Multitasking: The Switch Cost Your Brain Pays Every Time You Change Tasks

Introduction

In 1995 two psychologists at the University of Cambridge did something generous for the people in their experiment. They gave them a head start.

The task was simple. A pair of characters appeared on a screen, a letter and a digit. Sometimes you had to say whether the digit was odd or even. Sometimes you had to say whether the letter was a vowel or a consonant. The two jobs came in a fixed rhythm, two of one and then two of the other, so everybody knew exactly when the change was coming.

Robert Rogers and Stephen Monsell then stretched the pause before each trial. As it grew to about 0.6 seconds, the extra time needed to change jobs shrank, because people could get ready in advance. And then it stopped shrinking. Even with 1.2 seconds to prepare, a large cost was still there on the first trial of the new task, and only on that trial [1].

More than a second. For a task a child could do. Known in advance.

That leftover cost is the most interesting fact about multitasking, and you will almost never read about it. What you will read instead are big round numbers: 40 percent of your day wasted, 23 minutes to refocus, 2 percent of people who can really do it. One of those is not in the paper it is credited to.

This article goes through what was actually measured. It separates switching between tasks from doing two at once, because they fail in different ways, and it follows the famous Stanford finding about heavy media multitaskers through a decade of replications.

1927: A Stopwatch and Two Kinds of List

The first switching experiment was done with paper and a stopwatch.

In 1927 A. T. Jersild, a graduate student at Columbia, gave people columns of numbers. On some lists they did one thing all the way down, adding 6 to every number, say. On others they had to alternate, adding 6 to one number and subtracting 3 from the next. The alternating lists took noticeably longer, even though every single sum was just as easy as before. Jersild published it as a monograph called Mental Set and Shift, and it has no DOI because it predates them.

He also noticed something the field would come back to decades later. When the two tasks used different kinds of material, so that the item itself told you what to do, alternating cost little or nothing.

Hold on to both halves of that. Switching costs you when the situation does not tell you which rules apply. It costs much less when the world does the reminding for you.

The Waiting Room in Your Head

Four years later a different problem turned up in a different lab. It looks like the same problem. It is not.

In 1931 C. W. Telford asked people to respond to two signals that came very close together, and the second response came out slower than it should have, as if the system needed a moment to recover after the first one, and he called this a refractory phase [2]. In 1952 the British psychologist A. T. Welford gave it the name that stuck, the psychological refractory period, and proposed that the mind has a single channel that can only make one decision at a time [3].

By 1994 Harold Pashler could review four decades of these experiments and describe a "stubborn bottleneck" at the stage where you choose what to do in response [4]. Seeing two things at once is fine. Moving two hands at once is often fine. Deciding two things at once is where the queue forms.

So there are two ways to multitask, and they hit two different limits.

You can alternate: write a sentence, check a message, write a sentence. Here the cost is the switch itself, the work of swapping one set of rules for another. Or you can overlap: listen to a podcast while you read, talk while you drive. Here the cost is the queue, because anything that needs a decision has to wait its turn.

Most advice blurs these together. Batching your email helps with the first problem and does nothing for the second. Start with the switch.

The Head Start That Was Not Enough

Go back to the Cambridge experiment, because the design was cleverer than it looks.

Earlier studies compared whole alternating blocks with single-task blocks, the way Jersild did. But an alternating block is harder in two ways: you change rules, and you also hold two sets of rules in mind the whole time. The 1995 rhythm of two and two let Rogers and Monsell compare a switch trial with a repeat trial inside the same block, with the same memory load [1].

That comparison gave the field its basic unit. The switch cost is the extra time and the extra errors on the trial where the task changes, compared with the trial where it stays the same. The same comparison exposed a puzzle. If the cost were only about getting ready, enough time should erase it. It did not. The part that survived a full 1.2 seconds of warning appeared only once the new stimulus was on the screen, as if some of the change could only be finished when there was something real to act on.

Other labs pushed at this. In 2000 Nachshon Meiran and colleagues showed people a cue before each trial saying which task was next, and found the switch cost fell sharply as the gap between cue and target grew [5]. Preparation is real. You can do some of the work ahead of time. But not all of it, and in 2002 Sander Nieuwenhuis and Monsell offered a humbling explanation of why. Their model fitted the data from two experiments by assuming that on some trials people simply fail to use the preparation time they were given [6]. The residual cost, on this reading, is partly an average of good trials and lapses. You could get fully ready. You just do not, every time.

Voluntary switching looks the same. In a 2004 study where people chose for themselves which of two tasks to do next, Catherine Arrington and Gordon Logan measured a switch cost of 310 milliseconds when the pause between trials was 100 milliseconds, falling to 94 milliseconds when the pause was a full second [7]. Even with free choice, people also repeated the same task more often than chance. Left alone, you avoid switching. You seem to know it costs you.

Where the Old Task Hangs On

If you pull the switch cost apart, it turns out to be several costs stacked together, and they interact rather than simply adding up.

Part of it is setting up the new rules. In 2001 Joshua Rubinstein, David Meyer and Jeffrey Evans ran four experiments and found that switch costs grew with the complexity of the rules being swapped in, and modelled the switch as two stages, one that changes the goal and one that loads the rules for it [8]. A 2024 study found that the time cost scales with how different the two sets of rules are: the further apart the tasks, the slower the first response [9].

But the new task is not being written onto a blank page. In 2000 Glenn Wylie and Alan Allport showed that the task you just left keeps pulling. Their switch costs depended mainly on the task being switched from, not the one being switched to, which is what you would expect if the old rules linger and have to be fought off [10]. They called this task-set inertia. You have met it. It is the moment you reply to a work email in the tone of the text message you just sent. The mind also fights back in the other direction. Ulrich Mayr and Steven Keele reported in 2000, across six experiments, that going back to a task you abandoned a moment ago is slower than moving to a third task you have not recently used [11]. Doing A, then B, then A again costs more than A, then B, then C. The explanation is that leaving a task involves actively suppressing it, and that suppression has not worn off when you try to return. They called it backward inhibition.

Put those together and a morning of short hops gets expensive. Every return trip pays for where you just were and for the brake you put on where you are going.

Two large reviews reached the same broad verdict. In 2010 Andrea Kiesel and colleagues, and separately André Vandierendonck and colleagues, concluded that the switch cost reflects both active reconfiguration and interference from what came before, with longer preparation reliably shrinking it and never quite eliminating it [12] [13].

Is It Really a Switch at All?

Not everyone accepted that anything was being reconfigured.

In 2003 Gordon Logan and Claus Bundesen pointed out an awkward confound: in most cued experiments a new task also meant a new cue on the screen, and when they separated the two, much of the apparent switch cost went with the cue [14]. Maybe people were not rebuilding their mental rules at all. Maybe they were just slower to read an unfamiliar cue, and what looked like a switch cost was really a cue-repetition benefit.

Monsell and Guy Mizon answered in 2006 with six experiments using two different cues for each task, so the cue could change while the task stayed put [15]. Their first experiment reproduced the sceptics' result. The next two found a substantial task-switch cost that shrank with preparation. The difference came down to how often the task changed. When switches were rare enough that people did not bother getting ready until they had to, the evidence for a genuine reconfiguration step was clear.

Most researchers now accept both: part of the cost is memory priming, part is a real act of changing course, and the split depends on the task.

What the Brain Is Doing

Brain imaging agrees that there is more than one thing going on.

In 2003 Todd Braver and colleagues used fMRI to separate the two costs the behavioural work had identified. One is transient, the spike on the trial where the task actually changes. The other is sustained, the background load of being in a block where a change could come at any moment. Different regions tracked each of the two, a double dissociation suggesting that switching and being ready to switch are separate jobs [16]. A 2021 EEG study of 197 participants, run after enough practice to make the tasks routine, found the same split in brain rhythms: switch costs and mixing costs were each linked to their own network of theta-band activity [17].

The practical reading is simple. Living in a state where you might be interrupted costs you something even on the minutes when you are not. The mixing cost is the price of the open door.

Two Things at Once

Now the other kind of multitasking, the overlap.

The most famous demonstration that practice can loosen the bottleneck comes from 1976. Elizabeth Spelke, William Hirst and Ulric Neisser trained two students to read short stories while writing down words that were being dictated to them [18]. At first their reading collapsed. After months of daily sessions they read at their normal speed and understood what they read, while taking dictation. Two people, not a sample, but a real existence proof.

You know this already. Walking and talking stopped competing long ago, because walking no longer needs decisions. That is the process our piece on how habits become automatic describes: a routine that once needed attention hands itself over to a system that does not.

Whether the bottleneck can ever vanish is still argued. David Meyer and David Kieras built a computational model in 1997 in which the mind can apply the rules for two tasks at the same time, and the limits sit in the eyes, hands and voice rather than in a central decision stage [19]. In 2001 Eric Schumacher and colleagues reported that after fairly modest practice, at least some participants reached "virtually perfect time sharing" on two simple choice tasks [20].

Defenders of a hard bottleneck reply that those tasks were unusually easy and used separate senses and hands, so they may have slipped past the bottleneck rather than removed it. Mariano Sigman and Stanislas Dehaene tried to reconcile the two traditions in 2006 by running dual-task and switching conditions together, and argued that both a passive queue and an active control stage are involved [21]. Dario Salvucci and Niels Taatgen's 2008 threaded cognition model takes a middle road: several streams of thought can run at once, but they share one step-by-step procedural resource, and that is where they collide [22]. Christopher Wickens, whose multiple resource theory shaped aviation and car design, puts the everyday version plainly in a 2021 review: tasks interfere most when they compete for the same kind of resource, such as two visual tasks or two verbal ones [23].

So yes, you can do two things at once, if one of them is thoroughly practised and they use different channels. Two new things that both need thought will queue.

The Phone Call in the Car

Nowhere does that matter more than at the wheel.

In 2001 David Strayer and William Johnston put people in a driving simulator and gave them things to listen to. Radio broadcasts and audiobooks caused no measurable harm. Neither did simply repeating back what was said on a handheld phone. A real conversation did. Whether the phone was handheld or hands-free, people missed twice as many simulated traffic signals and reacted more slowly to the ones they caught [24].

The problem was never holding the phone. It was that a conversation keeps asking you to decide what to say next.

In 2006 Strayer, Frank Drews and Dennis Crouch compared the same drivers when they were talking on a phone and when they had drunk enough alcohol to reach a blood alcohol level of 0.08 percent [25]. The patterns differed. Drunk drivers followed closer and braked harder. Phone drivers braked later and had more collisions. The authors concluded that, with driving conditions controlled, the impairment from phoning "can be as profound" as driving drunk. A 2006 meta-analysis by William Horrey and Christopher Wickens pooled 23 studies and found clear costs, concentrated in reaction time rather than in keeping the car in its lane, with hands-free phones performing about as badly as handheld ones [26]. That last point is the one that should change behaviour. Taking your hands back does not give you your attention back.

Where the 40 Percent Came From

Here is the number you have probably seen most. Switching tasks can cost up to 40 percent of your productive time.

It is usually attributed to the American Psychological Association, and behind that to David Meyer, the same Meyer from the 2001 study. That 2001 study measured switch costs in milliseconds per switch, and showed they grow with rule complexity [8]. The 40 percent figure is an estimate of what those small costs could add up to over a day of heavy switching. It is not a measurement of anyone's day.

That does not make it silly. A few hundred milliseconds, paid hundreds of times, plus the errors, is not nothing. But nobody watched office workers and found 40 percent of their time gone. Treat it as an expert's order of magnitude.

The 23 Minutes That Are Not in the Paper

The second number is more specific, which makes it more persuasive and more of a problem.

You will often read that it takes 23 minutes, sometimes 23 minutes and 15 seconds, to get back on task after an interruption, credited to a 2008 study led by Gloria Mark at the University of California, Irvine, with Daniela Gudith and Ulrich Klocke. That paper exists and it is worth reading. It does not contain that number.

What it found was odder and more interesting. People who were interrupted completed their tasks in less time than people who were not, with no difference in quality [27]. The authors' reading was that people "compensate for interruptions by working faster, but this comes at a price": more stress, more frustration, more time pressure and more effort.

That is a truer story than 23 minutes. Interruptions make you pay in a different currency.

Laboratory work on resuming after a break fits this. Erik Altmann and Gregory Trafton's 2002 memory-for-goals model treats a suspended task as a goal that fades unless it is rehearsed or cued [28]. In three experiments published in 2008, Christopher Monk and colleagues found that longer and more demanding interruptions produced longer resumption times, in a pattern that matched that fading over the first minute [29].

When an interruption lands matters, and so does how often it comes and who it lands on. A 2023 study found that interruptions in the middle of a subtask led to slower resumption and higher reported workload than interruptions at the boundary between subtasks [30]. A 2024 EEG study of 34 participants found that repeated interruptions did more damage than single ones [31], and another that year, comparing 32 younger and 28 older adults, found the older group lost more to interruption and was the only group that benefited from knowing one was coming [32].

The pattern matches the lab. The longer you are away, the more of the goal fades.

Not all interruptions come from outside. In a 2016 study, people who interrupted themselves finished the main task more slowly than people interrupted from outside, and pupil measurements suggested that deciding to break away cost about a second each time [33]. The glance at your phone that you chose is not free just because you chose it. Nor do you have to pick the phone up. In 2015 Cary Stothart and colleagues found that simply receiving a notification, with no response, disrupted performance on an attention task by about as much as actually using the phone for a call or a text [34]. Our longer piece on how notifications fragment learning goes further into why a buzz you ignore still pulls attention away.

Attention Residue

There is a name for the part of you that stays behind.

In 2009 Sophie Leroy, then at the University of Washington, called it attention residue. Across her studies, people who left a task unfinished and moved on to the next one kept thinking about the first, and did worse on the second as a result [35]. Finishing, or at least reaching a clean stopping point, let people move on more fully. Nine years later Leroy and Theresa Glomb tested a fix. Across four studies, people who knew they would have to return to an interrupted task under time pressure carried more residue into the interrupting task and performed worse on it. A short "ready-to-resume" plan, briefly noting where they were and what they would do on return, reduced the residue and protected performance [36].

It is a small intervention, and one of the few in this literature tested directly.

There is a twist worth knowing. Not every interruption hurts thinking. A 2024 natural experiment used a supply-chain shortage that unexpectedly stopped production at several plants. Employees who were suddenly left with idle time produced 58 percent more ideas over the following three weeks than employees who kept working [37]. Interruptions that forced people to switch to a different task, rather than leaving them idle, had the opposite effect. The switch is the cost, not the break.

The Stanford Study and the Decade After It

Now to the finding most people have heard about.

In 2009 Eyal Ophir, Clifford Nass and Anthony Wagner at Stanford built a questionnaire measuring how often people use several media at once, texting while watching TV, browsing while listening to music. They picked 41 students, 19 heavy media multitaskers and 22 light ones, and put them through a set of attention tests [38].

The result made headlines around the world. Heavy media multitaskers were more easily distracted by irrelevant items on screen and by irrelevant items in memory. And, surprisingly, they were worse at switching tasks, the very thing you would expect them to practise most. Ophir told Stanford's news office at the time: "We kept looking for what they're better at, and we didn't find it."

It was a small study, and the replications have not been kind.

In 2013 Meredith Minear and colleagues found no evidence that heavy media multitaskers were worse at task switching or at ignoring distraction [39]. The same year Reem Alzahabi and Mark Becker found the opposite of Ophir's switching result. In both experiments, heavy media multitaskers switched between tasks better, and were no different at doing two at once [40]. A 2012 study of 63 participants went further, finding that heavier media multitasking went with better ability to combine sight and sound [41].

Add Ophir back in and that is four studies pointing three different ways.

Then, in 2017, Wisnu Wiradhany and Mark Nieuwenstein ran two careful replications of their own. Across the 14 tests in the two studies, with an average statistical power of 0.81, only five showed the expected effect, and only two of those survived a stricter Bayesian analysis [42]. Their meta-analysis of 39 effect sizes found a weak association that disappeared once they corrected for the tendency of small studies to report bigger effects. Their conclusion was blunt: the results "lead us to question the existence of an association" in laboratory tests of this kind.

Later work has mostly pointed the same way, and one study pointed further. A 2021 meta-analysis by Douglas Parry and Daniel le Roux pooled 118 assessments and found the overall association between media multitasking and cognitive control was small, and varied with how both were measured [43]. A 2021 study designed specifically to separate the parts of the switch cost found no relationship at all between media multitasking scores and task switching [44]. A 2022 community study of people aged 7 to 70 found that more everyday media multitasking went with better multitasking performance, not worse [45].

So is the whole thing a myth? Not quite, and this is where the story gets more useful.

When Melina Uncapher and Anthony Wagner reviewed the field in 2018, they noted that many studies find no differences at all, but that where differences do appear, they tend to involve sustained, goal-directed attention rather than switching as such [46]. In 2020 Kevin Madore and colleagues recorded brain activity and pupil size in 80 young adults as they tried to remember things, and found that lapses of attention just before remembering predicted forgetting, and a tendency to lapse explained the link between heavier media multitasking and worse memory [47]. A 2023 study of 924 participants across three samples found a negative link between media multitasking and sustained attention, which the authors describe as medium in size, with correlations of about .20 on questionnaires and .21 on a task [48].

Every one of those is a correlation. It could be that multitasking erodes attention. It could equally be that people whose attention wanders more are drawn to multitasking. A longitudinal study of 2,390 adolescents aged 11 to 16 found evidence of a harmful long-term effect only in the youngest group, not in middle adolescents [49]. A 2015 review of youth research found a small to moderate negative relationship overall and noted that evidence on the direction of cause was missing [50].

The honest summary: heavy media multitaskers are probably not worse at switching. They may be somewhat more prone to drifting off. Nobody has shown which way the arrow points.

The People Who Multitask Most

One finding here is sturdier, and it is uncomfortable.

In 2013 David Sanbonmatsu and colleagues at the University of Utah compared how much people multitask with how good they actually are at it, using a demanding memory-and-arithmetic test called operation span. In their sample, the people who multitasked most, including those who reported using a phone while driving, tended to be the worst at it [51]. They also rated their own multitasking ability far higher than it was. The authors suggested that many people multitask not because they are good at dividing attention but because they are less able to block distractions and stay on one thing. That lines up with a 2022 reanalysis of 15 data sets on which media people combine. Multitasking was most common for pairs that let you control when to switch, that do not vanish if you look away, and that do not demand a response [52]. In other words, people mostly multitask in the combinations that hurt least. Your sense that it works fine may come from choosing the easy pairs.

Then there are the supertaskers. In 2010 Jason Watson and David Strayer screened 200 people in a driving simulator while they also did a memory and maths test. Nearly everyone got worse at both. About 2.5 percent, five people, showed no measurable cost [53]. A 2015 brain-imaging follow-up found that these people used parts of the frontal cortex more efficiently than controls matched on working memory [54].

It is a real finding, often misquoted as "2 percent of people can multitask". It was 2.5 percent of 200 students, on one pair of tasks, in one lab. And given Sanbonmatsu's result, feeling sure you are one is weak evidence that you are.

What about the claim that women are better at it? A 2013 study of 120 women and 120 men found both slowed down when two tasks were interleaved, and the men slowed more, with an effect size of d = 0.27 [55]. In a second study of 47 women and 47 men, there was no significant difference in solving arithmetic, searching a map or answering general knowledge questions. A small difference on one lab task is not a rule about the sexes.

Age, Practice and Learned Readiness

Switching does change with age, but not in the way people assume.

A 2011 meta-analysis of 26 published articles found that older adults had a clear deficit in the global cost of juggling two tasks, and no specific deficit in the local cost of an actual switch [56]. An updated 2023 meta-analysis agreed that there was no switch-specific age impairment once general slowing was accounted for, and flagged both high variability and publication bias in the studies [57]. A 2023 imaging study of 71 younger and 175 older adults reached a similar conclusion from the brain's side: older adults compensated for switch costs but not for mixing costs [58].

So getting older makes the open door more expensive than the switch itself.

Can training help? In 2009 Julia Karbach and Jutta Kray trained three age groups, children of 8 to 10, young adults of 18 to 26 and older adults of 62 to 76, on task switching. All three improved on similar tasks, children and older adults most, and they also found gains on other executive tasks and on fluid intelligence [59]. A 2008 training study by Meredith Minear and P. Shah was more cautious: the mixing cost improved and transferred, but the switch cost itself did not reliably shrink [60]. Part of what looks like skill may be expectation. A 2024 fMRI study found that switch costs fell when switches became more frequent, as people learned from recent experience how ready to be [61]. Your brain tunes its readiness to the world you keep putting it in. That cuts both ways.

Multitasking in the Lecture Hall

If you are a student, this is where the research gets close to home.

In a 2013 simulated lecture, students who multitasked on a laptop scored lower on a test of the material. So did students who simply had a view of a classmate multitasking [62]. The cost spread along the row. In real university lectures, a 2019 study found that media multitasking rose as the lecture went on and was linked to worse learning, while mind wandering stayed stable and was not [63]. A 2021 survey of 1,445 university students in three Southern African countries found a weak negative association between media multitasking and grades [64]. Weak is the right word. It does fit the classroom experiments above.

The likely mechanism is the one this article started with. Every glance away from a lecture costs a switch back in. Divided attention also hurts how well new material gets stored in the first place, which we cover in attention and memory. And because working memory is limited, each switch eats into the room you had for the material itself, the heart of cognitive load theory.

A small and still preliminary study hints at a newer version of the problem. In 2025, 72 participants watched either 30 minutes of short-form swiping video, a documentary, or nothing, and then did a task-switching test [65]. The documentary and no-video groups used extra preparation time to cut their switch costs. The short-video group did not. One study is not a finding, but it is a reasonable question.

Do You Know What It Is Costing You?

A common claim is that multitaskers are blind to the cost. The evidence is split, and it is split in an interesting way.

In the lab, people are surprisingly good at sensing a switch cost. A 2022 study asked participants to estimate their own reaction times and found the estimates reflected switch costs and remained "strikingly accurate" [66]. Another 2022 study found that both measured and self-estimated switch costs shrank when switching was more frequent, and that the self-estimates guided how often people chose to switch in one of three experiments [67].

Yet outside the lab, as Sanbonmatsu's result showed, people rate their overall multitasking ability far above what they can do. The likely reconciliation is scale. You can feel a single switch. You do not feel the sum of four hundred of them, spread across a day, mixed in with the extra errors and the effort Mark's participants reported. A 2021 review makes a related point: media multitasking can feel productive while objective performance falls, and whether it "works" depends on which goal you measure it against [68].

What the Evidence Supports Doing

None of this means never doing two things at once. It means knowing which kind you are doing.

The most direct evidence is for the ready-to-resume plan. When you have to leave something unfinished, writing one line about where you are and what comes next reduced residue in four studies [36]. It takes moments and it gives the fading goal a cue, which is exactly what the memory-for-goals model says it needs. Where you stop matters as much as whether you stop. In the 2023 study, breaking at a natural boundary cost less than breaking mid-thought [30]. If you control the interruption, choose the boundary.

Fewer switches beat faster ones. The residual cost does not vanish with preparation, so the most reliable way to pay less is to switch less often. That is the logic behind batching messages into set times, and behind protecting long blocks for anything that needs thought. It is also why an environment that makes switching harder helps, which we look at in distraction-free study and the brain.

Silence matters more than distance. The 2015 notification study found that a buzz you do not answer still costs you [34]. That argues for turning notifications off, not just ignoring them. And be sceptical of your own sense that you are fine at it. In the 2013 Utah study, the people most confident in their multitasking tended to be the least able [51]. If you notice you are reaching for a second screen out of restlessness, that pull deserves attention in its own right. A 2022 study found no direct evidence that a bout of boredom drives people to media multitask, though a boredom-prone temperament went with more of it in one of two experiments [69]. What boredom may be signalling is the subject of our piece on boredom as a motivational signal.

The opposite of switching is not a technique. It is the long stretch of focus described in deep focus and, at its best, the absorption of flow. Both are easier to reach when fewer doors are open.

The Research at a Glance

The lab costs are small and solid. The everyday claims built on them are bigger and shakier.

A Century of Switching

The core effect was found early and has held for nearly a century. The dramatic claims arrived late and have mostly been trimmed back.

What Is Settled and What Is Not

Some of this you can say without hedging. Switching between tasks is slower and more error-prone than staying on one. Preparation shrinks the cost and does not remove it. Two tasks that both require decisions interfere, and phone conversations impair driving whether or not your hands are free. Interruptions cost more when they are long, demanding or badly timed.

Some of it is open. Whether heavy media multitasking changes attention, or whether people with wandering attention simply multitask more, is unresolved, with Ophir, Uncapher and Madore on one side and Minear, Wiradhany, Parry and le Roux on the other. Whether the central bottleneck can be trained away is still debated between Pashler's camp and Meyer, Kieras and Schumacher. Whether task-switching training transfers broadly depends on whose study you read. Supertaskers rest on one screening study.

And some of it you should stop repeating. The 40 percent is an estimate. The 23 minutes is not in the paper. Two percent of people is 2.5 percent of 200 students. Heavy multitaskers are probably not worse at switching.

Conclusion

Go back one last time to 1995 and that 1.2-second head start.

What Rogers and Monsell found is that a mind warned in advance, given more than a second, doing a trivial task, still pays to change direction. That cost is small. It is also stubborn, and you pay it again every time you switch.

That is the real case against constant switching, and it needs no inflated numbers. A day of short hops is paid for in small charges: new rules, the pull of the last task, the suppression you undo when you go back, goals fading while you are away. Mark's participants did not lose their work to interruptions. They kept up by working harder and feeling worse.

You cannot opt out of the switch cost. You can decide how many times a day you pay it.