Split attention in materials
A Socratic walk-through of split attention in materials — reasoned out one step at a time, not lectured.
The question we started with
THE QUESTION #Why does a diagram with its labels in a separate key teach less than the same diagram labelled in place?
Two handouts carry the same diagram of a hydraulic circuit and the same twelve sentences of explanation. On one, each sentence sits beside the component it describes. On the other, the sentences are numbered in a key below the figure. Nothing has been added or removed — only moved. Yet learners given the second version reliably do worse on a later test.
That is peculiar, if you think of a page as a container for information. The container holds the same contents either way. So what is the arrangement costing, and who is paying it?
Reasoning it through
REASONING #Let us watch someone read the keyed version. They meet label 7 on the drawing. They must hold "7" and its position in mind, drop their eyes to the key, scan for the matching number, read a sentence — and now hold that sentence in mind while returning to find the spot again. The sentence only becomes meaningful once it is reunited with the part it names.
Notice what is being consumed there. Not time, primarily, though it costs that too. What is consumed is the small, fragile store that holds the things you are currently thinking about. How small is contested — Miller's famous seven-plus-or-minus-two and Cowan's later estimate of about four chunks disagree — but no serious account makes it large, and the exact number is not load-bearing here. What matters is that holding a fragment while searching for its partner is itself an occupation of that store, and whatever is spent on it is unavailable for the thing we actually wanted: building an understanding of how the circuit works.
Cognitive load theory's name for that expenditure is extraneous load, and the adjective is the whole claim: this cost comes from the format, not from the subject. A hydraulic circuit is intrinsically hard. Splitting its explanation across the page makes it no harder, and yet makes it harder to learn — which is exactly the signature of a cost that lives in the presentation.
Now let us test that against the obvious rival explanations, because "the eyes have to travel" is a tempting story and it is not quite the right one.
If distance were the mechanism, the cure would be proximity alone, and the effect should vanish once the key sits close to the figure. But the same penalty appears when the two sources are separated in time rather than space — a narration played before an animation rather than during it — where there is no scanning at all. Whatever is being taxed survives the removal of eye movement, which points at memory rather than at gaze.
If slowness were the mechanism, then giving learners unlimited study time would erase the gap. Broadly, it does not: the format penalty shows up in comprehension and transfer, not merely in reading speed.
And here is the test that decides it. Take a diagram that is fully self-explanatory on its own, and integrate a text description into it. By the "closer is better" story, that should help. It does not — it makes learning worse. This is the redundancy effect, and it is the reason the split-attention account has to be stated carefully: integration is not a virtue in itself. It pays only when neither source means anything alone, so that the learner would otherwise have to perform the join in their head.
The analogy
THE ANALOGY #Think of assembling flat-pack furniture from instructions in which the drawing shows lettered holes and a separate sheet says what goes in each. You keep a letter in your head, cross the room to the sheet, carry a sentence back, and hope the panel has not rotated in the meantime. Print the words on the drawing at the holes and the same instructions become almost effortless — not because they say more, but because you no longer have to be the one holding both halves.
the furniture task lets you set a fragment down and go back for it, at the cost of only a walk; working memory has no such shelf, and a fragment displaced there is not delayed but lost, which is why the penalty is to learning and not merely to speed.
Clarifying the model
THE MODEL #Three refinements connect the steps.
First, the effect is about mutual unintelligibility. The design rule is not "put things close together" but "never require a learner to hold one source in mind while searching another in order to make either mean anything." That is why the same move — integration — is correct for a labelled diagram and wrong for a redundant one.
Second, it interacts with expertise, and the sibling explanation in this collection on expertise reversal is the general form of the point. A learner who already has a schema for hydraulic circuits does not need the sentences at all, so integrating them turns them into redundant material that must still be processed and reconciled. The split-attention effect and the redundancy effect are not two rules but one rule read at two levels of prior knowledge, and the practical response is the same: fade the support as the schema forms.
Third, the honest caveat. The empirical pattern — integrated formats beating separated ones for novices on unfamiliar material — is well replicated across many subjects and is one of the sturdier findings in instructional design. The explanation rests on cognitive load theory, whose central quantity is inferred rather than measured, usually from self-report or a secondary task, and which critics argue can accommodate almost any outcome after the fact. Reported effect sizes for spatial integration are large — around one standard deviation in Mayer's summaries, a figure I am recalling rather than deriving — and I would treat that magnitude as the weakest link in the chain, since it comes mostly from short controlled studies of well-structured material. The direction is solid; the size travels less well.
A picture of it
THE PICTURE #How to readThe horizontal axis is the design choice you control — how far apart the sources sit. The vertical axis is the property of the material that decides whether that choice matters: high up, neither source means anything without the other. Read the top row first, since that is the split-attention finding proper: the same content moves from the worst cell to the best purely by integration. Then read the bottom row, which is the control condition that rules out "closer is always better" — when each source stands alone, separation is harmless and integration actively hurts.
What became clearer
WHAT CLEARED #The page is not a container. Whatever the learner must join up, the learner must hold, and holding is the scarce resource. A key at the foot of a figure quietly delegates the joining to the reader, spending on bookkeeping the very capacity that understanding needed. Put the words where the thing is, and you have not simplified the subject — you have stopped charging for something that was never part of it. Unless the reader could already read each part alone, in which case you have started charging for something else.
Where to go next
ONWARD #- How the modality effect exploits a second channel, letting spoken words accompany a diagram without competing for the same store.
- Why worked examples and completion problems fade guidance on the same logic, one step at a time.
Key terms
TERMS #| Term | What it means |
|---|---|
| Extraneous load | demand on working memory created by how material is presented rather than by what it contains. |
| Split-attention effect | the learning penalty from separating two sources that must be mentally integrated to be understood. |
| Redundancy effect | the learning penalty from presenting material a learner can already do without, including a text that duplicates a self-explanatory diagram. |
| Spatial contiguity principle | Mayer's design rule that corresponding words and pictures be placed near each other on the page or screen. |
Every term the collection defines is gathered in the glossary.