Does Task Switching Really Cost Time
A popular attention book says the brain is network switching and that costs persist when a switch is predictable. The laboratory papers behind that idea give specific numbers: a few hundred milliseconds at most per switch, shrinking with warning and cues, rarely gone.
Answer to the question
Does task switching really cost time?
Yes, in the laboratory. Task switching slows responses on the first trial after a change, by tens of milliseconds to most of a second depending on the rules, cues and preparation time. Preparation shrinks the cost and often leaves a residual. How far those fractions of a second explain lost working time is a separate question.
Key takeaways
- Rubinstein and colleagues (2001) measured a switching cost of 975 milliseconds with twelve undergraduates sorting geometric objects, and 653 milliseconds in a 36-person arithmetic experiment. Switching from subtraction to addition cost 29 milliseconds.
- Rogers and Monsell (1995) found that lengthening the preparation interval up to 0.6 seconds reduced the cost of a predictable switch between digit and letter judgements. A reaction-time cost remained with 1.2 seconds of preparation.
- Kiesel and colleagues (2010) report a residual cost of 58 milliseconds at an 800 millisecond preparation interval. Brief cues sometimes removed the residual, and other studies found large costs after more than two seconds of preparation.
- In Rubinstein and colleagues' experiments the cost rose with rule complexity and fell with task cues. Strong incentives to prepare raised the share of fully prepared trials only marginally.
- Gazzaley and Rosen (2016) write that 'the brain itself is network switching' (p. 62). The reaction-time experiments supply the time costs, and the imaging describes the mechanism.
Gazzaley and Rosen's The Distracted Mind compresses the field of task switching into two sentences: when two goals both need cognitive control, "the brain itself is network switching," and switching costs persist even when the switch is predictable, growing with the complexity of the task. The experiments that match the second sentence report a cost close to nothing for one kind of switch and nearly a full second for another, both in one paper by Joshua Rubinstein and colleagues.
How Is a Switch Cost Counted in Milliseconds?
The method is older than the term. Arthur Jersild, nearly a century ago, had people alternate between two arithmetic tasks within a block and compared the time with doing one task alone, as Kiesel and colleagues describe the early work in their review of the field. Modern task switching experiments change the task on a schedule and compare trials where the task changes with trials where it repeats. The difference in reaction time is the switch cost. The slowdown across a whole mixed block, against a block of one task, is the mixing cost, which the same review files under alternation costs. The two are reported separately.
A digit judged odd or even, a letter judged consonant or vowel: those were the two tasks in Rogers and Monsell's experiments. The task changed on every second trial in five experiments and on every fourth in the sixth. The task switching cost fell on the first trial after a change, and it appeared even though the participants always knew the change was coming.
Four experiments reported in 2001 put figures on the cost, each defined as the difference in mean reaction time between blocks that alternated two tasks and blocks that repeated one. Joshua Rubinstein and colleagues used 12, 36, 36 and 24 young adults, according to the American Psychological Association's release. In the first, undergraduates sorted geometric objects by changing rules, and the switching cost was 975 milliseconds.
The arithmetic experiments gave smaller figures. A second experiment, with 36 people solving arithmetic problems, found 653. In a third, one direction of switching was close to free: from subtraction to addition problems the cost was 29 milliseconds, while the complexity effect was a mean difference of 960 milliseconds. On repeated blocks in the second experiment, multiplication and division took 3,597 milliseconds longer than addition and subtraction, which puts the switch cost at a few hundred milliseconds on top of trials that last several seconds.
Can Warning Remove the Cost?
Give people time to prepare and the task switching cost shrinks. In Rogers and Monsell's 1995 experiments, lengthening the interval between a response and the next stimulus, up to 0.6 seconds, reduced the cost, which they read as partial reconfiguration before the stimulus arrived. With 1.2 seconds of preparation a reaction-time cost still remained on the first trial of the new task. The 2010 review calls this a residual switch cost and reports that it survives "ample time for preparation" (p. 855). Schneider and Logan found one of 58 milliseconds at an 800 millisecond preparation interval, as the review reports it (p. 858).
Rubinstein and colleagues' own survey of earlier work notes mean costs of about 180 and 135 milliseconds after intervals of 20 and 1,100 milliseconds. Cues and complexity move the number in opposite directions. Rubinstein and colleagues report that switching-time costs increased with rule complexity and decreased with task cuing. Incentives to prepare did little. Nieuwenhuis and Monsell found that strong rewards for preparing raised the share of fully prepared trials only marginally, as the review summarises it (p. 856).
| Condition | What happens to the cost | Where reported |
|---|---|---|
| Preparation interval up to 0.6 s | Falls | Rogers and Monsell (1995) |
| Preparation of 1.2 s or more | A residual remains | Rogers and Monsell (1995); Kiesel et al. (2010) |
| Task cues | Falls | Rubinstein et al. (2001) |
| More complex rules | Rises | Rubinstein et al. (2001) |
| Strong incentives to prepare | Fully prepared trials rise only marginally | Nieuwenhuis and Monsell, via Kiesel et al. (2010) |
| Brief cues | Residual cost sometimes removed | Verbruggen et al. (2007), via Kiesel et al. (2010) |
The explanations of the leftover divide into two families. Two-stage models say a second reconfiguration step can begin only once the stimulus appears, so no amount of advance warning completes it (Kiesel et al., p. 855). Interference accounts put the leftover in the other task, still pulling on the response (p. 860). Both find support in the 1995 data. Rogers and Monsell attribute the residual to a reconfiguration process triggered only by the stimulus, and they also found that costs rose when a stimulus carried a character linked to the task not currently in force.
The empirical edge cases are where the families strain. Verbruggen and colleagues found that brief cues reduced the residual cost substantially and in some cases completely. Steinhauser and colleagues obtained substantial residual costs despite brief cues. Lien and colleagues found large costs even with more than two seconds of preparation, for some stimulus-response rules (all as summarised on p. 856). The review's own reading is that there are usually substantial true costs, and that they decrease if participants prepare (p. 860).
What Do the Papers Say About the Network Claim?
The book's p. 79 sentence is a fair compression of the two behavioural results above: costs that survive a predictable switch, and costs that grow with complexity. The p. 62 sentence is a claim about the brain, and its evidence is imaging from the authors' own laboratory. A 2011 study from that laboratory, as the University of California, San Francisco described it, had younger adults hold a natural scene in mind while a face appeared mid-delay. They disengaged from the memory network when interrupted, shifted resources to the face, then reconnected with the memory network afterwards. Older adults managed neither, which the authors linked to poorer recall.
That is a neural description of a switch, and the time costs come from the reaction-time experiments, not from the scans.
What a Few Tenths of a Second Add Up To
The American Psychological Association's topic page on the subject describes the costs of task switching as "relatively small, sometimes just a few tenths of a second per switch", which add up when people switch back and forth repeatedly. That is the measured quantity, and it is the one the experiments above report.
Whether those tenths of a second account for the lost feeling of a working day is a different question, and the interruption studies answer a different version of it. [Why Attention Residue Lingers at Work](https://thewhyweproject.com/articles/why-attention-residue-lingers-at-work) covers what carries over into the next task, and [Self-Interruption Is Half the Story](https://thewhyweproject.com/articles/self-interruption-is-half-the-story) covers who starts the switch. [The Borrowed Premise](https://thewhyweproject.com/articles/the-borrowed-premise) traces what happened to the larger percentages that circulate with this research.
Common questions
Does task switching really cost time?
Yes, in the laboratory. Task switching slows responses on the first trial after a change, by tens of milliseconds to most of a second depending on the rules, cues and preparation time. Preparation shrinks the cost and often leaves a residual. How far those fractions of a second explain lost working time is a separate question.
What is the difference between a switch cost and a mixing cost?
A switch cost compares trials where the task changes with trials where it repeats, within a mixed block. A mixing cost compares a mixed block with a block of one task. Kiesel and colleagues call the second a global cost and the first a local one.
Can preparation remove the cost completely?
Usually not. Residual costs survive long preparation in several experiments, though brief cues reduced them substantially, and sometimes completely, in one set of studies.
Is multitasking the same thing as task switching?
The Distracted Mind says task switching, not multitasking, better describes what the brain does when a person believes they are multitasking. Its p. 62 claim is that two goals needing cognitive control alternate between brain networks rather than running together.
Sources
Jersild, A. T. (1927). 'Mental set and shift.' Archives of Psychology, No. 89. Described as in Kiesel et al. (2010), p. 850.
Rogers, R. D., & Monsell, S. (1995). 'Costs of a predictable switch between simple cognitive tasks.' Journal of Experimental Psychology: General, 124(2), 207-231. doi:10.1037/0096-3445.124.2.207
Rubinstein, J. S., Meyer, D. E., & Evans, J. E. (2001). 'Executive control of cognitive processes in task switching.' Journal of Experimental Psychology: Human Perception and Performance, 27(4), 763-797. doi:10.1037/0096-1523.27.4.763
Kiesel, A., Steinhauser, M., Wendt, M., Falkenstein, M., Jost, K., Philipp, A. M., & Koch, I. (2010). 'Control and interference in task switching: A review.' Psychological Bulletin, 136(5), 849-874. doi:10.1037/a0019842
Clapp, W. C., Rubens, M. T., Sabharwal, J., & Gazzaley, A. (2011). 'Deficit in switching between functional brain networks underlies the impact of multitasking on working memory in older adults.' Proceedings of the National Academy of Sciences, 108(17). doi:10.1073/pnas.1015297108
American Psychological Association. 'Multitasking: Switching costs.' Topic page, apa.org/topics/research/multitasking.
American Psychological Association (2001). 'Is Multitasking More Efficient? Shifting Mental Gears Costs Time, Especially When Shifting to Less Familiar Tasks.' Press release, 5 August 2001
Gazzaley, A., & Rosen, L. D. (2016). The Distracted Mind: Ancient Brains in a High-Tech World. Cambridge, MA: MIT Press, pp. 62, 79.