The Science of Learning Movement Skills

One of my ongoing projects is to learn more about how research in learning is applied within specific fields. You can read my review of books summarizing the literature on language learning here and learning physics here.

To that end, I read Motor Learning and Performance, written by the eminent researchers Richard Schmidt and Craig Wrisberg.

This textbook is wide-ranging and full of interesting tidbits that don’t fit neatly into the overarching theory the authors propose. (For instance, did you know that movement accuracy tends to get worse as we move faster—except this trend reverses when muscles are above 70% of their peak force. Thus, if you want to strike a ball more accurately, surprisingly, you’ll do better if you swing your hardest!)

But instead of digging into exceptions, today, I’d like to review the central paradigm of motor skills argued for by the authors and suggest some implications for improving how we learn to move.

But First, What Exactly is a Motor Skill?

We all know that learning to do algebra is different from learning to play tennis. But what, exactly, is the difference?

In a sense, all skills we learn are movement skills. Even writing, hardly the typical domain of high-school jocks, is only possible through coordinated movement of your fingers to produce the pencil marks or keystrokes needed.

Similarly, few athletic skills are entirely devoid of intellectual content. Deciding how to return a tricky tennis serve or figuring out the best path to ski down a mountain all require fast, sophisticated judgements. Finesse, not just fitness, is central to athleticism.

Performing any skill, whether it’s athletic or intellectual, breaks down into roughly three components:

• Perception. Information from the outside world and sensations from inside your body must be processed to interpret and understand the situation you face.
• Decision. Memory and information processing must combine to figure out what you should do.
• Action. The skill must be executed through moving the body, which can be as simple as uttering a command to a subordinate or writing the answer to a calculation, or as complicated as as playing a solo concerto.

The domain of explicit motor skills, then, involves situations where the third element is a substantial point of difficulty, because the action requires high degrees of speed, accuracy or physical strength.

My overall impression from Schmidt and Wrisberg is that there is an essential continuity between learning movement skills and learning other kinds of skills. Therefore, instead of reviewing all the elements of their theory that overlap with what I’ve discussed elsewhere, I’ll focus on the aspects more particular to motor skills.

The Conceptual Model of Motor Skills

Schmidt and Wrisberg’s textbook continually builds on a central diagram that illustrates the overall theory of how we perform motor skills:

There’s a lot to unpack in this image, but the basic idea is that movement involves three stages: perception, decision and action, which are embedded in various feedback loops with your body and the environment.

A key element in this diagram is the importance of timing. Signals about what you perceive and how to move must travel between your brain and muscles. Relaying this information takes time. These physiological limits restrict what kind of feedback and decision-making processes can take place during the execution of any particular movement.

Schmidt and Wrisberg describe multiple forms of feedback that occur at different timescales:

• Spinal cord reflexes. The quickest loop occurs in the short-latency reflex (SLR, sometimes called the M1 response). Here an unexpected change in muscular contraction sends a signal up your nerve to the spinal cord, where a single synaptic connection sends back an appropriate motor nerve response. It takes only 30-50 milliseconds and is both unconscious and inflexible.
• Prepared reflexes. Taking 50-80 milliseconds, the long-latency reflex, sometimes called the M2 response, is slower than the SLR but more amenable to deliberate preparation. Here the instructions to “hold the weight steady” or “let go if you feel additional pressure” would modulate the response. But, like SLR, it is still an unconscious reflex with limited flexibility.
• Voluntary adjustments. After 120-180 milliseconds, information has time to travel to the brain and receive deeper processing. Hick’s Law, which relates the delay in reaction time to the number of possible choices, operates here, suggesting that cognition is now involved (even if the fastest actions may not have much conscious deliberation).

For skills that take place over a longer timeframe, like threading a needle, we can use a closed-loop system of feedback, where the full range of sensations can be used to adjust our movements while performing the task.

In contrast, for skills that take place over short intervals of time, feedback is too slow. Thus, our brains need to plan the entire action in advance, with limited possibility for adjustment if those actions turn out to be incorrect.

For instance, a baseball pitch can travel up to 90 miles per hour, meaning the entire time between the ball being thrown and it reaching the plate is less than 500 milliseconds. The batter needs 120-180 milliseconds for voluntary movement preparation and another 140-160 milliseconds to swing the bat. That means the batter must decide if and how to swing the bat before the ball has traveled halfway to the plate!

Planning Movements in Advance: Generalized Motor Programs

The timing constraints on open-loop movements imply that much of our movements must be prepared in advance. One theory for how we do this is that we construct motor programs. These programs act like little scripts telling our muscles when to move in order to produce the right actions.

If the motor program theory is correct, it also has major implications for learning motor skills. Since motor programs are the building blocks of skilled action, learning motor skills likely involves acquiring a large library of these programs (as well as the perceptual and decision-making facility to employ them in the right situations).

What exactly is a motor program?

One possibility is easy to rule out. If motor programs are the building blocks of skill, they are not organized in terms of explicit instructions for how to move each muscle.

Consider signing your name. This quick, fluent action is presumably stored in a motor program somewhere in your brain. The idiosyncrasies of this movement are what make your signature unique. If you sign your name in a checkbook and on a chalkboard, the two signatures maintain the same characteristics.

However, if you think about it carefully, the muscles involved in making the movements are completely different—writing on a checkbook mainly involves moving your fingers and wrist, whereas writing on a chalkboard mainly involves moving your shoulder and elbow while your wrist and hand stay largely fixed.

Thus, whatever a motor program is, it has to be more abstract than simple commands to contract particular muscles. It has to represent the idea or desired outcome of a movement, while presumably lower-level parts of the central nervous system are charged with implementing it.

Schmidt’s contribution to this theory was the notion of a generalized motor program. He argues that motor programs are stored in the brain as abstract structures. Some of the aspects of the programs are fixed, but there are also parameters that we can adjust on the fly to modify the movement for the current situation.

What aspects of motor programs are fixed, and which are free parameters?

We’ve already explained that the exact muscles involved in producing a particular movement are probably a free parameter (explaining the identical signatures on chalkboards and checkbooks). Amplitude is probably another (write the same signature big or small). Force, speed and trajectory are also factors that look like free parameters, rather than being fixed.

One element that potentially does appear to be fixed is the rhythm and relative timing of a movement. In one experiment, participants learned a task in which they practiced pressing keys in a particular order under specific timing requirements. After hundreds of trials, participants were then asked to produce the sequence of keystrokes as fast as possible. While they shortened the overall time to perform the trained routine, the rhythm of key presses remained the same (even though they were not asked to reproduce the rhythm learned in training).

This suggests that changes to the relative timing of a complex motor program may require learning a new motor program, rather than simply applying a different set of parameter values to an existing one. A coach who wants a person to use a different rhythm of actions to produce a tennis serve may have a much bigger job ahead than the coach who just wants the player to hit harder or higher.

How Can We Learn Movement Skills More Efficiently?

Given the conceptual model Schmidt and Wrisberg present, and the theory of generalized motor programs, what can we say about learning movement skills?

Variable practice beats repetitive training for flexible skills.

One area of active research in both intellectual and motor skills is the value of varied practice. In many studies, variable practice results in more durable or generalizable learning than more repetitive forms of practice.

Two types of variability deserve note:

1. Random practice (vs. blocked). Suppose you need to practice both a forehand and backhand tennis stroke. One strategy would be to drill forehand shots for a while and then switch to backhand shots. Another would be to randomize which shot you need to take, mixing both types of action together. Research generally supports the idea that the latter practice schedule will be more effective for learning, even if it tends to result in worse immediate performance.
2. Varied practice (vs. consistent). In contrast to simply mixing together different types of tasks in training, varied practice involves changing up the aims of the trained movement. Consider hitting a golf ball at the driving range vs. playing a round of golf. At the driving range, you repeatedly hit the ball off the same tee, compared to hitting it from multiple locations to different distances along the course as you play. Varied practice tends to be more effective for generating more flexible motor programs that can adapt to new situations.

Random practice provides for more robust learning of the underlying program, and varied practice helps to generalize the motor program so it can be successfully parameterized in a wide variety of settings.

The main exception to this principle occurs in the very early stages of learning, when the movement is not yet fully understood. Cognitive load may be higher here, so adding extra complications may make it harder to grasp the underlying movement.

Identify the right amount (and kind) of feedback.

The importance of feedback is clear in the conceptual model Schmidt and Wrisberg discuss. For closed-loop skills and complex performances, we adjust our actions based on multiple loops of feedback from the environment.

Given the importance of feedback, it might seem that more is always better. But this is not the case. Schmidt and Wrisberg note a few constraints on feedback, noting where it can do more harm than good:

• Concurrent and instantaneous feedback may distort performance. Feedback provided during the execution of a skill may result in a different skill being learned than the one intended. As the authors write, “concurrent visual feedback is usually disastrous for learning,” adding, “completely different neural pathways are used.” Similarly, feedback that is given instantly after a performance (rather than after a few seconds of delay) may inhibit performers from learning and processing intrinsic signals from the environment directly.
• More feedback is better, but not on every attempt. Higher absolute levels of feedback tend to improve performance, but higher ratios of feedback to no-feedback attempts don’t always do so. Better results often occur with external feedback on only a portion of trials.
• Simpler tasks benefit from sparser feedback. Summary feedback, where feedback is aggregated over multiple trials, often outperforms feedback given after each attempt. The degree of aggregation, however, depends on complexity—novel or highly complex skills benefit from less aggregation, whereas simpler skills benefit from more.
• Only one piece of advice at a time. Corrective suggestions should be kept as simple as possible, so as to not overwhelm a performer’s limited working memory bandwidth.

Focus your attention outside your body.

Since the motor programs that form the basis for skilled movements embody an abstract “idea” of the movement, not specific commands to individual muscles, paying too much attention to your movements can be counterproductive.

A wide range of studies find that an external focus of attention, i.e., paying attention to the goal of the movement rather than the movement itself, is more successful for learning many different motor skills. For instance, in one study, golfers told to focus on the movement and weight of their club did better learning to make a chip shot than golfers told to focus on their grip and arm movements.

Despite this generally valuable advice, there’s still some uncertainty in the research literature about exactly which elements of the external focus deserve more attention. In one study, golfers learned to swing better when focusing on the movement of their club, rather than the resulting movement of the ball. Yet, in a different study tennis players learned better when they were told to focus on the trajectory of their shot, rather than the movement of the approaching ball.

Further Thoughts and Reading

Overall, I found Schmidt and Wrisberg’s textbook to be a good resource covering many basic principles of motor skills, especially in emphasizing some aspects that differ from the more academic and intellectual skills that I typically write about.