A learner finishes a practice session with a smooth, accurate attempt. That is useful information about performance at that moment. A stronger learning question is whether the skill remains available later, and whether it can be used under the conditions that matter.
For a Canadian physiotherapy student reading a research paper or discussing a supervised learning activity, this distinction changes which results deserve attention. It also prevents an unfair conclusion: a person who performs less well when a prompt disappears has not necessarily forgotten everything they practised.
This article teaches interpretation through an invented computer task. It does not prescribe a patient's practice schedule, recommend withdrawing necessary assistance or establish clinical competence.
Name the question before choosing the test
Acquisition performance describes what is observed during practice. A retention test revisits the practised skill after an interval. A transfer test examines performance under a deliberately changed task or condition. The delay and the change must be specified; the labels alone do not describe an adequate design.
Kantak and Winstein's focused review argues that delayed retention or transfer performance is more informative about motor learning than performance at the end of practice. Its abstract supports that distinction; it does not identify one universally correct delay or a clinical threshold for successful learning.
All three tests still produce observations of performance. Learning is inferred from a pattern of results, rather than directly measured as a visible quantity inside a person. Soderstrom and Bjork's integrative review explains why immediate success can mislead judgments about durable capability. A delayed test improves the question; it does not make fatigue, assistance, measurement error or task conditions irrelevant.
Worked example: the prompt changes the comparison
Consider a wholly fictional teaching dataset. Two cohorts of 12 adult volunteers practise moving a computer pointer through a target sequence. Each reported result is the cohort's mean accuracy score, expressed as the percentage of attempts meeting the same predeclared timing and accuracy rule. Higher scores are better.
Cohort A receives a highlighted route during practice. Cohort B practises without that highlight. Every test block uses the same number of attempts. Everyone completes every block; there is no target-task practice between sessions. No allocation method, participant-level data or uncertainty estimates are supplied.
At the immediate and delayed probes, the trained sequence, instructions, scoring and equipment are identical, and neither cohort receives the route highlight or corrective coaching. A separate transfer block changes the target spacing and uses the same scoring rule. These are educational assumptions, not a description of a real study.
| Observation | Cohort A | Cohort B | Conditions that matter |
|---|---|---|---|
| Baseline test | 40% | 41% | Trained sequence; no highlight |
| Final practice block | 92% | 81% | A has the highlight; B does not |
| Immediate test after practice | 70% | 76% | Trained sequence; no highlight for either cohort |
| Retention test 24 hours later | 68% | 74% | Same test conditions as the immediate test |
| Transfer block after retention | 57% | 63% | Changed target spacing; no highlight |
Start with the tempting headline: A finished practice 11 percentage points ahead of B, because 92 − 81 = 11. That calculation is correct. Describing A as the better learner is not supported by it. The practice conditions differ, and the result concerns supported execution at that time.
Now compare A's final practice block with its next-day result: 92 − 68 = 24 percentage points. Calling this “24 points of forgetting” combines at least two changes: elapsed time and removal of the highlight. It cannot isolate either contribution. Even the immediate drop from 92 to 70 does not prove a pure cue effect: the blocks occur in sequence, so practice, fatigue and order may also matter.
The more comparable immediate-to-delayed contrast is 70 to 68 for A and 76 to 74 for B. Each cohort's observed score decreases by two percentage points. These comparisons hold more test conditions constant, but do not prove that exactly the same amount of learning was lost. The observations have no supplied uncertainty estimates, and the example does not control every time-varying influence.
Separate persistence, superiority and generalization
Three useful statements follow from the table, with different limits.
Persistence: A's delayed score is 28 percentage points above its baseline; B's is 33 points above its baseline. Performance above the starting level remains observable under the specified test conditions. This is consistent with retained improvement. Without a suitable comparison design, the differences cannot be attributed entirely to the assigned practice condition.
Between-cohort comparison: B's retention score exceeds A's by six percentage points. That describes these cohort means. It does not establish a statistically reliable difference, identify which individuals benefited or show that practising without a highlight caused the difference. The allocation process is unknown. A six-point difference is also not a six-percent relative improvement: those are different calculations.
Transfer: B also scores six points higher on the changed-spacing task. The block asks a different question from retention. There is no pre-practice measurement of this transfer task and no untrained comparison group, so the table does not establish how much either cohort's transfer-task performance improved because of training. Nor does success with one new spacing establish generalization to everyday activities.
In this example, the retention block occurs before transfer. Any practice or fatigue from that earlier block is part of the transfer-test context. A research claim about transfer needs to describe that order, and consider whether its design separates the effect of training from the effect of intervening tests.
Make the test match the intended use
Removing a prompt is informative when the intended question concerns performance without that prompt. It is not automatically the right outcome for every task.
Imagine instead that a permanent on-screen guide is part of the intended workstation. A later test with that guide could be directly relevant to successful use of the workstation. A separate no-guide test would answer a narrower question about unsupported execution. Labelling the latter the only “real” learning would silently change the goal.
For clinical discussions, the same logic is a reason to state the intended conditions and consult the responsible supervisor. It is not permission for a student to remove a walking aid, communication support or safety assistance to create a harder test. A study's experimental manipulation is not an individual patient's assessment plan.
Likewise, difficulty is not an outcome measure. More errors during practice are not, by themselves, evidence of better learning. The review literature's distinction between immediate and later results does not justify making every task more difficult. Read the actual retention and transfer evidence, the population and the conditions before extending a conclusion.
Transfer exercise: repair the claim
A fictional report gives this sequence for one cohort practising another computer task:
- Baseline, without a route highlight: 45% accuracy.
- End of highlighted practice: 95%.
- Immediate test without the highlight: 70%.
- Test 48 hours later, with the highlight restored: 90%.
The report concludes: “Only five percentage points were forgotten, and the skill transferred successfully.”
Identify two problems and propose one additional observation. Then explain what that observation would still leave uncertain.
Feedback: The five-point calculation compares highlighted practice with a later highlighted test. It describes performance under those supported conditions; it does not quantify unsupported retention or isolate forgetting. The immediate no-highlight score cannot be compared cleanly with the later highlighted score to infer a 20-point recovery of independent skill.
There is also no stated transfer test: the report supplies no deliberately changed task or condition intended to test generalization. Restoring a practised highlight does not establish successful transfer simply because it happens later.
If unsupported execution is the question, a delayed no-highlight test matched to the immediate probe would add relevant evidence. Its timing, instructions and position relative to the highlighted test should be planned explicitly. Testing itself creates experience, so a probe inserted after further supported attempts is not equivalent to a probe before them. The added observation would still not establish long-term durability beyond the tested interval, causal superiority over another practice method or applicability to a patient population.
When reading your next motor-learning paper, mark four details beside each result: when, which task, which support and which comparison. Those details make a learning claim assessable.
Sources
- Kantak and Winstein (2012), Learning-performance distinction and memory processes for motor skills: a focused review and perspective — PubMed abstract consulted; publisher full text was not accessed for this article.
- Soderstrom and Bjork (2015), Learning Versus Performance: An Integrative Review — author-hosted full text; relevant conceptual and motor-learning sections consulted. This article's datasets and exercises are original constructions, not results reported in either review.