The Psychological Refractory Period

Abstract

The psychological refractory period is the delay imposed on the second of two responses when the two tasks that require them follow one another too closely. As the interval between the two stimuli shrinks, the second reaction time climbs while the first is left untouched. The regularity is among the most reliable in the timing of human performance, and it is the central evidence for a bottleneck at the stage of selecting a response: perception and motor execution can overlap between tasks, but the choice of what to do cannot. Competing accounts recast the bottleneck as graded capacity sharing or as a strategy the performer adopts, and practice narrows the cost without erasing it. Three interactive demonstrations model the bottleneck, the locus-of-slack logic that locates a stage, and the effect of practice.

Keywords: dual-task interference, response selection, reaction time, cognitive bottleneck

When a person must make two speeded responses to two stimuli that arrive in quick succession, the second response is delayed in direct proportion to how closely the two overlap, a slowing first mapped systematically by Welford and named by him the psychological refractory period (Welford, 1952). The effect is easy to produce, needs only two simple tasks and a variable gap between them, and yields a slope so orderly that it has anchored theories of the mind's central limits for seventy years. Its interpretation bears directly on the broader question of divided attention: whether the apparent inability to do two things at once reflects a fixed structural gate, a resource shared by degree, or a schedule the performer chooses. This article traces the finding from Telford's first report of a refractory phase to the neural and computational models that now account for it.

What the Psychological Refractory Period Is

The psychological refractory period is a slowing of the second response in a pair of overlapping speeded tasks. A participant is given two stimuli, each mapped to its own rapid response, and the experimenter varies the stimulus-onset asynchrony, the time by which the second stimulus follows the first. When that interval is long, the two tasks are performed almost as if each were done alone. When it is short, the response to the first task proceeds normally but the response to the second is delayed, and the delay grows as the interval shrinks. The second task is not performed less accurately so much as later, as though it had been made to wait its turn.

The name is borrowed by analogy from the refractory period of a neuron, the brief interval after firing during which it cannot fire again, but the analogy is loose and the mechanism is cognitive rather than a literal recovery period. What is refractory is not a cell but a central operation that cannot serve two tasks at once. The effect is defined operationally by its signature: the second reaction time rises as the interval falls, with a slope near minus one over the range where the two tasks compete, and the first reaction time stays flat. That pattern, reproduced across an enormous variety of stimulus and response combinations, is what any account of the phenomenon must explain.

The Overlapping-Tasks Paradigm

The experimental arrangement that isolates the effect is the overlapping-tasks, or dual-task, paradigm. Two stimuli, often a tone requiring a keypress and a visual symbol requiring a vocal response, are presented with a controlled asynchrony, and both responses are required as fast as possible. The design deliberately keeps each task simple, so that the interference cannot be blamed on the difficulty of either task in isolation, and it varies only the interval between them. Telford's early study of successive responses first noted that a response made soon after another was slower, a refractory phase in voluntary reactions (Telford, 1931), and Welford turned the observation into a paradigm and a theory by treating the interval as the independent variable (Welford, 1952).

The dependent measure of interest is the second reaction time as a function of the asynchrony, and its shape is diagnostic. Over the range of short intervals it falls along a line of slope near minus one, meaning each reduction of the gap adds an equal amount to the second response; past a knee it levels off, because beyond that point the first task has finished with the shared stage before the second task needs it. The comprehensive review of this literature established that the pattern holds for tasks drawn from different sensory modalities and different response systems, which rules out competition for a particular input channel or a particular set of muscles and points instead to a central limit (Pashler, 1994).

The Response-Selection Bottleneck

The dominant account is the response-selection bottleneck. It divides each task into three successive stages: a perceptual stage that identifies the stimulus, a central stage that selects the response, and a motor stage that executes it. The perceptual and motor stages can run in parallel across the two tasks, but the central stage can serve only one task at a time. When the second stimulus arrives while the first task still occupies the central stage, the second task's response selection cannot begin; it waits until the stage is free. That enforced wait is the psychological refractory period, and because a shorter interval means a longer wait, the model predicts the observed one-for-one slope directly (Welford, 1952; Pashler, 1994). Figure 1 plots the predicted second reaction time against the interval, and the demonstration that follows lets the reader vary the interval and read off the resulting delay.

The Locus-of-Slack Logic

The bottleneck model makes a sharp, testable prediction about where in the second task a manipulation will show its effect, and the test is the locus-of-slack logic. If a stage of the second task that precedes the bottleneck is prolonged, the extra time can be absorbed during the wait the second task already spends idling, so at a short interval the manipulation has no effect on the reaction time; the added time is hidden inside the slack. If instead a stage that follows the bottleneck is prolonged, there is no slack to absorb it, and the extra time passes through to the reaction time at every interval. The dissociation places a stage before or after the bottleneck without any independent measure of its duration (Pashler & Johnston, 1989).

The prediction has been confirmed repeatedly. Degrading the second stimulus, a perceptual manipulation that lengthens an early stage, has little effect on the second reaction time at short intervals but a full effect at long ones, exactly as absorption into slack requires; manipulations of response selection or execution add to the reaction time regardless of interval (Pashler & Johnston, 1989). This under-additivity at short intervals is among the strongest pieces of evidence for a discrete central stage, because it is difficult to reconcile with a model in which the two tasks simply share a graded resource with no privileged serial step. The second demonstration implements the logic, letting the reader add time before or after the bottleneck and switch the interval between short and long to see which delay survives.

Capacity Sharing and Strategic Alternatives

The strict bottleneck is not the only account, and two families of alternative challenge it. The first replaces the all-or-none serial gate with graded capacity sharing: rather than one task waiting while the other holds the central stage, a limited central resource is divided between them, so both proceed at once but each more slowly. A central capacity-sharing model can reproduce the refractory slope by allocating most of the resource to the first task early on, and it accommodates findings, such as occasional effects of the second task on the first, that a strict bottleneck does not predict (Tombu & Jolicoeur, 2003). The debate between queuing and sharing turns on subtle features of the reaction-time distributions, and a critical evaluation concluded that no single-bottleneck account fits every result, though the data equally resist a pure sharing view (Navon & Miller, 2002).

The second alternative locates the limit not in the architecture but in the performer's strategy. The executive-process account embodied in the EPIC architecture holds that there is no structural bottleneck at all; the apparent serial constraint is a cautious scheduling policy that participants adopt to avoid errors, and with suitable instructions and incentives it can be abandoned (Meyer & Kieras, 1997). A related executive-control theory casts dual-task coordination as the deliberate management of task sets, in which the cost of overlap reflects the control operations needed to hold and switch between two task configurations rather than a fixed gate (Logan & Gordon, 2001). The strongest evidence for this view is the demonstration that a few participants, given extensive training and tasks that do not force a response conflict, can achieve virtually perfect time-sharing with no dual-task cost, which a fixed bottleneck should forbid (Schumacher et al., 2001). Defenders of the bottleneck reply that such cases are rare and depend on special conditions, and that the more parsimonious reading of the typical result is a structural limit rather than a strategy (Ruthruff, Pashler, & Klaassen, 2001).

Practice and the Residual Cost

If the refractory period were merely a strategy, sufficient practice should abolish it; if it were structural, practice might shrink it but should leave a residue. The evidence favours the second outcome. Extended training reduces the size of the psychological refractory period substantially, as both tasks and especially their central stages speed up, but a measurable cost typically survives even after many sessions (Van Selst et al., 1999). The reduction is real and sometimes large, so practice matters, but the failure to reach zero in most participants is the finding that keeps the structural account alive. The third demonstration models this course, letting the reader move from an unpracticed to a highly trained performer and watch the cost fall toward a floor rather than to nothing.

Why practice reduces the interference at all has itself been dissected. The speed-up is not uniform across stages; practice compresses the central operation that constitutes the bottleneck, so the second task waits less even though it still waits (Ruthruff, Johnston, & Van Selst, 2001). On this reading the residual cost is the irreducible remainder of a central stage that has been made fast but not instantaneous and not parallel. The account connects the refractory period to the wider literature on automatic processes, in which practice moves a task from slow, capacity-limited control toward fast, low-cost execution without necessarily making it fully parallel with everything else.

The Neural Basis

The bottleneck acquired a specific neural substrate when time-resolved imaging tracked where in the brain two overlapping tasks are forced into a queue. Using rapid event-related functional magnetic resonance imaging to separate the responses to the two tasks, researchers found that regions of lateral prefrontal cortex represented the two tasks serially rather than in parallel, holding the second task's central processing until the first had cleared, precisely the profile a central bottleneck predicts (Dux et al., 2006). The finding gave the abstract central stage a location and showed that the queuing is visible in the brain, not merely inferred from reaction times, and it fits the broader survey of the brain's information-processing limits (Marois & Ivanoff, 2005).

Complementary work characterized the temporal dynamics of the bottleneck and tied it to conscious, serial processing. By parsing a cognitive task into its component durations, imaging and modelling localized a stage whose engagement by one task excludes the other and traced how the second task's central processing is postponed (Sigman & Dehaene, 2005). A later study distinguished the brain regions that operate serially during dual-task performance from those that continue to run in parallel, sharpening the claim that only a specific central operation is refractory (Sigman & Dehaene, 2008). Converging evidence identified a common frontoparietal network as a unified attentional bottleneck shared across otherwise different tasks (Tombu et al., 2011). The refractory period thus became one of the clearest behavioural windows onto the cognitive control functions of the prefrontal cortex.

Applications

The psychological refractory period is not confined to the laboratory, and its most consequential application is to divided attention in the real world. When a driver who is steering and monitoring the road must also respond to a second demand, such as a phone conversation or a warning signal, the central bottleneck predicts that one of the responses will be postponed even if neither eyes nor hands are occupied by the other task. A study of braking while conversing found exactly this central interference: the additional task delayed responses to driving events in a way that could not be explained by the hands or eyes being busy, implicating the same response-selection stage the refractory period isolates (Levy et al., 2006). The result is a caution against the intuition that hands-free multitasking is safe, because the limit is central rather than peripheral.

The same logic bears on the design of any setting in which people must respond to closely spaced signals, from cockpits to control rooms to human-computer interfaces. Where two decisions may be required within a few hundred milliseconds of one another, the second will be delayed, and the delay is a property of the operator's central architecture rather than a lapse that training or motivation can wholly remove. Understanding the refractory period therefore informs how tasks are sequenced, how alarms are timed, and how much can safely be asked of an operator at once, a practical extension of the study of divided attention.

Measurement

Quantifying the psychological refractory period rests on the function relating the second reaction time to the interval between the stimuli, and the informative quantities are read from that function rather than from any single condition. The slope of the falling limb indexes the strength of the serial constraint: a slope near minus one signals a full bottleneck, while a shallower slope implies partial sharing or partial overlap of the central stage. The location of the knee estimates how long the first task occupies the shared stage, and the height of the flat portion recovers the second task's single-task baseline. Table 1 lists the principal measures and what each reveals.

What the Effect Does and Does Not Show

The psychological refractory period is often taken as proof of a single, fixed bottleneck that limits all mental work, and that reading overreaches. The effect demonstrates a serial constraint on response selection under the specific conditions of two speeded, forced-pace tasks, but it does not license the conclusion that a lone gate governs cognition generally. The capacity-sharing models fit much of the same data with a graded resource rather than a discrete gate (Tombu & Jolicoeur, 2003), and the demonstrations of near-perfect time-sharing show that the constraint can be loosened under the right training and task structure (Schumacher et al., 2001). The honest summary is that a central serial limit is the default for these tasks, not an inviolable law of the mind.

A second caution concerns generality across tasks and populations. The size of the refractory period depends on the tasks paired, their difficulty, the compatibility of their stimulus-response mappings, and the practice a performer brings, so a number obtained in one setting does not transfer cleanly to another (Pashler, 1994). This sensitivity is a strength for the analysis of central processing, which is genuinely configured by task demands, but a liability for anyone who wants a single dual-task cost to stand for a person's general capacity for multitasking. The effect is best understood as a controlled probe of the central stage of processing, revealing when and how response selection must be serialized, rather than as a fixed measure of how much a mind can do at once.

Worked Example

Consider the bottleneck model behind the first demonstration, with its default timing. The second task's perceptual stage takes 150 ms and can run in parallel; the first task occupies the central stage until 500 ms after the first stimulus; and the second task's own central selection and response together take 400 ms. The second reaction time is the sum of that 400 ms and however long the second task must wait for the central stage, which is the larger of its own perceptual time, 150 ms, and the time until the stage is free, 500 ms minus the interval. In symbols the second reaction time is 400 plus the greater of 150 and 500 minus the interval, and the single-task baseline, with no wait beyond perception, is 150 plus 400, or 550 ms.

Set the interval to 100 ms. The stage is free at 500 minus 100, or 400 ms, which exceeds the 150 ms perceptual time, so the second task waits until 400 ms and the reaction time is 400 plus 400, or 800 ms, a slowing of 250 ms above the 550 ms baseline. Lengthen the interval to 200 ms and the stage is free at 300 ms, still above the 150 ms perceptual time, so the reaction time is 400 plus the greater of 150 and 300, or 700 ms, a slowing of 150 ms. Each 100 ms added to the interval has removed 100 ms from the second reaction time, the one-for-one slope that defines the effect. The wait vanishes when 500 minus the interval drops to the perceptual time of 150 ms, at an interval of 350 ms; beyond that knee the reaction time sits at its 550 ms baseline. The demonstration lets the reader confirm that the second reaction time tracks this line exactly and flattens at the knee.

Discussion

The psychological refractory period has kept a central place in the study of attention because it turns an abstract question, whether the mind can truly do two things at once, into a measurable slope. The answer it returns is qualified: for two speeded tasks that overlap in time, the selection of a response is serialized, so the second response waits, and the wait is as long as the overlap. That much is among the most reproducible findings in experimental psychology. What the wait reveals about the underlying architecture has been contested at successively finer grains, from Welford's serial gate to graded capacity sharing to strategic scheduling to a frontoparietal network that queues tasks in real time, and each account has had to reproduce the same orderly function that Welford first described.

The trajectory shows a familiar retreat from the language of fixed faculties toward dynamics. The early bottleneck was a structural given; the mature picture is of a central operation that is fast but not instantaneous, largely serial but loosenable with practice and favourable task structure, and realized in prefrontal circuitry that can queue or share depending on what the tasks demand. The residual cost that survives thousands of trials is the durable core of the phenomenon, and the cases of near-perfect time-sharing mark its outer limit. Read this way, the refractory period is not a wall the mind runs into but a description of how a limited central stage schedules competing demands, and its dependence on task, practice, and instruction is not noise around a constant but the very thing the paradigm exists to expose.

Glossary

Additive-factors method.

The inference of processing stages from whether the effects of two manipulations add or interact, the logic underlying the locus-of-slack test.

Bottleneck.

A processing stage that can serve only one task at a time, forcing a second task that needs it to wait until the first has finished.

Capacity sharing.

An account in which a limited central resource is divided by degree between two tasks, so both proceed at once but each more slowly, as an alternative to a serial gate.

Central stage.

The response-selection step between perception and motor execution, identified as the locus of the bottleneck in dual-task performance.

Dual task.

A procedure requiring two speeded responses to two stimuli presented close together in time, the arrangement that produces the refractory period.

EPIC.

A computational cognitive architecture in which dual-task limits arise from executive scheduling strategies rather than a fixed structural bottleneck.

Locus of slack.

The idle wait the second task spends before the bottleneck, into which a prolonged pre-bottleneck stage can be absorbed without cost at short intervals.

Overlapping-tasks paradigm.

The experimental design that presents two tasks with a variable stimulus-onset asynchrony and measures the second reaction time as the interval changes.

Perfect time-sharing.

Dual-task performance with no cost relative to single-task performance, observed rarely and under special training, taken as evidence against a fixed bottleneck.

Prefrontal cortex.

The frontal region whose lateral portions imaging implicates as the site where two overlapping tasks are processed serially.

Psychological refractory period.

The slowing of the second of two responses when the two tasks overlap closely in time, greater the shorter the interval between the stimuli.

Response selection.

The central operation of choosing which response a stimulus requires, proposed to be the stage that cannot serve two tasks at once.

Slope.

The rate at which the second reaction time changes with the interval; a value near minus one over the short-interval range signals a full serial bottleneck.

Stimulus-onset asynchrony.

The interval by which the second stimulus follows the first, the independent variable that governs the size of the refractory period.

Task set.

The configuration of perception and action toward a particular task, which must be maintained for each task in a dual-task pair.

Key Researchers

Alan Welford named the psychological refractory period and, in his 1952 review, turned Telford's early observation of a refractory phase into a single-channel theory of central processing. No verified ORCID, faculty, or Wikipedia page could be confirmed this session for the historical figure, so he is not listed among the linked researchers below; his contribution is documented in the References.

Harold Pashler. Distinguished Professor of Psychology at the University of California, San Diego; wrote the definitive review of dual-task interference and, with Johnston, the locus-of-slack logic that isolates a central bottleneck. ORCID - Faculty Page - Google Scholar - Wikipedia

David E. Meyer. Professor of Psychology at the University of Michigan; with Kieras built the EPIC architecture and the adaptive-executive account, arguing the bottleneck is a strategic schedule rather than a fixed structural limit. Faculty Page - Google Scholar - Wikipedia

Rene Marois. Professor and Chair of Psychology at Vanderbilt University; provided the time-resolved imaging that localized a central bottleneck to lateral prefrontal cortex. ORCID - Faculty Page - Google Scholar

Paul E. Dux. Professor of Psychology at the University of Queensland; first author of the time-resolved fMRI isolation of the central bottleneck and later work on how training reshapes it. ORCID - Faculty Page - Google Scholar

Stanislas Dehaene. Professor at the College de France and director of the NeuroSpin centre; with Sigman characterized the temporal dynamics of the central bottleneck within a global-workspace account of serial processing. ORCID - Faculty Page - Google Scholar - Wikipedia

Eric Ruthruff. Professor of Psychology at the University of New Mexico; showed with Van Selst and Johnston that practice shrinks but does not abolish the refractory period, and dissected why practice reduces central interference. ORCID - Faculty Page - Google Scholar

Pierre Jolicoeur. Professor of Psychology at the Universite de Montreal; with Tombu formulated the central capacity-sharing model that reframes the bottleneck as graded division of a limited resource. Faculty Page - Google Scholar

Frequently Asked Questions

What is the psychological refractory period?
It is the slowing of the second of two responses when the two tasks that require them follow one another too closely; the shorter the interval between the stimuli, the greater the delay (Welford, 1952).

Why is the second response delayed?
Because selecting a response is a central operation that can serve only one task at a time, so the second task must wait for the first to release that stage before its own response can be chosen (Pashler, 1994).

What is stimulus-onset asynchrony?
It is the interval by which the second stimulus follows the first, the variable the experimenter manipulates; as it shortens, the second reaction time rises along a line of slope near minus one (Pashler & Johnston, 1989).

Is the bottleneck structural or strategic?
Both accounts have support; capacity-sharing and strategic-scheduling models fit much of the data, but the survival of a cost after extensive practice favours a largely structural central limit (Meyer & Kieras, 1997; Ruthruff, Pashler, & Klaassen, 2001).

Can practice eliminate the psychological refractory period?
Practice reduces it substantially, and a few trained participants approach perfect time-sharing, but in most people a measurable residual cost survives extended training (Van Selst et al., 1999; Schumacher et al., 2001).

Where in the brain is the bottleneck?
Time-resolved imaging localizes the serial queuing to lateral prefrontal cortex and a shared frontoparietal network, where two overlapping tasks are processed one after the other rather than in parallel (Dux et al., 2006; Tombu et al., 2011).

How does it differ from a general capacity limit?
The refractory period reflects a serial constraint on response selection specifically, revealed by the under-additive locus-of-slack pattern, rather than a uniform slowing of all processing that a single graded resource would produce (Tombu & Jolicoeur, 2003).

Why does it matter for driving?
Because the limit is central, a second task such as a conversation delays responses to the road even when the eyes and hands are free, so hands-free multitasking is not cost-free (Levy et al., 2006).

References