Cognitive Psychology

Dual-Task and Limited Capacity Laboratory

Two easy tasks. Each of them, on its own, is barely a task at all. Put them on screen at the same moment and something has to give. Which of the two gives, and by how much, is more informative than the fact that anything gives.

Illustrative browser timing — a demonstration, not a measurement of anybody

Learning objective

By the end you should be able to measure a dual-task cost against its own single-task baseline, and say what an asymmetric cost implies.

About 30 minutes. Nothing you do here is saved or sent anywhere.

Before you start — how this is paced

Nothing is timed out and nothing flashes. Each display waits for your answer, blocks can be stopped between trials, and no part of the page animates, so there is nothing for a reduced-motion setting to switch off.

A dual-task cost is a difference from a baseline. Both single-task blocks come first, so that there is one. The worked example reaches every conclusion on this page without running anything.

  1. Answer the prediction below — it unlocks the laboratory.
  2. Choose how hard the second task is, and whether it needs an answer on every item.
  3. Run block 1: letters alone. Then block 2: numbers alone.
  4. Run block 3: both at once. Only now is there a cost to measure.
  5. Change the settings and run another set of three to compare.
  6. Prefer not to perform it? Load the worked example: a documented simulated dataset with a fixed seed.

First, a prediction

The first task is a letter: press Vowel or Consonant. The second is a number: press Odd or Even. Alone, either is trivial. In the third block both appear at the same instant and both need an answer.

What will adding the second task do to the first one?

The laboratory

Three blocks, in order. Use the buttons or the keys printed on them: F and J for the letter, D and K for the number. Nothing is timed out.

Key terms
Single-task baseline
How you performed on one task with nothing else running. Blocks 1 and 2 exist to establish one for each task.
Dual-task cost
The change in a task's performance from its own baseline when the other task is running alongside it. There are two of them here, one per task.
Asymmetric cost
When the two costs are not the same size. It is why each task is reported against its own baseline rather than combined into one number.

One task, the other task, then both

Results are withheld until a block ends, because a running score on screen would change how you answer the next trial.

  1. 1 Letters alone
  2. 2 Numbers alone
  3. 3 Both at once

Your task Answer the question above to unlock this.

Set up the second task

This is the overlap manipulation. Counting still needs the number to be read and held; it does not need a choice response on each item.

Run

A simulated dataset, seeded. Not anybody's data.

Answer the question above to unlock the laboratory.

Challenge — capacity, bottleneck, both or neither?

Two families of account. A general-capacity account says the two tasks draw on a shared, limited pool, so anything that makes either more demanding costs the other. A structural bottleneck account says one specific stage — usually choosing a response — can only serve one task at a time. What matters is whether the two tasks need that stage at once. They are not interchangeable, and some findings separate them.

Five findings, each to be assigned to the account it favours
Finding What does it favour?

What this demonstrates

A cost is a difference, and differences need baselines

The number that matters is not how fast you were in the dual block. It is how much slower you were than in your own single-task block, on the same day, with the same fingers. Comparing your dual-task time with somebody else's is meaningless: the two were never referred to the same baseline.

The cost is usually asymmetric, and that is the finding

People protect one task at the expense of the other, and which one they protect depends on instruction, on which stimulus they treat as first, and on their own strategy. A symmetric cost is a special case, not the default. An asymmetric cost immediately raises the question a symmetric one hides: who decided which task to protect, and how?

Two families of explanation that are not the same

A capacity account treats attention as a pool that can be divided, so difficulty anywhere costs performance everywhere. A bottleneck account treats one processing stage as strictly serial, so the cost appears only when two tasks need that stage at the same moment. The two make different predictions. Making the second task harder without changing its response requirement should hurt on a capacity account and matter much less on a bottleneck account. Removing the per-item choice response should help enormously on a bottleneck account and only moderately on a capacity account. The two settings in this laboratory are exactly those two manipulations.

Automatisation, and why "multitasking" is the wrong noun

Dual-task costs shrink with practice, sometimes dramatically, and the reduction is specific to the tasks practised rather than a general improvement in dividing attention. That is the strongest reason to be suspicious of "multitasking ability" as a trait: what improves is a pair of skills and the coordination between them, not a pool that got bigger.