Mathematics: Worked examples and the transfer gap

What you'll be able to do: Get the real benefit from a worked solution, and know precisely what it will and won't buy you, because in mathematics the answer to that is unusually well established and unusually often ignored.

[EVIDENCE PENDING, one claim verified, marked ✓]

What the evidence actually says

Studying a complete worked solution is, for a novice, often more efficient than attempting a problem unaided. Sweller and Cooper (1985) found learners given problems to solve took roughly six times longer than those studying worked examples, with worse post-test performance. ✓ (Cognition and Instruction, 2(1), 59-89, verified.)

That's a large effect, it was found in algebra specifically, and it's why worked examples are everywhere in mathematics teaching.

Now the part that gets dropped, and that changes how you should use them:

The benefit did not extend to post-test problems that varied even modestly from the studied ones. ✓

Worked examples buy speed and accuracy on near items. They do not buy transfer. And transfer is what an exam requires, because an exam contains problems you haven't seen.

So the honest position: a worked example is an efficient way to acquire a method and a poor way to learn to select it. Everything below is machinery for getting the first while not mistaking it for the second.

How to study one properly

Reading top to bottom is nearly worthless. This takes about three times as long per example and is worth more than six read passively.

1. Attempt it badly for two minutes first. Even a failed attempt changes what you notice, you'll read for the bit you got stuck on.
2. Cover the solution; reveal one line at a time. Predict each next line before you see it. This turns reading into retrieval.
3. Annotate every step with why, not what. "Multiply both sides by x" is what. "Because we need to clear the denominator before factorising" is why.
4. Mark the decision points. In a typical integration by parts, exactly one step is a decision (choosing u) and the rest follows. Most students cannot identify the decisions, which is precisely the information they needed.
5. Ask what would have to be different for a different method to be right. This is the only step that builds selection.
6. Do a near-transfer problem immediately. Non-negotiable.
7. Then do one that looks similar and needs something else.

In that solution, which steps were genuine decisions and which were forced once the decision was made? For each decision, what was the alternative and why was it rejected?

The fading sequence: how to stop needing them

The exit route, and the answer to expertise reversal:

Give me the same type of problem four times:
1. fully worked, decision points labelled;
2. worked except the last two steps;
3. only the first step given;
4. nothing.
Don't move on until I've completed each. Tell me where I stall.

Then ask for the first step to be the missing one rather than the last. Missing the end tests whether you can carry a method through; missing the start tests selection, which is what you actually need and what nobody practises.

When to stop using them

The advantage shrinks and reverses as expertise grows, the expertise reversal effect. Once you can attempt a problem, studying the worked solution costs you the retrieval you'd have got.

Practical rule for mathematics: worked examples for genuinely new methods, attempts for anything you can half-do. If you can write the first line, write it.

Across the syllabus

Algebra. Where the original evidence comes from, and where the effect is strongest.

Integration. Study two, then move to completion problems with the substitution blanked. The substitution is the only decision.

Proof. Worked proofs are read as narratives and should be read as decision sequences. Ask which line was a choice.

Statistics. The worked example usually demonstrates computation; the difficulty is test selection. Worked examples help least here and students use them most.

Linear algebra. Good for procedures (row reduction, determinants) and poor for the conceptual content, which needs representation work (article 07).

Mechanics. The decision is the diagram, and worked examples usually present it already drawn. Cover the diagram and produce it yourself first.

Series. Worked examples of convergence tests are close to useless without the selection question, because the test is given.

Complex numbers. Good for procedure, and pair with the representation choice, which form makes this easy?

Pitfalls

1. Reading instead of predicting. Produces fluency and no capability.
2. Annotating what instead of why. You'll have a transcript.
3. No near-transfer afterwards. The research is explicit that the benefit doesn't extend on its own.
4. Assuming transfer. Ten worked examples do not prepare you for an unfamiliar problem. Budget separate practice for selection.
5. Continuing past competence. Once you can attempt it, attempt it.
6. The tell: you can follow any solution and start none. That's the exact signature of worked examples used without the transfer half.

Try this today

Take a worked example you've already read and thought you understood. Cover it.

Reveal one line at a time, predicting each. Then mark which steps were genuine choices, most solutions contain one or two.

If you can't find them, you've been reading the arithmetic.