Learning by explaining

Why does preparing to teach a topic teach you more than preparing to be tested on it?

Tell a student they will be tested on Thursday and they will study. Tell them instead that on Thursday they must teach it to someone else, and they study differently — and, on average, end up knowing more, even when the teaching never happens and the test they eventually sit is the same one.

The usual explanation is that you have to know something better to teach it. But that describes the standard, not the mechanism: it tells us the bar is higher without saying what the learner does differently to clear it. So what actually changes between preparing to answer questions and preparing to be the person answering them?

The first suspect is effort. Perhaps teaching is simply more motivating — the prospect of standing in front of someone is more frightening than a mark, so people work harder. This is testable, and it does not survive well. If expectancy were the engine, then merely announcing "you will teach this" should produce most of the benefit. In the work of Fiorella and Mayer, the expectancy on its own bought only a small advantage; the substantial gains came from students who actually delivered the explanation. So the expectation matters mainly through what it makes you do.

Then what does it make you do? Consider what each kind of preparation is rehearsing.

A test can be met with an unordered bag. You will be asked six things and you must have six answers available; nothing requires that the answers be connected to one another. An explanation cannot be delivered as a bag. It has to be sequenced, which forces a decision the bag never demands: what depends on what, and therefore what has to come first. Building that spine is itself learning — it creates relations between items that studying-for-recall leaves lying unconnected.

Second, notice where explanations break. Texts are full of unstated steps; the author knew why step three followed step two and did not say. Reading over those gaps costs nothing. Speaking across one is impossible — you arrive at the seam and have to build the bridge yourself. That generative move is well studied under its own name, the self-explanation effect, and it accounts for a good share of what is happening here.

Third, and easily missed: a rehearsal of teaching is normally done without the text. It is therefore a retrieval attempt, and a merciless one, because a gap in an explanation announces itself the moment your voice stops. Reading, by contrast, produces fluency, and fluency is what learners mistake for knowledge. Which is the more useful signal — the feeling that the page made sense, or the sound of your own sentence failing halfway?

Fourth, teaching requires you to model a mind that does not yet know this. Where would a beginner stop? What would they find arbitrary? Asking that drives you back to precisely the assumptions you had skipped over, because those are the ones a novice cannot skip.

An honest caveat about all four: the individual mechanisms — retrieval, self-explanation, organisation — are well supported, while "learning by teaching" as a distinct effect is more mixed, with modest and inconsistent results across studies. It may be best understood as a convenient package that reliably triggers the others, rather than as a fifth ingredient of its own.

Three refinements are worth keeping.

The gain comes from generating the explanation, not from having one. Reading a beautifully clear explanation written by someone else is close to rereading; the work you learn from is the work you do.

Explaining to a real person adds something an imagined listener cannot: unanticipated questions, which probe exactly the places you did not think to probe. It also subtracts something, because a polite listener nods and lets a gap pass. The imagined beginner is harsher and less informative; the real one is gentler and better informed.

And the whole effect depends on the source being shut. Explaining with the material in front of you collapses back into fluent rereading, and it feels productive, which is the trap: the activity looks identical from outside and does almost none of the same work.

• Whether teaching a weaker student and teaching a peer produce different gains for the teacher.
• Why explaining to a computer agent that cannot ask questions still helps some learners.
• How this relates to writing, which imposes the same demand for a spine but removes the audience entirely.

Every term the collection defines is gathered in the glossary.