Feedback timing

Why can telling a student the answer too quickly weaken what they learn?

Fast feedback sounds unarguable. Catch the error before it sets, correct it while the attempt is fresh, and do not let anyone practise a mistake. Software that grades a keystroke at a time is sold on exactly this logic. And yet, in a fair number of experiments, learners given the answer immediately do worse on a test two days later than learners made to wait. They also feel they learned more. Something in the intuition is off — so what exactly does the delay give a learner that speed takes away?

Begin by asking what a learner is doing in the gap between attempting an answer and being told. It looks like nothing much. Actually it is a search: rummaging through what is stored, half-finding it, testing a partial answer against the question. That search is effortful and often unsuccessful.

Now notice a fact that reorders the whole problem. Retrieving something from memory is not a neutral read. It changes the memory — successfully pulling a fact back makes it more retrievable later, and the harder the pull, the larger the strengthening.

So consider what immediate feedback does. The verdict arrives before the search has run — often before it has properly begun. The learner reads the answer, recognises it, and the recognition feels like knowing. But the effortful retrieval never happened, so the strengthening it would have produced never happened either. Speed did not merely fail to help; it deleted the mechanism.

There is a second, older explanation with different machinery, and it is worth keeping both. If feedback comes instantly, a learner's wrong answer is still vividly in mind and competes with the correction. Wait a while and the error has faded, so the correction arrives with less to fight — and the reader gets a second, spaced encounter with the material rather than one continuous episode. Interference reduced, spacing gained.

Both accounts predict the same odd pattern, and the pattern is what matters: immediate feedback often produces better performance during practice and worse retention afterwards. That signature turns up outside the classroom too. In motor learning, frequent immediate information about each attempt improves performers while it continues and leaves them worse once it is withdrawn — they have been leaning on it instead of building their own error detection.

Which points at the deeper idea. Conditions that make practice feel smooth and successful are not the same as conditions that make learning last. Slowing acquisition down — by spacing, by interleaving, by withholding the answer a while — often costs performance now and buys retention later. These are the desirable difficulties, and the awkward part is the word desirable: the learner experiences them as the method not working.

The result is real but conditional, and stating it as a rule — "delay your feedback" — gets it wrong in an important class of cases.

Delay helps when there is something to retrieve. If a learner has partial knowledge, a gap forces a search, and the search is the whole benefit. If they have none — an unfamiliar term, a convention with no derivable answer — then waiting produces no search, only a longer period of not knowing, and immediate telling is better.

Delay also assumes the feedback will actually be processed when it comes. A verdict handed back a week later, on work the learner no longer remembers producing, is not delayed feedback in the useful sense; it is feedback that arrives after the learner has stopped caring about the question. The effect requires the learner to re-engage.

And there is a case where immediacy plainly wins: when the attempt itself is being encoded. A wrong grip, a mis-formed vowel, an unsafe procedure — if repetition of the action is what practice consists of, letting the wrong version run is expensive to undo, and correction should come at once. The retrieval argument concerns learning facts and concepts, where the wrong answer is a guess to be revised, not a movement being trained into the body.

Two further honesty notes. The delayed-feedback advantage is not universal even in the fact-learning studies; it varies with materials, with the length of the delay, and with whether the final test is soon or distant. And a large meta-analysis of feedback interventions found that a substantial minority of them made performance worse, chiefly when the feedback drew attention to the person rather than the task — a reminder that timing is only one dial, and not the most important one.

• How the guidance hypothesis explains why frequent feedback in motor practice degrades performance once withdrawn.
• Why learners' own judgments of how well they are learning track fluency rather than retention.

Every term the collection defines is gathered in the glossary.