3.4 — Mixing, Connecting, and the Truth About Two Channels

Two groups of children practise throwing beanbags into a bucket.

One group practises only from three feet. The other practises from two feet and four feet, and never from three feet at all.

Then everyone is tested at three feet. The group that never practised at that distance does better.

This result, from a 1978 study, is the clearest demonstration of what varied practice does. The group that practised the exact target distance learned one throw. The group that practised around it learned the relationship between distance and force, and could therefore produce a throw they had never made.

This page covers the three remaining techniques with real evidence behind them, and finishes with the most popular idea in education, which has none.

Interleaving: mix the problems

3.2 introduced it; here is how to actually do it.

Blocked practice does all of type A, then all of type B, then all of type C. Interleaved practice mixes them: A, C, B, A, B, C.

The evidence is consistent and the size is often large. In one study of university students learning to compute volumes of four different solids, the interleaved group performed worse during practice and, on a test a week later, got roughly three times as many correct as the blocked group.

Why it works. In a blocked set, you never have to decide which method applies — the heading told you. Then the real problem arrives with no heading. Interleaving trains the step that blocked practice skips, which is recognising what kind of problem you are looking at. In most subjects that step is the hard part; once you know it is a related-rates problem, the mechanics are routine.

It also forces spacing automatically, since each type is revisited after a gap.

Where it applies and where it does not. Interleave things that are related but need to be distinguished — problem types in mathematics, similar-looking diseases, chords, verb tenses, painters, drug classes. Do not interleave completely unrelated subjects in a way that just adds switching cost (2.3); mixing chemistry and Bengali grammar every four minutes is not interleaving, it is task-switching.

And there is a sensible ordering. Get the basic method for each type first, in a small block, then start mixing. Interleaving from the very first exposure, before you can do any of them, is just confusion.

The cost you must accept. Interleaved practice feels worse, produces more errors, and is slower. Students consistently rate it as less effective than blocking while performing better on it. If you wait until it feels right, you will never use it.

Elaboration: connect it to what you know

The strongest single predictor of whether something sticks is how well it connects to your existing knowledge (3.1). Elaboration is the deliberate building of those connections, and it is mostly a matter of asking two questions.

"Why?" — repeatedly, until it bottoms out in something you already accept. Called elaborative interrogation in the research, and it reliably improves retention over just reading the fact.

Take: deep sleep happens mostly in the first half of the night. Why? Because sleep pressure is highest at the start and dissipates. Why does that shift the composition? Because the drive for deep sleep is what accumulates while awake, and REM is regulated more by circadian timing than by pressure. You now cannot forget it easily, because it is attached to two other things you understand, and either one can lead you back to it.

"What does this remind me of?" — deliberate analogy. New material connected to something familiar has a retrieval route from the moment it arrives. The caution is that analogies are always partly wrong, so the useful habit is to state both halves: this is like X in that…, and unlike X in that…. The second half is where most of the understanding is.

Self-explanation is the version with the best evidence in problem-solving. After each step of a worked example, say why that step was taken. Students who do this learn substantially more from the same examples than students who read them straight through — and they transfer better to new problems, which is the outcome that matters.

And concrete examples are elaboration in the other direction. Abstract principles are hard to hold and hard to apply. The reliable method is: the principle, then two or three examples from genuinely different areas, then the principle stated again. Different areas matters — three examples all from cricket teach you about cricket. When the surfaces differ and the structure is the same, the structure becomes visible, and that is what transfers.

Dual coding: two channels, one message

Working memory has partly separate capacity for verbal and visual material. Present information in both, and you use more capacity without more interference — and you create two retrieval routes to the same content.

The evidence for well-designed diagrams alongside text is good. But this only works under specific conditions, and getting them wrong makes it worse rather than better.

The two must be integrated, not merely present. Text on one page and the diagram on the next forces you to hold one while looking for the other, and that split attention destroys the benefit. Labels go on the figure. The explanation goes beside the part it explains.

Do not present the same words in two forms. Reading text aloud while the identical text is on screen is worse than either alone — the two verbal streams compete for the same channel. This is why a presenter reading their slides is uniquely ineffective, and it is a documented effect rather than a matter of taste.

Decoration hurts. Images that are interesting but unrelated to the content reduce learning. So do animations that move for the sake of it, and background music with words in it.

The strongest version is making the diagram yourself. Drawing what you understand — a flow, a comparison, a structure — combines dual coding with retrieval and generation at once, and it exposes gaps instantly, because you cannot draw a relationship you have not understood. A page of your own boxes and arrows, drawn from memory, is worth ten pages of highlighted text.

Learning styles: the popular idea with no evidence

Somewhere between 80 and 95 per cent of teachers in surveyed countries believe that students learn better when taught in their preferred style — visual, auditory, or kinaesthetic. It is one of the most widely held beliefs in education.

It has been tested properly and it does not hold.

The test is specific and it is the right one. Identify learners' preferred styles, teach some in their preferred mode and some in a mismatched mode, then test everyone on the same material. The prediction is an interaction: visual learners should do better with visual instruction than auditory learners do. Study after study designed this way has found no such interaction. A major review in 2008 found almost no studies using an adequate design, and the ones that existed produced negative results. Later work has not changed the picture.

What is true, and gets confused with it.

People genuinely have preferences. They are real preferences. They just do not predict what works.

Material has a best format, and it is a property of the material, not the person. Geography needs a map. A chemical reaction needs a diagram. A poem needs to be heard. Everyone learns the layout of a building better from a plan than from a description, whatever their supposed style. Ask what the content demands, not what the learner prefers.

Everyone benefits from multiple formats, which is dual coding, and it is a different claim.

Why this matters practically rather than as a debunking exercise: believing you are a particular type of learner does real damage. It gives you a reason to avoid formats you find hard — which are frequently the ones you most need practice with — and it supplies a ready explanation for failure that removes the incentive to change method. "I'm a visual learner" ends an investigation that should have continued.

What to do with this

Mix your practice deliberately. When working problems, do not do fifteen of the same type. Do a mixed set, and if the source is organised in blocks, shuffle it yourself — write problem numbers on slips, or work backwards through mixed chapters.

Ask "why is that true?" three times on anything important. Chase it until you reach something you already believe. This is slower and it is the difference between knowing a fact and owning it.

Find two examples from different areas for every principle you learn. One from your own work, one from somewhere unrelated. If you cannot produce the second, you have probably learned the example rather than the principle.

Draw it from memory. Once per topic, close the book and draw the structure — what connects to what, in what order, with what causing what. Then compare with the source. This is retrieval, generation and dual coding in one operation, and it is the highest-density technique in this Part.

Explain each step of a worked solution before reading the next line. Cover the rest of the page and predict.

And drop the idea that you are a certain kind of learner. Ask instead what this particular material needs, and use that. The only stable difference between learners that predicts outcomes is what they do, and every technique in this Part is available to anybody.

Next: 3.5 turns from learning material to building skill — what ten thousand hours actually bought, why most practice produces no improvement at all, and what separates the two.