Task switching and switch cost: what the research shows
Task switching is the study of what happens when a person changes from one simple task to another. The main finding is the switch cost: responses on the first trial after a change are slower and less accurate than when the task repeats. Jersild reported the slowdown in 1927, and Rogers and Monsell measured it trial by trial in 1995.
The game above is Task Shuffle. It shows one digit and asks one of two questions about it, and the question changes every few digits.
What is task switching?
Task switching is an experimental method in which the same kind of stimulus can be handled under two different rules, and the rule in force changes during the session. A digit might be judged as odd or even on some trials and as below or above five on others.
Every trial is one of two kinds. On a repeat trial the task is the same as on the trial before. On a switch trial it has changed. The switch cost is the difference in performance between them.
In the versions described here the stimuli fit both tasks. A digit does not tell you which question to answer, so you have to keep track of the rule yourself or read a cue.
What did Jersild find in 1927?
Jersild found that people worked more slowly when they alternated between two tasks than when they did one task at a time. His monograph, “Mental set and shift”, was published in Archives of Psychology in 1927 and is credited as the first use of the task-switching method.
His comparison was between whole blocks of work: one task throughout, or two tasks in rapid alternation. That design shows that alternating is slower overall. It does not show what happens on any single trial, which was hard to time before computers.
What did Rogers and Monsell add in 1995?
Rogers and Monsell measured the cost on individual trials and showed that most of it falls on the first trial after a switch. Their paper, “Costs of a predictable switch between simple cognitive tasks”, appeared in the Journal of Experimental Psychology: General.
Each display in their experiments held a pair of characters. One task was to classify the digit as even or odd. The other was to classify the letter as a consonant or a vowel. The task changed on a fixed schedule that people knew in advance: every second trial in five of the experiments and every fourth trial in the last one.
Three results from the paper’s abstract are still cited:
- Giving people longer between their response and the next stimulus reduced the switch cost, up to an interval of 0.6 seconds. People could get partly ready for the new task ahead of time.
- A large cost remained even with 1.2 seconds to prepare, and it appeared on the first trial of the new task only.
- The cost was larger when the display included a character linked to the task that was not in force.
Rogers and Monsell took the second result to mean that part of the adjustment cannot be done in advance and waits for a stimulus of the new task to arrive.
Monsell’s 2003 review in Trends in Cognitive Sciences sums up the wider literature the same way. Responses are slower and more error-prone right after a switch. Preparation reduces the cost and does not remove it. He attributes the cost partly to the time needed to reconfigure for the new task and partly to carry-over from the task just left.
How do our three switching games work?
All three show a stimulus that fits two rules, mark which rule is in force, and change the rule every few trials.
| Game | The two tasks | How the rule is shown |
|---|---|---|
| Task Shuffle | Odd or even; below or above five | A chip above the digit, and the answer tiles relabel |
| Rule Flip | Sort by color; sort by shape | A chip above the item, and the bins show colors or shapes |
| Priority Switch | Name the shape; odd or even for the number | A frame around the tile that counts |
Each round lasts 30 seconds and each right answer is one point. A right answer holds the board for 170 milliseconds before the next trial. A wrong one holds it for 520 milliseconds.
Two details set these games apart from the Rogers and Monsell design. The changes are not on a fixed schedule: the length of each run is drawn at random from a small range. And the cue arrives together with the stimulus, so there is no interval in which to prepare. In their terms, the games sit at the short end of the preparation range, which is where they measured the largest cost.
How often does the rule change in each game?
The three games share one piece of scheduling code, so the rule changes at the same rate in all of them. The rate depends on your level and on how far into the round you are.
A rule holds for a run of trials whose length is picked at random between a minimum and a maximum. Runs that begin after the fourteenth trial of a round are one trial shorter at both ends. Your level starts at 0, goes up after two rounds in a row with at least 80 percent accuracy, and comes down after a round under 50 percent.
We read the run lengths from the code. To get the share of switch trials, we generated 5,000 rounds per level for each game with a short program and counted. The three games agreed to within 0.2 percentage points.
| Level | Run length, first 14 trials | Run length after that | Switch trials among trials 2 to 14 | Switch trials from trial 15 on |
|---|---|---|---|---|
| 0 to 1 | 4 to 6 | 3 to 5 | 16.5% | 25.0% |
| 2 to 4 | 3 to 5 | 2 to 4 | 22.4% | 33.2% |
| 5 to 7 | 2 to 4 | 1 to 3 | 31.0% | 49.8% |
| 8 and up | 1 to 3 | 1 to 2 | 48.6% | 66.4% |
At level 0 the rule changes on about one trial in six early in the round and one in four later. At level 8 and above it changes on two trials in three once the round is under way, so a repeat is the less common case.
How do the games differ in conflicting items?
They differ in how often the stimulus points to different answers under the two rules. Rogers and Monsell found a larger cost when the display carried a character tied to the other task, and this is the part of their design that the games vary.
In Task Shuffle the digits decide it. Odd and below five are both on the left tile, and even and above five are both on the right. So 1, 3, 6 and 8 have the same answer side under either question, and 2, 4, 7 and 9 change sides. The eight digits are picked with equal chance, which makes half of all trials conflicting at every level. Our count came out between 49.9 and 50.0 percent.
Rule Flip and Priority Switch set the share in code, and it rises with level. From level 2 they add a second rule: on every switch trial, the item’s two features are forced to point at different answers.
| Level | Task Shuffle | Rule Flip and Priority Switch |
|---|---|---|
| 0 | 50% | 50% |
| 1 | 50% | 60% |
| 2 to 3 | 50% | 73% |
| 4 | 50% | 80% |
| 5 to 7 | 50% | 84% |
| 8 and up | 50% | 89% |
The last column is the measured share of conflicting items over 5,000 generated rounds per level, rounded to a whole percent. It climbs past the base rate of 60 or 70 percent because switch trials get more frequent and every one of them is a conflict.
What follows from this when you play?
- Check the rule on every trial. The cue appears with the stimulus and never before it. The answer tiles change with it, so they show the rule as plainly as the chip does.
- Expect faster changes after the fourteenth trial. At level 0 that means a rule lasts three to five trials where it used to last four to six.
- In Task Shuffle, 1, 3, 6 and 8 are safe. They sit on the same side for both questions. The digits 2, 4, 7 and 9 are where an answer under the old question is wrong.
- In Rule Flip and Priority Switch from level 2, the first item after a change always conflicts. An answer by the old rule on that item is always wrong, and it costs 520 milliseconds where a right answer costs 170.
- In Priority Switch, find the frame before reading either tile. The unframed tile is a decoy and from level 4 it points at the other answer on at least 70 percent of turns.
The switch cost described in the research is a laboratory measurement taken under controlled timing. A score in these games is the number of right answers in 30 seconds and says nothing about a person’s ability. People get better at the games with practice. Whether that carries over to other activities is not established, and we cover the evidence in what the research shows about brain training games.
A related task hides the rule and gives no cue at all. We describe it in the Wisconsin Card Sorting Test explained, and Card Sort is our playable version.
Common questions
What is task switching in psychology?
Task switching is a laboratory method in which a person moves back and forth between two simple tasks, such as judging a digit as odd or even and judging it as high or low. Researchers compare trials where the task repeats with trials where it has just changed.
What is a switch cost?
A switch cost is the difference in speed and accuracy between a trial where the task has just changed and a trial where it is the same as the one before. Responses right after a change are slower and more error-prone.
Does preparation remove the switch cost?
No. Rogers and Monsell (1995) found that more time to prepare made the cost smaller, up to about 0.6 seconds, but a large cost was still there with 1.2 seconds to prepare. It showed up on the first trial of the new task only.
Sources
- Rogers, R. D., & Monsell, S. (1995). Costs of a predictable switch between simple cognitive tasks. Journal of Experimental Psychology: General, 124(2), 207-231.
- Monsell, S. (2003). Task switching. Trends in Cognitive Sciences, 7(3), 134-140.
- PsyToolkit experiment library: Task switching (describes Jersild, 1927, Mental set and shift, Archives of Psychology, 89)
- Wikipedia: Task switching (psychology)