Practice feedback

Why does taking away the coach's feedback during practice improve performance a week later?

A coach who calls the result after every attempt is doing the job as everyone understands it. And through the session it works: the learner given feedback on every trial is visibly better than the learner given it on one trial in three.

Come back a week later, hand both the task with nobody watching, and the ranking inverts. The one told least performs best. That is strange, because the extra information was accurate, timely and free. Where can accurate information go wrong — and what is it displacing?

Ask first what a learner has to end up owning. Not a movement: a movement is different every time — different wind, different fatigue, different starting position. What has to be built is the capacity to tell, from inside, whether the attempt was right and what to change if it was not. The skill is a comparison, and the comparison has to run without a coach, because most of the time there will not be one.

Where does the raw material for that comparison come from? From the attempt itself: the felt path of the limb, the sound of the contact, where the eyes were when it left. Those signals arrive on every trial whether anybody speaks or not.

And now the mechanism. When an external verdict arrives immediately and reliably, what is the point of consulting them? The question is answered before you ask it. The learner reads the verdict, adjusts, and never has to interrogate their own sensory record — so the internal comparison, the thing that had to be built, is never exercised. The feedback did not add a benefit on top; it substituted for the process that produces the skill. That is the guidance hypothesis, named by Salmoni, Schmidt and Walter in the 1980s: the information guides performance while present, and is leaned on rather than learned from.

There is a second, less obvious cost. A verdict about the last attempt is partly about the last attempt's noise — the gust, the slip, the twitch. A learner who corrects toward every verdict is chasing that noise, making large trial-to-trial swings around a target that had not actually moved. Reduce the feedback and the learner is forced to average across several attempts before adjusting, which is closer to what the underlying pattern deserves.

Both accounts predict the same signature: better during practice, worse after it — and they make one prediction that separates guidance from ordinary explanations. If frequent feedback prevents an internal comparison from forming, then ask learners, before the verdict, to say how far off they think they were. The reduced-feedback group should be better at that estimate, even though they have been told less. That is the standard finding, and it is what would refute the account: were the frequent-feedback group equally good at judging their own errors and still worse at retention, guidance would be the wrong mechanism and the explanation would have to move to attention or consolidation instead.

Three refinements connect the steps, and one of them limits the finding sharply.

The first is that "less feedback" is not the intervention. Reducing frequency is only one of several schedules that produce the same retention advantage: summary feedback covering a block of trials, bandwidth feedback given only when the error exceeds a tolerance, self-controlled feedback the learner asks for. What they share is that the learner has to consult their own record before any verdict arrives. Frequency is a proxy for that; the mechanism is the required self-consultation.

The second is that this is a claim about retention and transfer, never about practice-session performance. Judging a coaching method by how the session looks is exactly the error the effect exposes, and the learner's own sense of progress misleads in the same direction — the smooth session feels like the productive one.

The third is a real boundary. Much of the original work used simple laboratory tasks, and the picture from complex skills is different: as a task's difficulty rises, learners generally need more feedback, not less, and the reduced-frequency advantage weakens or reverses. A beginner given almost nothing to go on does not build an internal reference; they build a stable wrong one.

Two honesty notes. The effect sizes vary a great deal across tasks and the original demonstrations used small samples, so this is not a result to quote with a number attached. And the wider motor-learning literature it sits in has had its own replication trouble — several well-publicised claims about autonomy, enhanced expectancies and self-controlled practice have failed in recent pre-registered attempts. The guidance effect is better supported than those, and it is still safest to hold it as a robust direction rather than a calibrated dose.

• Why a learner's own sense of how well practice is going tracks fluency rather than retention.
• How the same logic applies to instructional scaffolds that must be faded rather than simply removed.

Every term the collection defines is gathered in the glossary.