Mathematics: Interleaved problem sets

What you'll be able to do: Practise in a way that feels worse, scores lower, and leaves you able to tell which method a problem needs, which is what the exam is actually testing.

The specific failure this fixes

You finish the chapter on integration by substitution. Twenty problems, all substitution, and by problem eight it's automatic. You feel you've mastered it, and in a narrow sense you have.

Then the exam gives you an integral with no chapter heading, and you don't know whether it wants substitution, parts, partial fractions or a trig identity.

In a blocked problem set you never once had to decide. The heading decided for you, twenty times. You practised execution and the exam tests selection, and nothing in your revision touched it.

Mathematics is the subject where this gap is widest, because the methods look similar on the page and the difference is recognition. Once you know it's integration by parts, doing it is routine.

What interleaving does about it

Mix problem types within a session, unlabelled. Now every problem requires the decision, because nothing tells you which family it's from.

The evidence here is clean and counterintuitive: interleaved practice produces worse performance during practice and better performance on later tests. Learners consistently rate it as less effective while doing better from it. The discomfort is the mechanism.

Expect your accuracy to drop. That's the method working, and the first session is genuinely unpleasant.

How to run it in mathematics

1. At least three types, and preferably five. Two alternating becomes its own detectable pattern.
2. Unlabelled. The moment a problem is tagged, you're back to blocked practice with extra steps.
3. State your method and why, before solving. This is what makes interleaving worth more than shuffling, the selection has to be conscious or you can't learn from getting it wrong.
4. Include material from previous months, which adds spacing to interleaving.
5. Track selection errors separately from execution errors.

Give me 12 problems mixed randomly from [topics]. Don't label which type each is. Before each, I'll tell you which method I'll use and why, correct my SELECTION before I solve, and tell me what in the problem should have signalled the right method. At the end, separate my errors into selection and execution.

That separation is the single most useful output. "Wrong answer" splits into I chose the wrong method (which needs discrimination practice) and I chose right and made an arithmetic slip: which needs something completely different, Most students respond to both with "revise harder".

The natural interleaving sets in mathematics

Integration. Substitution, parts, partial fractions, trig identity, recognition. The canonical case, these look similar and differ in what you notice first.

Differentiation. Product, quotient, chain, implicit, logarithmic. Same structure of difficulty.

Solving equations. Factorising, quadratic formula, completing the square, substitution, graphical. Students learn "the method for this chapter" and never choose.

Series. Which convergence test? This is almost entirely a selection problem and is taught as a list of tests.

Probability. Conditional, independent, complementary, binomial, normal approximation. Selecting is the whole skill and word problems deliberately obscure it.

Trigonometry. Which identity? A mixed set is worth ten blocked ones.

Vectors. Dot product, cross product, projection, or a scalar equation, the choice is the content.

Proof. Direct, contradiction, induction, counterexample. Undergraduates often meet these blocked and then face an exam that asks "prove or disprove".

When not to interleave

Important, because interleaving is oversold:

• Not on first contact. You need enough competence with each method individually to have something to choose between. Block briefly, then interleave.
• Not for automating a procedure. If you're building speed on one technique, differentiating polynomials, say, blocked repetition is correct.

Block to acquire, interleave to discriminate.

Pitfalls

1. Labelling the problems. Keeps the discomfort, removes the benefit.
2. Two types only. Alternation is a pattern you'll detect.
3. Quitting because accuracy dropped. That's the expected result. Judge it a week later.
4. Not stating the method first. Without the conscious selection you're just doing a shuffled problem set.
5. The tell: your accuracy on mixed sets matches your accuracy on blocked sets. It isn't mixed enough, or you're recognising types from phrasing.

Try this today

Take three integration methods you've studied separately. Ask for twelve mixed, unlabelled problems.

Before each one, say which method you'll use and why, out loud, before touching it. Count only the selection errors.

That number is the one your blocked practice never gave you.