Switch Cost

What it is

Switch cost is the accounting correction that makes interruption legible. A two-minute interruption is not a two-minute loss, because the two minutes are the cheapest part of it. What costs is getting out of the first task and back into it, and the mental resources consumed doing so. This is why an afternoon fragmented into eight ten-minute pieces produces less than an afternoon of eighty uninterrupted minutes, even though the arithmetic says they are the same.

The reason the cost exists at all is capacity. Everything you are currently thinking about sits in working memory, a small and temporary store through which everything on its way to long-term memory has to pass. George Miller's famous 1956 paper put its size at about seven items. That figure is out of date and should not be repeated. Current estimates are lower, around three to four, following Nelson Cowan's 2001 revision, and this page carries the corrected number rather than the memorable one. Nicholas Carr, whose The Shallows is the popular vector for most of this material, is careful to state the revision rather than reach for the magical seven, which is more than most popular accounts manage. His image for it is a thimble filling a bathtub: the bath is long-term memory and effectively unbounded, the thimble is what you can carry across at once, and information arriving faster than the thimble can move it is not stored faster, it is spilled. Every switch empties and refills the thimble.

The best-evidenced consequence in our notes is not the one people expect. In Foerde, Knowlton and Poldrack's 2006 study in PNAS, participants learned a categorisation task either undistracted or while simultaneously monitoring a stream of tones. Both groups eventually learned the task about equally well. The difference appeared afterwards. The distracted group was markedly worse at applying what they had learned to new cases, because the distraction had shifted learning out of the flexible declarative memory system and into a more rigid procedural one.

That is the finding worth keeping, and it is more interesting than "multitasking makes you worse at things." Multitasking did not stop the learning. It changed the kind of learning, into a form that does not transfer. You end up able to do the exact thing you practised and unable to do the adjacent thing, which is a failure mode that is almost invisible from the inside, because performance on the practised task looks fine.

There is a real carve-out, and it should be stated rather than buried, because the blanket version of this claim is overstated. Two tasks competing for the same sensory or cognitive channel do interfere, and people are measurably slower and more error-prone. Two tasks using different channels can genuinely run together. Two podcasts, one in each ear, is impossible. An audiobook while walking is not. Nir Eyal makes this carve-out in Indistractable after conceding the standard finding first, which is what makes it credible rather than convenient, and it sits comfortably alongside Poldrack's result: what same-channel distraction degraded was the transferability of learning, not the capacity to learn.

The everyday environment most of this happens in is worth naming. On Eyal's account, around 70 per cent of American offices are open-plan, and he cites a 2016 meta-study of more than 300 papers on open-plan distraction. Neither the percentage nor the meta-study is named with an author in the text we have, so treat both as his report rather than as verified figures.

In effect

The categorisation task with the tones

The research case is Foerde, Knowlton and Poldrack (2006), and the value of it is the precision. Two groups learned the same categorisation task, one of them while counting tones. Both learned it. Only the undistracted group could apply it to new cases. If you want a single sentence for why divided attention matters in education and training specifically, it is that the distracted learner acquires a version of the skill that works only where it was acquired. Note that this is a laboratory task, not a semester of coursework, and the extrapolation to schooling is ours rather than the authors'.

The lecture hall, and the screen two rows in front

Sana, Weston and Cepeda's 2013 study in Computers & Education is the one with the uncomfortable finding. Laptop multitasking during a lecture harmed not only the person doing it but students seated nearby who could see the screen. The cost of a switch, in other words, is not confined to the person switching. This is the most concrete everyday demonstration in our notes that attention is an environmental condition and not only a personal discipline, and it is the reason a policy about laptops is not merely paternalism about the laptop's owner.

Push notifications, and the 15 per cent who change the settings

The product-design case is the deliberate manufacture of switches. Eyal reports that Apple sent 7.4 trillion push notifications in 2013, and quotes Adam Marchick of the mobile-marketing firm Kahuna saying that under 15 per cent of users ever change their notification settings. Tony Stubblebine's three-category audit of a phone, Primary Tools, Aspirations and Slot Machines, exists because the third category is built to initiate switches rather than to be opened deliberately. The commercial logic is clean: an app that can trigger a switch does not have to compete for a decision to open it. The 7.4 trillion figure and the 15 per cent are Eyal's reporting and are not traced by us to a primary document.

What it does not say

It does not say all multitasking is equivalent. The channel matters. Two demands on the same sensory or cognitive channel interfere; two demands on different channels can run together. Advice that treats walking with an audiobook as the same act as reading email during a meeting is not describing this literature.

It does not say multitasking prevents learning. The Poldrack result specifically did not find that, and the popular version that says it does is a misreading. Both groups learned. What differed was whether the learning transferred.

It does not carry the magical number seven. Working memory capacity is now estimated at around three to four items, not seven. The 1956 figure is one of the most repeated numbers in popular psychology and it has been superseded for decades. Any page, including this one, that leans on capacity limits should use the corrected estimate.

The wider literature has not been assessed here. This page is marked primary-cited, which in our vocabulary means the citations behind it are real and correctly attributed but nobody has audited how well the surrounding field has replicated. Foerde, Knowlton and Poldrack (2006) and Sana, Weston and Cepeda (2013) are correctly cited. Whether the broader task-switching literature has held up under the replication reckoning is a question this page does not answer.

It does not have a tidy origin story. See below.

Who developed it

Switch cost has no single originator, and this page declines to name one. Our vault note records Arthur Jersild as having described task-switching cost in 1927, and records that the term itself is standard vocabulary in cognitive psychology rather than any one person's coinage. Both of those are honest, and neither adds up to an originator in the sense this site uses the word. The specific findings the page rests on belong to their own authors: Russell Poldrack with Karin Foerde and Barbara Knowlton for the memory-systems result, and Faria Sana with Tina Weston and Nicholas Cepeda for the lecture-hall result. Nicholas Carr is the popular vector rather than a source, and Nir Eyal supplies the carve-out. None of Jersild, Poldrack, Foerde, Knowlton, Sana, Weston or Cepeda has a page on this site yet.

---