Worked examples and cognitive load
Also known as the worked example effect, cognitive load theory, expertise reversal
Beginners learn more from studying full solutions than from struggling with problems — and experts learn less from them.
The claim Working memory is small enough that unguided problem-solving overloads novices; showing them complete solutions frees capacity for learning, and the advantage reverses as they gain expertise.
Good evidence The worked-example and expertise-reversal effects are well replicated; the wider theory's core construct — cognitive load itself — remains hard to measure directly.
The one uncontroversial fact about human cognition is that working memory is tiny. A few items, held for seconds, and everything novel has to pass through it before it can reach the effectively unlimited store of long-term memory. Cognitive load theory, developed by John Sweller from the 1980s onward, takes that bottleneck seriously and asks what instruction should look like given it.
The most useful result to fall out is the worked-example effect. Give a novice a problem to solve and most of their capacity goes into search — trying operations, checking whether they got closer, backtracking. That search is effortful and it teaches almost nothing, because the effort is spent on finding an answer rather than on noticing the structure of the solution. Give the same learner a fully worked solution to study, and they learn more in less time.
This sits awkwardly beside the desirable-difficulties literature, and the tension is real rather than apparent. The reconciliation most researchers accept is that difficulty helps when it is difficulty of the thing being learned, and hurts when it is difficulty of everything else — the interface, the search, the notation, the working memory bookkeeping. Sweller's term for the second is extraneous load.
The finding that stops this becoming a universal rule is expertise reversal. As a learner's knowledge grows, the guidance that helped them starts to hurt: they now have to reconcile the explanation with the schema they already possess, and that reconciliation is itself load. The instructional support that is optimal for a beginner is actively worse than nothing for an expert. There is no single correct level of scaffolding — only a correct level for a given learner at a given moment.
Cognitive load theory has real critics, and the sharpest criticism is methodological rather than philosophical: cognitive load is usually measured by asking people how hard something felt, which is the same self-report the illusions-of-learning literature shows to be unreliable. The instructional effects replicate; the explanation for them is less settled than textbooks imply.
What to do differently
- When starting something genuinely new, study complete solutions before attempting problems — and study them as objects, asking why each step follows.
- Fade the support as you improve: full solution, then a solution with the last step missing, then the last two, then nothing.
- Separate the difficulty of the material from the difficulty of the tools. Fighting an unfamiliar interface, notation or app is load that teaches you nothing.
- Do not split attention across sources — a diagram whose labels are in a paragraph elsewhere forces you to hold one while finding the other.
- Notice when guidance starts to feel like a detour. That feeling is expertise reversal, and it is the signal to drop the scaffolding.
Limits and trade-offs
- Taken too far it becomes an argument for never struggling, which contradicts a large and well-supported literature on desirable difficulties. Both are true, of different learners at different stages.
- The central measure — how much load a task imposes — is usually a rating scale, and self-reported effort is exactly what other parts of this field show people misjudge.
- It is a theory of instruction, so it assumes someone competent is designing the sequence. A self-directed learner has to do that design themselves, badly, from inside the subject.
- Worked examples work well for problems with a correct procedure and much less well for open, ill-structured tasks such as writing.
What this means for a language
This is the strongest case for the much-derided textbook. A carefully sequenced course is an attempt to control load — vocabulary introduced before the grammar that needs it, one new thing at a time — and for an absolute beginner in a distant language that sequencing does real work that immersion in native material cannot. Expertise reversal is also why the same textbook becomes unbearable at intermediate level: the scaffolding that made the first hundred hours possible is now the thing slowing you down. Knowing when to drop the course is not impatience, it is the effect operating on schedule.
The research
- Sweller, van Merriënboer & Paas (2019), Educational Psychology Review
The originators' twenty-year restatement of cognitive load theory, including the worked-example and expertise-reversal effects and the conditions on each.
Related
Read next
-
Cognitive load theory restated by its originators, including the effects that limit it — most importantly expertise reversal, where the support that helps a beginner starts to hurt an expert.
- Why Don't Students Like School? A Cognitive Scientist Answers Questions About How the Mind Works Book
Nine principles, each argued from the cognitive science rather than asserted. The chapter arguing that thinking well requires knowledge — not just skills — is the clearest statement of why 'learn how to think' cannot be taught in the abstract.
-
Thirty-two primary papers summarised and then argued with. The best bridge if you want to move from popular books to reading the literature yourself, and it does not hide where the authors’ own position is contested.
-
A short free report pairing cognitive-science questions with what the research answers and what it implies for teaching. The second edition adds memory, attention, motivation and misconceptions. Read this if you read nothing else.