Why we need variability of movement

When it comes to producing skillful sport performance, we tend to think that we need to achieve a very consistent way of moving. In that perspective, less movement variability means better performance. This comes from a common assumption that consistent performance is an essential element of skill. But there are two components of performance: the movement, and the outcome. Traditional teaching methods focus on the movement, with ideal patterns that have to be learned, rehearsed, and reproduced. We therefore tend to treat deviations as errors, that come from some kind of lack of control. Here we will challenge these assumptions by focusing on the consistency of outcome. We will argue that variability is a necessary component of movement, allowing for better control, adaptability and learning. If we want consistent outcomes, we need variability of movement.

In this article, we will leave out of the discussion activities with aesthetic elements, like gymnastics or dance, where it is not obvious that there is an outcome separate from the movement itself. In these activities, the patterns of movement matter in a more direct fashion than say, football or athletics, so we will examine them in another article.

Variability is omnipresent #

To start, let us state the simple fact that variability of movement is everywhere, while not seeming to hinder our capacity to move with precision and efficiency. Even when rehearsing a technique, every attempt will look a bit different. A bit more speed, a bit less flexion at the knee. But luckily this does not make us incapable of attaining our objectives. While studying expert blacksmiths, the Russian neurophysiologist Nikolaï Bernstein remarked that the trajectory of the arm and the hammer where very different from strike to strike, while the blacksmiths were still capable of striking with precision. Bernstein coined the phrase “repetition without repetition”[1] : the blacksmiths were repeating the same gesture, but the patterns of movement were not identical. Because we move in complex and dynamical environments, every trial and every task are different. That our actions have some variability is therefore something we should expect. We struggle with repetitive and stereotypical movements, and they can be a sign of pathology, like in Parkinson’s disease[2]. This is also linked to what we call the degrees of freedom problem: we have so many different ways of moving available to us, how do we select one specific solution among them ?

Hammer strike cyclogram

Let’s keep this in mind: the trajectory of the strike of the blacksmiths was variable, but the endpoint was not. Results should be invariant, but this does not require invariable movements. Let’s take another example: in ping-pong, the variability of the position of the paddle is quite high at the beginning of the movement, in order to adapt to the position of the ball[3]. But the variability gets really small at the moment of the strike, because that is the moment when precision is essential. If this still seems a bit alien to you, let’s go into why this might happen and see if we can make it sound reasonable.

Variability allows for adaptation #

As we’ve already alluded to, there is some variability in the environments we interact with. We also have to face internal variability (usually called “noise”): this might come from factors like fatigue or the imperfection of transmissions in our nervous system. This means there cannot be a 1:1 relationship between plans of movements in our heads, and their executions. Some variability will happen eventually.

If we accept this, then there must be some way that we compensate for this internal and external variability. Take the blacksmiths: the only way they could use the same pattern of movement each time was if their arms were in the same starting position before every strike. But this would be very hard to do, because when they strike, the hammer rebounds in different ways on the surface. So they need some variability of movement to compensate for the noise, errors and uncontrolled rebounds. The arm takes different trajectories because it doesn’t always start at the same point, but the blacksmith still wants to keep the strike end-point constant. Let’s take another example: you’re trying to figure out your run-up in order to jump from a take-off board[4]. If you keep your stride length constant, you probably won’t end up with your foot on the board. In order to do that, you have to constantly adapt your stride length to your speed, and compensate for unpredictable factors like the wind, your fatigue, etc. One part of the system varies, so other parts must adapt to keep a precise and stable result. I call this the “funnel” model (see graph).

Two models of variability

This means that variability makes our movements more flexible and adaptable. It helps to compensate for internal (fatigue…) and external (wind…) changing factors. It allows us to attain our goals consistently in complex, changing and unpredictable contexts. It allows to react to perturbations and have alternative solutions at hand in cases of uncertainty. Let’s say your shoulder is injured: it would be better to perform using more of the elbow and less of the shoulder. Or the opponent you’re facing is uniquely tall: if you’re not able to adapt the trajectory of your throws, you won’t be very helpful for your team. In the blacksmith example, variability might also be functional to avoid overuse injuries, by putting muscles and joints into play differently at each strike.

The idea of the single ideal pattern of movement can also be put into question because it doesn’t seem like a very efficient strategy of controlling our movements. Think of it this way: your main goal is to strike the hammer with precision at the endpoint. Insuring that the trajectory is precise too is like adding a separate subgoal. Allowing for variability and flexibility is better, because it allows to put the maximum resources (attention, energy, etc.) into actually attaining our goals consistently, rather than attaining our goals while at the same time ensuring that our movements look the same[5]. Variability does not come from errors or lack of control: it comes from an optimal control strategy. Variability is tolerated as long as it does not interfere with task success. It might help to think of this as a way of solving the degrees of freedom problem: motor control is simplified because we don’t need to control every single independent element. Some can be left to vary and compensate for the variability of others. Only the global results matters. As Mark Latash puts it: “We are blessed with abundance [of movement solutions]; so, let us not waste time trying to eliminate it.”[6]

Variability can enhance learning #

There is also some evidence that variability improves learning. A study has shown that learners with a higher level of variability learned faster than those with lower variability[7]. There is also an experimental method called differential learning[8], which was also shown to be pretty effective for motor learning. The basic idea of this method is that every trial is done with a random variation of the movement (right arm at 45°, left leg extended, etc.). There is no repetition or reproduction of an ideal model and no feedback or demonstration, but still, the learners improve their performance. We can see variability as a sign of exploration of different movement solutions, while low variability can mean being stuck in inefficient or inflexible solutions.

Brisson et al. have also shown that sometimes it’s not even possible to identify an optimal pattern of movement for a certain task[9]. And when they compared learning by trying to emulate the pattern of the subject with the best results, versus trying to reproduce the pattern you used in your previous personal best attempt, the results were slightly better in the second case.  In another study, Brisson et al. showed that you don’t necessarily need to identify optimal patterns of movement to give useful feedback to learners[10]. When given information about the way they performed and of their results, they were able to explore different patterns and discover those that gave them the best results. It is not obvious then that we should try to imitate the patterns of expert athletes.

What are the implications ? #

With all the above, we should now understand that prescribing precise and stereotypical movement solutions risks being over constraining. Let’s say you insist that your learners run with a precise stride length: there will be a trade-off, where being precise while running will require compensatory variability and might diminish precision at the take-off board (the end-point). I would call this the “sieve” model (see graph), for lack of a better term. And if you give your learners a strict pattern to imitate, there will be less exploration, and therefore less chance to find novel solutions. This has consequences for the flexibility and adaptability of movement, and maybe for creativity too[11].

I would advocate for a principle of minimal intervention[12]. We should identify parameters that are most relevant for task success and safety, and only correct these parameters while being tolerant of variability in the other parameters. Let’s say we find out that in landings, knees caving in puts the learner at a high risk of injury[13]: this is a parameter that we need to keep in check. But on the other side it doesn’t matter if your learner bends his knees at 33° or 95° when landing: it only matters that he dissipates forces safely. Focusing him on the reduction of his landing sound, or having him land barefoot would be two useful methods for this. Which bring us into the territory of external focus of attention and the constraints-led approach. Instead of focusing on the body and specific patterns of movement, we should be focused on the tangible effects of our actions. This should allow for more self-organisation and variability in movement, while keeping the outcome stable. Instead of prescribing patterns of movement, we should design learning situations that can be actively explored, with constraints that guide learners towards the most efficient solutions. We can also try to induce perturbations by changing these constraints, in order to force learners to explore different solutions than those they’re getting (too) comfortable with. We should probably also avoid repetitive drills in simplified contexts, and allow for more practise in complex and changing situations. This can be achieved by using realistic environments, tasks with opponents or partners, learning different skills together rather than in isolation, or adding some randomness (like the differential learning approach, see above).

Conclusion #

In this article, we have shown that variability of movement is omnipresent. It is also necessary for the flexibility and adaptation of movement: it allows for consistent outcomes in variable environments. It also seems to play a role in learning. All of this suggests we should be tolerant of variability, abandon repetitive drills and the imitation of ideal patterns of movement, and not mistake the means (technique) for the end (functional outcomes).

There seems to be some exceptions though. What about disciplines and activities which are not directed to outcomes in the environment, but rather focus on the body ? Or to put it differently, where the goal of the movement is the movement itself ? Surely, in dance or gymnastics, we want to rehearse the movements so they look a specific way, so variability of the pattern of movements does not seem to be desirable. This, we will leave for another article. But I’ll end with a question. Think of activities where the movement outcome is important, like football, parkour or athletics, or even everyday life. Are there situations were we, as practitioners, coaches or spectators, focus on the pattern of movements rather than the outcome ? And if so, why ?

  1. Bernstein N, « On dexterity and its development », in: Latash Mark L. et Turvey Michael T. (éds), Dexterity and its development, Mahwah, N.J, L. Erlbaum Associates, 1996. ↩︎

  2. Davids K., Button Chris et Bennett Simon, Dynamics of skill acquisition: a constraints-led approach, Champaign, IL, Human Kinetics, 2008. ↩︎

  3. Bootsma Reinoud J. et Wieringen Piet C. W. van, « Timing an attacking forehand drive in table tennis », Journal of Experimental Psychology: Human Perception and Performance 16 (1), 1990, pp. 21‑29. ↩︎

  4. Montagne Gilles, « Le contrôle des mouvements finalisés en sport », Bulletin de psychologie Numéro 475 (1), 2005, pp. 7‑10. ↩︎

  5. Liu Dan et Todorov Emanuel, « Evidence for the Flexible Sensorimotor Strategies Predicted by Optimal Feedback Control », Journal of Neuroscience 27 (35), 2007, pp. 9354‑9368. ↩︎

  6. Latash Mark L., « The bliss (not the problem) of motor abundance (not redundancy) », Experimental Brain Research 217 (1), 2012, pp. 1‑5. ↩︎

  7. Wu Howard G., Miyamoto Yohsuke R., Castro Luis Nicolas Gonzalez et al., « Temporal structure of motor variability is dynamically regulated and predicts motor learning ability », Nature Neuroscience 17 (2), 2014, pp. 312‑321. ↩︎

  8. Schöllhorn Wolfgang Immanuel, Beckmann Hendrik, Janssen Daniel et al., « Stochastic perturbations in athletic field events enhance skill acquisition », in: Renshaw Ian, Davids K. et Savelsbergh Geert J. P. (éds), Motor learning in practice: a constraints-led approach, 1st ed, London ; New York, Routledge, 2010, pp. 69‑82. ↩︎

  9. Brisson T. A. et Alain C., « Should Common Optimal Movement Patterns Be Identified as the Criterion to Be Achieved? », Journal of Motor Behavior 28 (3), 1996, pp. 211‑223. ↩︎

  10. Brisson Therese A. et Alain Claude, « Optimal Movement Pattern Characteristics are Not Required as a Reference for Knowledge of Performance », Research Quarterly for Exercise and Sport 67 (4), 1996, pp. 458‑464. ↩︎

  11. Santos Sara, Coutinho Diogo, Gonçalves Bruno et al., « Differential Learning as a Key Training Approach to Improve Creative and Tactical Behavior in Soccer », Research Quarterly for Exercise and Sport 89 (1), 2018, pp. 11‑24. ↩︎

  12. Todorov Emanuel et Jordan Michael, « A Minimal Intervention Principle for Coordinated Movement », Advances in Neural Information Processing Systems 15, 2003, pp. 27‑34. ↩︎

  13. Note: it seems obvious that knees caving in are bad. But maybe it’s a valid pattern in specific situations, with certain body types, or with elite practitioners.. At the same time, it’s not evident that instructions are the best intervention. Maybe a CLA approach, or strength and flexibility training would fit better. Therefore we shouldn’t be too quick to overcorrect. ↩︎