Why Therapy Is Designed That Way — An Introduction to Motor Learning Principles

There is a set shape to practice in the rehabilitation clinic. How many times you perform the same movement, in what order you mix various tasks, and when and what the therapist tells you are all part of the design. This article covers the principles that design relies on. And we first look at how sparsely papers that tested those principles describe how they practiced.

The primer "Neuroplasticity — A Brain That Builds New Roads Beside the Broken Ones" introduced motor learning as the process by which movement skill changes to last a long time through repeated practice. Task specificity—the property where the exact movement and activity practiced improves the most—was also covered there. That piece got up to what to practice. This article takes the next step, looking at how to practice the same task. That means in what order they are done, how time is divided, and what feedback is given.

The immediately preceding piece, "When Does Recovery Happen — What Time Did, and What Treatment Did", also ended by leaving the same question. In a trial that got acute stroke patients moving out of bed early, the number of times out of bed per day (session frequency) and the total amount moved per day (daily dose) had directions associated with good outcomes that were opposite each other. Papers that report only "total treatment time" erase this distinction. How to divide the same amount is one of the practice conditions this article covers.

Before we look at those principles, there is a fact we must know first. Stroke arm training trial papers describe only about half or even fewer of the details on in what order, how divided, and with what feedback practice was done. This article starts from there.

Why We Need to Know — Practice Not Written Down Cannot Be Copied

Practice conditions are design elements that determine how a given task is practiced. This includes how many times it is performed, in what sequence it is mixed, when rests are taken, and what information is provided.

An example of a treatment that bundles these conditions together is arm training after stroke. Upper Limb Task-Oriented Training (UL-TOT) is a complex intervention that raises upper limb motor recovery by applying practice conditions related to motor learning principles (PMID 41238370). The upper limb means the arm and hand, and a complex intervention is a treatment where multiple elements intertwine to operate as a single unit. Because such treatments have many elements, they must be reported in detail so other research teams can try the same treatment over again. TIDieR, a reporting guideline telling researchers to write down an intervention so others can copy it exactly, requires this (PMID 41238370).

A 2025 scoping review measured how well this requirement is kept. A scoping review is a review that draws a map of what studies on a topic covered and to what extent. Searching five databases from 2000 through 2024 yielded 23,802 papers, leaving 166 (PMID 41238370).

Three unfamiliar terms appearing in the table are unpacked first. Practice variability is the degree to which the same task is practiced while changing conditions. Practice order is whether multiple tasks are practiced clustered one by one or mixed together, and practice distribution is whether practice time is crammed into one session or divided with rests. Feedback is the information returned from the outside regarding the movement or its result.

Lining up the reporting rates from highest to lowest yields the following (PMID 41238370).

The three items consistently sagging at the bottom are practice order, practice distribution, and feedback type. While what was chosen and how much was practiced are almost all written down, in what order, how it was divided, and what was returned are written down only around half the time or even less. The reviewers conclude that practice distribution, order, and feedback were reported haphazardly, and propose creating a standard checklist for writing down practice conditions. Only then can researchers repeat this treatment and apply it reliably to specific patient groups (PMID 41238370).

The same blank spaces are seen in swallowing rehabilitation as well. As seen in the primer, neuroplasticity is the ability of the brain to reorganize itself by creating new neural connections, and researchers have organized the conditions under which this happens well into 10 principles. There is a review that compared 20 swallowing exercise protocols in healthy older adults against these principles and motor learning principles one by one. "Use it or lose it", "repetition matters", and "salience matters" were well kept. On the other hand, gaps were identified in the application of the concepts of intensity, specificity, and transference. Transference is when practiced skill carries over to unpracticed tasks or situations. Among the five practice conditions of motor learning, except for target complexity, there were gaps from the recommendation guidelines, and the same gaps were confirmed across all four feedback conditions (PMID 41176157). This means it is not just an arm training problem, and the empty appearance of feedback overlaps across both fields.

Rather than memorizing principles, the habit of checking whether those conditions are written in the paper you are reading comes first.

The Big Picture — Four Axes for Designing Practice

In motor learning, therapists can manipulate roughly four things.

• What is practiced — task selection, task specificity, transfer
• How much is done — dose, number of repetitions, intensity. As we saw in the primer, dose is how much practice is performed, and is typically measured by practice time or repetition count. This axis is the subject of the next piece, "How Much Is Enough — An Introduction to Dose-Response in Rehabilitation".
• How it is arranged — practice order (random practice, where tasks are mixed, versus blocked practice, where one task is done all at once), practice distribution (massed practice, done all at once, versus distributed practice, broken up into parts), variability, and difficulty progression
• What is given back — type, timing, and frequency of feedback

Comparing this with the previous table, what gets written down haphazardly is the practice order and practice distribution of the third axis, as well as the entire fourth axis. Even within the third axis, variability and difficulty progression are recorded well. But the core claim of this field is that those unrecorded conditions separate learning from non-learning. Contextual interference, which we will see later, is an example of how a single practice order changes the remaining skill.

Core Concepts — Skill in Practice and Skill That Stays

Performance is the skill seen right in the middle of practice, and learning is the change that remains even after practice ends. Doing well during practice and what stays behind later are separate things. That is why motor learning research does not evaluate by performance at the end of practice, but by retention and transfer. Retention is measuring after stopping practice and letting time pass to see how much skill remains, and transfer is seeing whether that skill appears in untrained tasks or conditions. Without this distinction, "getting better during treatment" is misread as "having learned."

Contextual interference is a phenomenon where practicing multiple tasks mixed in a random order makes performance look worse during practice, but leaves more staying power later. On the opposite side is blocked practice, where one task is done entirely in a row before moving on to the next. Because mixing them makes things look less smooth during practice, blocked practice looks better if you only measure performance, and the difference only shows up when you measure retention or transfer.

Explicit learning is learning where you consciously build a strategy and correct errors, and implicit learning is learning where the relationship between sensation and movement is readjusted without you being conscious of it. Looking at a screen to see that your stride is short and deliberately widening your step is the former, and your body naturally adjusting its stride while doing so is the latter. In daily life and rehabilitation, the two are usually used together (PMID 39230337).

There are also other branches of learning. Motor adaptation is adjusting a movement you already know to fit changed conditions, and de novo learning is establishing a control method from scratch that did not exist before. The two types have long been thought to differ in both behavior and brain mechanisms (PMID 38632252). One study trained the same participants on both tasks. One was a rotation task (adaptation) where hand movements appeared rotated by a certain angle on a screen, and the other was a mirror-reversal task (de novo learning) where left and right appeared reversed like a mirror. There was no difference based on training order, and no basis for a difference was found in learning rate and asymptote. Learning rate is how fast you improve, and asymptote is the level you reach as you continue practice. On the other hand, in mirror-reversal learning, the variation among participants was larger and movement initiation was slower. Aftereffects were observed only after rotation training (PMID 38632252). An aftereffect is the trace of the body continuing to move in the adjusted direction for a while even after the changed condition is removed. The authors write that these results broaden our understanding of what separates learning a new skill from adjusting a known movement (PMID 38632252). This means that even if the rate of improvement looks the same, the remaining traces and the variation between people can differ depending on the type of learning.

What It Looks Like in Practice — Does Contextual Interference Show Up in the Clinic Too?

The effect that mixing up practice is better was clear in the laboratory, but smaller and not statistically confirmed in real-world settings (PMID 39205981).

Since the first study in 1966, the effect of practice schedules on retention and transfer has been studied, accompanied by both support and skepticism. A 2024 meta-analysis sorted this out. A meta-analysis is an analysis that gathers the results of multiple studies and recalculates them together. It found 1,287 papers, performed full-text screening on 300, included 42 in the systematic review, and put 34 into the meta-analysis (PMID 39205981).

The results are written as SMD (standardized mean difference). SMD is a value obtained by dividing the difference between the averages of two groups by the degree of data scatter, allowing results measured by different tests to be compared on a single yardstick. The larger the value, the greater the difference. "Statistically significant" means that it is hard to view the difference as having arisen by chance alone.

• The overall contextual interference effect on transfer was a moderate SMD of 0.55, favoring random practice.
• In both laboratory and real-world settings, the direction of the effect favored random practice.
• However, in laboratory studies, the moderate effect size was statistically significant (SMD 0.75), whereas in applied studies—that is, real-world studies—the effect size was small and not statistically significant (SMD 0.34).
• Age analysis was significant only in adults and older adults, and both groups favored random practice. The effect was moderate at 0.54 for adults, large at 1.28 for older adults, and negligible at 0.12 for younger participants.

From these results, two points touch directly on rehabilitation clinical practice.

First, the group with the largest effect is older adults (1.28). This age group overlaps with people receiving rehabilitation. However, this does not mean the participants in this analysis had neurological conditions.

Second is the gap between the laboratory at 0.75 and the real world at 0.34. The effect obtained under controlled laboratory conditions did not appear to the same extent in real-world studies outside the lab, and this gap is the real problem of this entire tier. It means that the effect diminishes at the bottleneck where the principle is transferred, rather than in the principle itself. The real-world studies in this meta-analysis do not necessarily mean rehabilitation treatment rooms, but the same bottleneck is crossed when moving into a treatment room. And as seen in the previous section, those treatment room conditions are described in only about half or fewer of the papers.

Common Misconceptions

There is a notion that "principles derived from healthy people can be applied directly to patients." A study investigating walking learning in chronic hemiparetic patients directly tested this assumption. Hemiparesis is a condition where one side of the body becomes weak or fails to move at all, as seen in the primer "What a Stroke Actually Is — Blocked, Burst, and the Clock", and chronic refers to the prolonged period after onset when recovery has slowed down.

The researchers measured the two types of learning separately. Implicit learning was measured using split-belt adaptation. A split-belt is a device where two treadmill belts move at different speeds, and walking on it causes the body to adjust its stride without conscious awareness. Explicit learning, namely the use of strategies, was measured by providing visual feedback during split-belt walking to have participants consciously correct stride errors. After 40 steps, the visual feedback was removed and participants were instructed to walk comfortably to minimize explicit contributions. Then, a multi-rate state-space model was used to separate the contributions of the two processes for each person (PMID 39230337). This model is a computational method that estimates outwardly visible changes in walking by dividing them into the sum of a fast-moving process and a slow-moving process.

When compared to age- and sex-matched neurologically normal control groups, chronic stroke patients showed deficits in both explicit and implicit contributions. The authors note that this result contrasts with previous studies that examined each process separately. They also add that because post-stroke gait rehabilitation relies on both types of learning, future work must uncover how to design rehabilitation to maximize overall motor learning (PMID 39230337). This means we cannot assume that one remaining type of learning will compensate for the other.

There is also a belief that "the more feedback is given, the better." Which form of feedback is most effective has not yet been settled. A study on children with cerebral palsy illustrates this. Cerebral palsy is a lifelong disability affecting movement and posture caused by damage to the developing brain. Augmented feedback is information added from outside what the body senses on its own, such as a therapist's words or results displayed on a screen. A systematic review of 12 papers on this topic notes that evidence supports the expectation that children generally benefit from feedback provided during or after movement performance. However, due to the heterogeneity of the studies, it is difficult to draw firm conclusions about the relative effectiveness between feedback types (PMID 35416108).

The same review makes two recommendations for rehabilitation practice: not providing feedback on every trial, and reducing the frequency while letting the child decide when to receive it (PMID 35416108). What is difficult to conclude is the superiority among feedback types, whereas the direction of reducing frequency is what the reviewers recommend for practice. Because this recommendation comes from studies on children with cerebral palsy, this review does not answer whether the same approach applies to other patient groups.

Returning to the reporting rate table from earlier, only 37% of the 166 papers recorded feedback frequency. This is part of the reason why it is difficult to summarize the relative effectiveness of feedback. Studies that do not report frequency cannot be compared with one another.

To Go Deeper

• When reading a paper — before looking at the results, look first in the methods for whether the practice order, practice distribution, type of feedback, timing, and frequency are written down. If they are not written down, that trial is hard to repeat as is, and hard to compare with other trials.
• When looking at the results — separate whether it is performance at the end of practice, or a retention or transfer assessment. The distinctions from the earlier articles dividing what it was measured with ("How Do We Measure Recovery — An Introduction to Rehabilitation Outcome Measures"), how deep the recovery is ("How Deep Is the Recovery — Between Compensation and True Recovery"), and whether time did it or treatment did it ("When Does Recovery Happen — What Time Did, and What Treatment Did") are used just the same here.
• When translating principles — do not use laboratory effect sizes as clinical expectations. Context interference is an example that showed that gap in numbers (0.75 versus 0.34).
• Upcoming articles — if this article dealt with "how," what remains is "how much." "How Much Is Enough — An Introduction to Dose-Response in Rehabilitation" covers what dose is measured with and how much is actually done in the treatment room. "Able But Unused — An Introduction to Learned Non-Use and CIMT" covers an example where these principles are bundled into a single treatment paradigm.

Source Papers

• Motor learning principles reported in stroke trials of upper limb task-oriented training: a scoping review. PMID 41238370 (2025)
• The effect of contextual interference on transfer in motor learning - a systematic review and meta-analysis. PMID 39205981 (2024)
• Explicit and implicit locomotor learning in individuals with chronic hemiparetic stroke. PMID 39230337 (2024)
• Principles of neural plasticity and motor learning in swallowing exercise protocols for older adults: A mapping review. PMID 41176157 (2025)
• Effectiveness of different extrinsic feedback forms on motor learning in children with cerebral palsy: a systematic review. PMID 35416108 (2023)
• Distinct learning, retention, and generalization patterns in de novo learning versus motor adaptation. PMID 38632252 (2024)