- RNP
- Motor Learning
- From practice to the field: why a skill transfers or not
- Transfer and contextual interference: what makes a session look good often keeps the movement from sticking
SeriesFrom practice to the field: why a skill transfers or not2 of 6
Transfer and contextual interference: what makes a session look good often keeps the movement from sticking
A good session doesn't tell you what will last. Retention, transfer of learning, contextual interference: what makes a movement hold, and how far.
The session that goes well and the session that produces learning are rarely the same one. The problem is that you only see the first.
Hello to you, movement professional,
Take a training log reread at the end of a cycle: six weeks of work on a change-of-direction plant step with a team-sport player. Line after line, everything goes up. The reference points stabilize, verbal feedback decreases, and the last session of the block is so clean you could have filmed it for a demo. Three weeks later, in a friendly match, the plant step reappears exactly as it was before the block, with the same hesitation and the same lost time. Nothing had been neglected. What had been measured, however, was not what everyone thought was being measured.
This gap has a name, the distinction between performance and learning, and it has been documented for more than thirty years (Schmidt and Bjork, 1992). What you observe during a session is momentary performance, sensitive to fatigue, to instructions, to context, to immediate repetition. What you are trying to produce is a lasting change in capability, which can only be read later and elsewhere. The two often move in the same direction, which makes confusing them comfortable. In one well-identified family of situations, they move in opposite directions, and that is where the blocks that are most satisfying to run produce the least.
The objection comes right away: this reasoning would let you botch your sessions and call it science, since anything that goes badly would become proof of deep learning. It is a fair objection, and the framework lends itself easily to that use. It falls, though, for a simple reason: the framework is only usable if it rests on a delayed measure, set in advance, that can contradict it. Without a planned retention test, nobody can tell a demanding block from a poorly built one, you included. The rest of this article hinges on that condition.
Retention and transfer of learning: what the session doesn't measure
Everyday vocabulary conflates two things. Acquisition performance is what happens during practice, session after session. It is directly observable, it feeds both your client's sense of progress and your own, and it is the only feedback most blocks ever collect.
Learning, on the other hand, is defined as a relatively lasting change in the capability to produce an action, and it is not directly observable. It is inferred through two delayed measures. Long-term retention means reassessing the same task after a period without practice, long enough for the transient effects of the session to have worn off. Transfer of learning means assessing a task or context that was not trained, and looking at what carried over.
These two measures are the only ones that decide. A block that improves acquisition performance without improving either of them has produced nothing more than a session effect. The trap lies in the nature of the feedback: acquisition answers you every day, retention answers you once, weeks later, when the block is over and the calendar has already moved on.
That is why the lens offered by ecological motor learning shifts the unit of analysis. What gets trained, in this view, is a system's capacity to produce a suitable solution under changing conditions, far more than an isolated movement you would repeat until it sets. That capacity lives in the full sensorimotor loop, sensory input included, and it degrades as soon as the testing situation departs from the training situation. The useful question, then, is how far it can depart.
Contextual interference: what the order of trials changes
The reference experiment, by Shea and Morgan (1979), is simple to describe. Participants learn three different movement patterns. One group practices them in blocked order: all trials of the first pattern, then all trials of the second. Another group practices them in random order, switching patterns from one trial to the next. The total number of trials is identical.
During acquisition, the blocked group performs better, often clearly so. On delayed retention and transfer tests, the ranking flips. This result has been reproduced in many laboratories since the late 1970s, on sequencing, aiming and timing tasks (Magill and Hall, 1990), and it is among the most replicated effects in the field under laboratory conditions.
In applied settings, that robustness erodes: the effect becomes smaller, less consistent, and vanishes in some protocols. A meta-analysis that separates the two worlds finds a markedly weaker effect in applied studies than in laboratory studies, and weaker in young learners than in adults (Brady, 2004). A review of studies conducted in applied settings notes that most of them do not observe the expected benefit (Barreiros, Figueiredo and Godinho, 2007). Three moderators keep coming up. Similarity between the tasks compared: when they rely on organizations that are too close, there is nothing to discriminate and the effect fades (Magill and Hall, 1990). Task complexity relative to the person's level: for a beginner on an already demanding task, mixing adds a load that is no longer productive. Practice duration: an effect that needs sufficient volume to appear cannot be tested over two sessions.
What the effect really says is that trial order is a training variable in the same way as load or volume, and that it acts on a dimension neither load nor volume touches. The remaining question is through what mechanism, because the mechanism decides what you can carry over elsewhere.
The RNP training space, free for life
Get access to RNP training, no strings attached.
The two candidate mechanisms, and why they are not equivalent
Two hypotheses have competed for the explanation since the early 1980s, and neither has won.
The first is about elaboration (Shea and Zimny, 1983). When several tasks coexist in working memory, the system compares them, contrasts them, and builds representations that are more distinct from one another. Blocked practice offers nothing to compare, and produces less differentiated traces, which are therefore harder to retrieve when the context changes.
The other hypothesis, reconstruction (Lee and Magill, 1983), starts from working memory rather than contrast. Between two trials of the same pattern, the action plan remains available and simply needs to be replayed. As soon as another pattern is interleaved, that plan is partly lost and the system has to rebuild it on every trial. This repeated reconstruction would be what does the work, by engaging the very processes that will have to be engaged later, in the field, without warning.
The second hypothesis is the more interesting one for practice, because it ties learning to the system's ordinary functioning. Read through the tools of motor control, and this remains an interpretation, rebuilding an action plan amounts to re-engaging the motor decision instead of short-circuiting it. It means re-issuing a command along with its efference copy, and therefore reformulating a prediction of the expected sensory consequences, then comparing that prediction with the actual feedback. Reafference only becomes informative if a prediction was made beforehand, and a system that replays the same plan on a loop is making a prediction it no longer needs to refine. Predictive motor control gets trained when it is wrong enough to correct, not when it is right without effort.
Both mechanisms predict the same result in most published protocols, which explains why they still coexist. They diverge on one testable point, and we will come back to it at the end.
Desirable difficulties: a difficulty only helps if it targets the right process
The general framework that encompasses these observations came from the psychology of memory before it reached movement. It holds that certain practice conditions degrade immediate performance while improving retention, and calls them desirable difficulties (Bjork, 1994). Spacing repetitions rather than massing them, reducing and delaying feedback, varying conditions, requiring information to be retrieved rather than reread: the list is short and consistent.
All the weight rests on the word desirable, and it is the one people forget fastest. A difficulty is only useful if it targets the process you want to train. Adding fatigue, noise, instability or time pressure makes a session harder without guaranteeing anything. State specificity is a legitimate topic with its own rationale, and it does not follow the same logic as contextual interference.
The challenge point framework (Guadagnoli and Lee, 2004) defines the boundary. It holds that a task's functional difficulty depends on the person's level, and that there is a zone where the information generated by a trial is maximal. Below it, the trial tells the system nothing new. Above it, the trial generates more information than the person can process, and the surplus is lost. It is a theoretical framework, debated and hard to quantify in a session, yet it remains a useful safeguard against the quickest reading of the principle, the one that consists in making everything harder for everyone.
The direction of attention belongs to the same level of tuning. Attentional focus changes how the person processes the trial, and two protocols that are identical on paper do not produce the same learning depending on what the instruction points to. It is one of the cheapest levers to manipulate, and it is worth setting early rather than last.
Transfer is specific long before it is general
The other half of the problem is older. The idea that training a general function would improve everything that depends on it was tested at the very start of the twentieth century (Thorndike and Woodworth, 1901), and it did not hold up. What transfers from one task to another depends on how many elements they genuinely share, and far transfer is rare.
This is what makes transfer of learning so hard to achieve in strength and conditioning, a point the dedicated article on strength covers from a complementary angle. There, the specificity principle is described from the side of mechanical parameters: coordination, velocity, contraction type. A second layer is added here, on the side of learning conditions: the sources of information available during practice become part of what has been learned. This is the specificity of practice hypothesis (Proteau, Marteniuk and Lévesque, 1992). On aiming tasks, a person who trained while seeing their movement loses more when that vision is removed than a person who trained without it, and this dependence on visual feedback increases with the amount of practice instead of fading.
This hypothesis has a direct consequence: when performance drops as an aid is removed at the end of a block, what the drop reveals is learning that included the aid. Frequent verbal feedback may be part of that. This is what the guidance hypothesis proposes (Salmoni, Schmidt and Walter, 1984), taken up by work on the frequency of knowledge of results (Winstein and Schmidt, 1990): feedback given on every trial would support acquisition and penalize retention, because the person leans on the external feedback instead of processing their own signals. Yet a recent meta-analysis (McKay et al., 2022) finds no clear effect of reduced frequency, across a body of studies judged to be underpowered. The hypothesis remains plausible, but the direct evidence is thin.
The same logic governs task specificity in balance work, where the gap between the exercise and the target situation is the best predictor of what will fail to carry over. Standing on one leg trains you to stand on one leg.
What should vary, and what should stay stable
The previous reasoning leaves a tension unresolved. If transfer is specific, practice should stay as close to the target as possible. If learning requires variation, practice should move away from it. Both are true, and they operate at different levels.
What benefits from varying are the conditions of execution: starting position, angle of approach, signal timing, opponent, surface, task order. What benefits from staying stable is the intention of the task and its informational structure, in other words what the person is trying to do and what tells them whether they are succeeding. Repetition without repetition describes exactly this regime: the goal stays the same, the motor solution changes on every trial, and it is the variation in solutions that builds the ability to find a new one.
The constraints-led approach provides the lever for organizing this variation without leaving it to chance: modify the task, the environment or the person to make certain solutions more likely, rather than instructing the expected movement. Representative learning design provides the criterion that keeps the variation from drifting: the information that drives the solution in the session must be the information that drives it in the real situation. An exercise can be highly variable and barely representative, and it will produce solutions that serve no purpose anywhere.
That leaves the internal constraint. Exploring solutions requires the ability to free up and then recouple degrees of freedom, and that exploration relies on the quality of the body schema. An imprecise map shrinks the space of accessible solutions, and variation then does little more than cycle through two or three already familiar patterns.
Sensory inputs are part of what transfers
This is the point the RNP reading adds, and it is rarely addressed in the motor learning literature, which focuses mainly on practice conditions while treating sensory input as a stable given.
If the information sources available during practice are incorporated into what is learned, then the person's sensory weighting at the moment of learning determines what they will be able to transfer. Someone who relies heavily on vision to organize their stance learns a solution that contains this dependency, and that solution degrades as soon as vision is busy elsewhere, which is the normal situation in combat and duel sports. The vestibular system provides the gravitational reference that makes this redistribution possible, and its level of involvement at the time of learning is not a neutral parameter.
Residual primitive reflexes come in through the same door, and that door needs to be named without being widened. In RNP, an active residue links a specific trigger to an imposed tone pattern. The ATNR, the asymmetric tonic neck reflex, is triggered by neck movement, with a cervical proprioceptive component and a vestibular component that RNP keeps both open, since turning the head engages the neck and the labyrinth in the same motion. The STNR, the symmetric tonic neck reflex, and the TLR, the tonic labyrinthine reflex, impose the same family of constraint in other planes. A reflex is defined by its triggering stimulus, never by everything that activates when it occurs, and this distinction prevents most misreadings in this area.
In this reading, adding variation to a solution space restricted by an internal constraint mostly makes the same responses cycle faster. In the logic of the training, you first identify where the lights are out, you turn them back on, and only then do you make it hold through variation.
What this changes in how you build a block
Three decisions follow, and they are made before the block rather than at its end.
The first is to set the measurement before the block: a retention test scheduled after a delay without practice, on the trained task, and a transfer situation that has never been practiced, both fixed before the first session. Without this planning, the block can be neither validated nor invalidated, and it will be judged on the last session, which is the worst information available.
The second is to split the block into two phases: a short, blocked phase when the person is discovering a new organization and has nothing to discriminate yet, then a switch to random order once the patterns exist. The switch is decided by the person's ability to produce the form without constantly thinking about it, not by a number of sessions.
The third is to accept that session performance will drop, and to say so before it drops. This is the hardest part in practice, because the person being coached judges their coach on the feeling of progress, and a phase of varied practice feels worse than a blocked phase. Announcing the drop before it happens turns it into an expected signal.
Retention and transfer: what is established and what remains open
The following are established facts: the dissociation between acquisition performance and learning, the validity of delayed retention and transfer tests as the only measures of learning, the existence of the contextual interference effect in the laboratory, and the limited nature of far transfer between loosely related tasks.
The following sit at an intermediate level: the size of the contextual interference effect in applied settings, which is markedly smaller and more scattered than in the laboratory; the effect of feedback frequency on retention, which the most recent meta-analysis does not find robustly; identifying which mechanism is responsible, elaboration or action-plan reconstruction; the operational definition of the challenge point, whose value nobody knows how to set in a session; and the transposition to movement of a framework born in verbal memory research.
The following are field extrapolations, acknowledged as such: the three block-building decisions proposed above, the criterion for switching from blocked to random, and the reading that an internal constraint restricts the solution space accessible to variation. None of these three has been validated as a protocol.
On causality, observing that athletes trained in varied environments transfer better does not show that variation produces transfer. Total practice volume, the diversity of sports played in childhood and club selection act on both terms at once, and this common origin is enough to create the association without any causal link existing.
One last falsifiable marker can separate the two mechanisms. If the benefit of random practice comes from action-plan reconstruction, then blocked practice in which an unrelated activity, demanding enough to wipe the plan, is inserted between two trials should improve retention almost as much as random practice. If the gain only appears when the inserted activity forces a comparison between patterns, then elaboration is carrying the effect, with different practical consequences for how sets are put together.
RNP is not a medical procedure: it complements healthcare follow-up, it does not replace it.
What feels good in the session is judged far from the session, or not judged at all.
👉 I want to intervene in the right place.
Frequently asked questions about transfer and contextual interference
What is the difference between performance and motor learning?
Performance is what you observe during practice, at a given moment, under the influence of fatigue, instructions and immediate repetition. Learning is a lasting change in the ability to produce the action, and it can only be inferred through delayed retention and transfer tests. The two come apart in a well-documented family of situations (Schmidt and Bjork, 1992).
How do you measure transfer of learning in practice?
By defining, before the block, a situation that has never been practiced but is close to the target, and administering it after a delay without practice. The only non-negotiable point is that the assessment situation is set before the block; otherwise it will be chosen after the fact from among those that confirm the desired result.
Is random practice always better than blocked practice?
No. The advantage appears when the tasks being compared are sufficiently different, when the person already has a rough version of the patterns involved, and when practice volume is sufficient. In applied settings it is markedly smaller than in the laboratory, and smaller in young learners than in adults (Brady, 2004). For a beginner on an already demanding task, blocked order remains a reasonable starting point, and the switch to random is decided afterwards.
Should verbal feedback be removed during learning?
Nothing requires removing it. The guidance hypothesis holds that external feedback given on every trial supports session performance and penalizes retention, because the person leans on that feedback instead of processing their own signals. The most recent meta-analysis on feedback frequency (McKay et al., 2022) does not find this effect robustly: reducing and delaying feedback remains a reasonable option to check with a retention test, not a proven rule.
Where do residual primitive reflexes come into this?
In RNP, an active residue imposes a tone pattern tied to a given trigger: neck movement for the ATNR and the STNR, head position in space for the TLR. This pattern restricts the solution space that variation can explore. It is a reading framework consistent with known mechanisms, not an established experimental finding.
Sources
Schmidt R.A., Bjork R.A. (1992). New conceptualizations of practice: common principles in three paradigms suggest new concepts for training. Psychological Science, 3(4), 207-218.
Shea J.B., Morgan R.L. (1979). Contextual interference effects on the acquisition, retention, and transfer of a motor skill. Journal of Experimental Psychology: Human Learning and Memory, 5(2), 179-187.
Shea J.B., Zimny S.T. (1983). Context effects in memory and learning movement information. In R.A. Magill (Ed.), Memory and Control of Action (pp. 345-366). Amsterdam: North-Holland.
Lee T.D., Magill R.A. (1983). The locus of contextual interference in motor-skill acquisition. Journal of Experimental Psychology: Learning, Memory, and Cognition, 9(4), 730-746.
Magill R.A., Hall K.G. (1990). A review of the contextual interference effect in motor skill acquisition. Human Movement Science, 9(3-5), 241-289.
Brady F. (2004). Contextual interference: a meta-analytic study. Perceptual and Motor Skills, 99(1), 116-126.
Barreiros J., Figueiredo T., Godinho M. (2007). The contextual interference effect in applied settings. European Physical Education Review, 13(2), 195-208.
Bjork R.A. (1994). Memory and metamemory considerations in the training of human beings. In J. Metcalfe, A.P. Shimamura (Eds.), Metacognition: Knowing about Knowing (pp. 185-205). Cambridge (MA): MIT Press.
Guadagnoli M.A., Lee T.D. (2004). Challenge point: a framework for conceptualizing the effects of various practice conditions in motor learning. Journal of Motor Behavior, 36(2), 212-224.
Thorndike E.L., Woodworth R.S. (1901). The influence of improvement in one mental function upon the efficiency of other functions (I). Psychological Review, 8(3), 247-261.
Proteau L., Marteniuk R.G., Lévesque L. (1992). A sensorimotor basis for motor learning: evidence indicating specificity of practice. Quarterly Journal of Experimental Psychology Section A, 44(3), 557-575.
Salmoni A.W., Schmidt R.A., Walter C.B. (1984). Knowledge of results and motor learning: a review and critical reappraisal. Psychological Bulletin, 95(3), 355-386.
Winstein C.J., Schmidt R.A. (1990). Reduced frequency of knowledge of results enhances motor skill learning. Journal of Experimental Psychology: Learning, Memory, and Cognition, 16(4), 677-691.
McKay B., Hussien J., Vinh M.-A., Mir-Orefice A., Brooks H., Ste-Marie D.M. (2022). Meta-analysis of the reduced relative feedback frequency effect on motor learning and performance. Psychology of Sport and Exercise, 61, 102165.
By the LabO RNP team