localized the dual-task bottleneck to the response-selection stage. UC San Diego · Google Scholar · ORCID.
Christopher D. Wickens. Professor Emeritus at the University of Illinois Urbana-Champaign; developed multiple-resource theory and applied it to workload prediction in aviation and human factors. Illinois faculty · Google Scholar · ORCID.
Richard M. Shiffrin. Distinguished Professor of Psychological and Brain Sciences at Indiana University; with Walter Schneider drew the controlled-versus-automatic distinction that explains how practice reduces dual-task cost. Indiana University · Google Scholar · ORCID.
David L. Strayer. Professor of Cognition and Neural Science at the University of Utah; his simulator studies established that phone conversation impairs driving even hands-free. University of Utah · Google Scholar · ORCID.
Arthur F. Kramer. Professor of Psychology at Northeastern University and the University of Illinois; studies attentional control and dual-task training, including its plasticity in older adults. Northeastern University · Google Scholar · ORCID.
Key Terms
| Term | Definition |
|---|---|
| Divided attention | The allocation of limited mental resources to two or more tasks performed at the same time. |
| Dual-task cost | The drop in speed or accuracy when two tasks are performed concurrently rather than alone. |
| Psychological refractory period (PRP) | The slowing of the response to a second stimulus when it closely follows a first, increasing as the gap shrinks. |
| Stimulus-onset asynchrony (SOA) | The time between the onset of the first and second stimuli in a PRP procedure. |
| Central bottleneck | A single processing stage, response selection, that can serve only one task at a time. |
| Response selection | The stage that maps a stimulus onto a chosen response, identified as the locus of the bottleneck. |
| Capacity theory | Kahneman's account of attention as a single, graded pool of effort allocated flexibly across tasks. |
| Multiple-resource theory | Wickens' account in which separate resource pools by stage, modality, and code determine interference. |
| Controlled processing | Effortful, capacity-limited, serial processing that depends on the central stage. |
| Automatic processing | Fast, effortless, parallel processing that develops with practice and bypasses the bottleneck. |
| Switch cost | The slowing and added errors when a trial requires a different task set from the preceding one. |
| Task set | The configuration of rules and mappings needed to perform a particular task. |
| Executive control | The processes that schedule, coordinate, and reconfigure tasks during divided attention. |
| Time-sharing | Performing two tasks together by sharing or alternating limited resources between them. |
Frequently Asked Questions
What is divided attention?
Divided attention is the attempt to allocate limited mental resources to two or more tasks at the same time, as in driving while conversing. It is studied through the dual-task cost, the reliable finding that concurrent performance is slower or less accurate than performing either task alone (Kahneman, 1973).
What is the difference between divided and selective attention?
Selective attention prioritizes one stream of information and suppresses competing input, whereas divided attention spreads attention across two or more tasks that must be done together. Selective attention asks what gets through; divided attention asks how much two tasks cost each other when combined.
What is the psychological refractory period?
The psychological refractory period (PRP) is the delay in responding to a second stimulus when it follows closely on a first; the response slows as the gap between them shrinks. It is the key evidence for a central bottleneck at the response-selection stage (Pashler, 1994).
What is the bottleneck theory of attention?
Bottleneck theory holds that a single processing stage, the selection of a response, can serve only one task at a time, so when two tasks need it together one must wait. Pashler's experiments localized this serial stage and showed that perception and motor execution can run in parallel around it (Welford, 1952; Pashler, 1994).
What is multiple-resource theory?
Multiple-resource theory proposes that processing draws on several separate pools defined by stage, sensory modality, and code, so two tasks interfere in proportion to how much they share. It predicts that a visual-manual task and an auditory-vocal task can be combined with little cost (Wickens, 2002).
Can people really multitask?
Rarely, and not as well as they think. Most apparent multitasking is fast switching between tasks, which carries a measurable switch cost, and genuine parallel performance is limited to cases where tasks use separate resources or one is highly automatic. People who multitask most are not better at it (Monsell, 2003; Ophir, Nass, & Wagner, 2009).
Why is talking on a hands-free phone still dangerous while driving?
Because the interference is central, not manual. Conversation competes for the same response-selection and attentional resources that driving needs, so drivers look at hazards yet fail to register them, and hands-free phones are no safer than handheld ones in controlled studies (Strayer & Johnston, 2001; Strayer & Drews, 2007).
Does divided attention change with age?
Yes. Older adults show larger dual-task and switch costs than younger adults, beyond what general slowing predicts, reflecting an age-related decline in executive coordination. The deficit is partly trainable, narrowing with practice at allocating attention between tasks (Verhaeghen & Cerella, 2002; Kramer, Larish, & Strayer, 1995).
References
| 1 | Dux, P. E., Ivanoff, J., Asplund, C. L., & Marois, R. (2006). Isolation of a central bottleneck of information processing with time-resolved fMRI. Neuron, 52(6), 1109–1120. https://doi.org/10.1016/j.neuron.2006.11.009 |
| 2 | Kahneman, D. (1973). Attention and effort. Prentice-Hall. |
| 3 | Kramer, A. F., Larish, J. F., & Strayer, D. L. (1995). Training for attentional control in dual task settings: A comparison of young and old adults. Journal of Experimental Psychology: Applied, 1(1), 50–76. https://doi.org/10.1037/1076-898X.1.1.50 |
| 4 | Marois, R., & Ivanoff, J. (2005). Capacity limits of information processing in the brain. Trends in Cognitive Sciences, 9(6), 296–305. https://doi.org/10.1016/j.tics.2005.04.010 |
| 5 | Monsell, S. (2003). Task switching. Trends in Cognitive Sciences, 7(3), 134–140. https://doi.org/10.1016/S1364-6613(03)00028-7 |
| 6 | Navon, D., & Gopher, D. (1979). On the economy of the human-processing system. Psychological Review, 86(3), 214–255. https://doi.org/10.1037/0033-295X.86.3.214 |
| 7 | 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 |
| 8 | 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 |
| 9 | Rogers, R. D., & Monsell, S. (1995). Costs of a predictable switch between simple cognitive tasks. Journal of Experimental Psychology: General, 124(2), 207–231. https://doi.org/10.1037/0096-3445.124.2.207 |
| 10 | Schneider, W., & Shiffrin, R. M. (1977). Controlled and automatic human information processing: I. Detection, search, and attention. Psychological Review, 84(1), 1–66. https://doi.org/10.1037/0033-295X.84.1.1 |
| 11 | Shiffrin, R. M., & Schneider, W. (1977). Controlled and automatic human information processing: II. Perceptual learning, automatic attending and a general theory. Psychological Review, 84(2), 127–190. https://doi.org/10.1037/0033-295X.84.2.127 |
| 12 | Spelke, E., Hirst, W., & Neisser, U. (1976). Skills of divided attention. Cognition, 4(3), 215–230. https://doi.org/10.1016/0010-0277(76)90018-4 |
| 13 | 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 |
| 14 | Strayer, D. L., & Drews, F. A. (2007). Cell-phone–induced driver distraction. Current Directions in Psychological Science, 16(3), 128–131. https://doi.org/10.1111/j.1467-8721.2007.00489.x |
| 15 | Telford, C. W. (1931). The refractory phase of voluntary and associative responses. Journal of Experimental Psychology, 14(1), 1–36. https://doi.org/10.1037/h0073262 |
| 16 | Verhaeghen, P., & Cerella, J. (2002). Aging, executive control, and attention: A review of meta-analyses. Neuroscience & Biobehavioral Reviews, 26(7), 849–857. https://doi.org/10.1016/S0149-7634(02)00071-4 |
| 17 | Welford, A. T. (1952). The "psychological refractory period" and the timing of high-speed performance—a review and a theory. British Journal of Psychology, 43(1), 2–19. https://doi.org/10.1111/j.2044-8295.1952.tb00322.x |
| 18 | Wickens, C. D. (2002). Multiple resources and performance prediction. Theoretical Issues in Ergonomics Science, 3(2), 159–177. https://doi.org/10.1080/14639220210123806 |
| 19 | Wickens, C. D. (2008). Multiple resources and mental workload. Human Factors, 50(3), 449–455. https://doi.org/10.1518/001872008X288394 |
The three interactive figures on this page — the bottleneck, dual-task, and task-switching demonstrations — generate their trials and compute their reaction times live in your browser; no experimental dataset is bundled with the page. The empirical claims in the text are sourced to the references above.