Tuesday, 23 November 2010

What Does Coordinated Rhythmic Movement Have To Do With Anything?

In which I provide an answer to a question I get asked by everyone, including grant reviewers, students and random people who make the mistake of asking what I do for a living.

I've spent numerous recent posts talking about coordinated rhythmic movement. This is my bread-and-butter experimental task, my go-to example for studying all aspects of perception, action and learning. I'm branching out, now I have my own faculty position, but coordination is where it's at. 

The single most common question I get is "why study this? Surely it's just some fake movement task; I mean finger-wiggling, who ever even does that?" I wouldn't mind so much, but I even get this question from grant reviewers, scientists who should know the answer. Doing science properly is important, but communicating that my methods achieve this matters too, not least because today's funding climate demands it.

So, coordinated rhythmic movement: what the hell?

The most important thing people need to know is that, in a very specific sense, I could not care less about coordinated rhythmic movements. I don't study them because they are intrinsically interesting. They do, however have some useful properties which make them an ideal model system for the laboratory study of perception, action and learning. 

Model systems
In science (biology especially) a model system is a version of the thing you're actually interested in that you can use in your experiments. These are often animals; so, for example, a neuroscientist colleague here at Leeds studies spinal cord injury and recovery and does many of his experiments on rats. He can create controlled lesions, then test promising drugs and behavioural interventions such as locomotor training before taking them anywhere near a human. The point is that rats (the model system) and humans (the system of actual interest) overlap in terms of what the spinal cord contributes to locomotor control, and so, within some limits, what he learns from his experiments on rats transfers to humans. 

The primary reason to study the model system and not the system of interest is experimental control. There are often practical and ethical issues* as well (you can't deliberately lesion a human, for example, and spinal cord injury patients are hard to come by for research purposes). But the main reason is scientific: the model system is generally simpler, which helps when it comes time to interpret your results.

Perception/action systems
The human perception/action system contains massive redundancy, which affords it tremendous flexibility (while also posing interesting control problems). Think about locomotion, for a moment - walking. Imagine walking over rocky terrain, at varying speeds, while avoiding low hanging branches. A skilled walker can use the range of motion of their knees and hips and varying levels of muscular force to change their stride length from step to step, or to alter their height briefly to move under a branch. If a rock moves from under you, perturbing your stepping motion, you can often recover swiftly from even a large perturbation by compensating for the unexpected motion using new joint configurations. 

The entire point of the perception/action literature is to uncover how we achieve this flexibility in the face of perturbations from a noisy world. But if you started out studying locomotion in the wild, you'd swiftly run into two related problems: too much data and no clear idea what any of it means. So you start small, and slowly, and build up your knowledge of the general principles by which human movement is organised, and apply those to increasingly complicated versions of your task. Coordinated rhythmic movements just happen to be an ideal simple model system within which you can sensibly develop all the required techniques and skills. 

Coordinated rhythmic movements
I've already discussed the basic structure of coordinated rhythmic movements, and the two theoretical approaches to explaining that structure (the dynamic pattern hypothesis, and the perception/action hypothesis). Given that I don't think the former does any explanatory work (it's purely descriptive and has had numerous key empirical failures) I'm going to focus on how this task is a good model system for a perception/action approach.
  1. There are easy, valid measures of performance: as I explained in the first coordination post, the primary measure for this task is mean relative phase. This variable is straight forward to compute (from trigonometry, it's the arctangent of velocity over position**) and has a very straight forward validity to it: it is, precisely, a measure of the coordination between two rhythmically moving harmonic oscillators. It's a circular variable, which causes some analysis problems, but these are readily resolved.***
  2. According to these measures, performance in the task is structured: if performance was the same across relative phase, this task wouldn't be that useful. Science needs structure, differences, to get it's teeth into; these are the cracks at which you can prise something apart to see what's going on under the hood.
  3. The task dynamic is complete: while it's true that the task we study in the lab is a very simple action, it is complete in the sense that the task dynamic contains all the essential features of a perception/action task, and is thus a valid model system. Those features include
    •  voluntary control of limb movements
    • coordination of those limbs with each other, other people or other events in the world (e.g. a display)
    • online perceptual control of this coordination, i.e. informational coupling
  4. The experiments are easy to run: once set up, you can run endless permutations, manipulating all and any variables you think might be relevant for uncovering the control laws. For example, I have programmed an extensive Matlab toolbox to run my studies, which gives me complete control over the computer feedback displays and enables me to record data from various input devices. They can be a little time consuming if you are doing learning, but there's no getting away from that.
  5. You can study performance and learning within the same task dynamic: often, when studying learning, you need to provide people with a task you can guarantee they've had no exposure to; this usually means something a bit whacky, such as trying to produce a specific pattern of fingertip forces. Coordinated rhythmic movements come in a variety of types (ones you can do, ones you can't) that are all examples of the same basic task dynamic (although learning does extend this dynamic; see future posts). This means you can compare performance across them usefully.
  6. The numbers that come out are straight forward to interpret: the measures we produce (average and variability data about relative phase within and between a trial) simply and validly measure what we want to know: how stable was the coordinated movement that was produced? This means we can interpret the consequences of our manipulations (e.g movement frequency) for coordination stability and come to understand what this is telling us about the mechanism underpinning performance. There are several ways to produce these values (related to the fact that relative phase is a circular variable) which I've addressed in the literature and will discuss here as well.
  7. The task dynamic is amenable to modelling: my last post was detailing Bingham's perception/action model of bimanual coordinated rhythmic movements, as well as the modelling strategy. This strategy took the results of various experimental perturbations and used these to constrain the selection of the components that could be included in the model. This worked because of (6) above, and also because the components are simple enough that our dynamical systems toolbox contains suitable bits. The oscillators that we rely on are simple enough to be well understood; there are only a small number of such well understood pieces in dynamics and more complicated tasks would require more complicated elements to which we simply might not have access. This task is simple enough to serve as a test bed for the modelling strategy and we can use it to learn the process of building these types of models. Then, as the tools are developed that will allow us to attack more complicated problems, we will have a clear and readily understood example of the process we need to go through to get the model that we want. I know that I rely heavily on this task as an example whenever I'm approaching a new task. 
So what the hell? 
So what has this toy task got to do with real movements? Everything that matters, plus the advantage of being accessible in the lab. Empirically I use this task to study questions of perception/action learning, changes in this learning with healthy ageing as well as stroke, and any question in which the assembly, coordination and control of a perception/action task dynamic is the system of actual interest. I don't use it to actually try and rehabilitate stroke survivors (Grant Reviewer #2, I'm looking at you just now) and I don't think this task per se will help make older adults less likely to fall over (grant reviewers on my upcoming submission, this one's for you). But this task will help me ask rigorous questions and gain valid answers so I can understand the mechanisms of perception/action and how these change with, for example, old age. Only then can I address those other, perhaps more interesting questions.

And that is what science is all about.
*Obviously there are also ethical implications for using animal model systems, but I don't especially want to get into that here because it's not what I do.
**This deserves a post on it's own. It's a nice example of how, in science, you can build your own variables using simple techniques.
***This should get a post too. It actually took the field some time to start addressing the problems caused by this circularity and the solutions have been fun to come up with.


  1. Andrew, its seems I can't post a message to your blog post on Dr Daryl Bem and his psi research. I've posted my response @ Against Open Mindedness (Practical Ethics Blog).

    I wanted to debate with you, but it seems like that you don't want to debate.

  2. I'd put my money on a blog website error. Andrew does invariably want to debate all aspects of everything with everyone who will listen!

  3. Just out of curiosity, does the phase shifting from 180 to 0 deg behavior manifest in any other task? I wonder if this particular, and rather showpiece, factor has hamstrung the generalizability of the paradigm for those not involved in RBC research..

  4. Gav, I'm not sure what you mean. Phase transitions (qualitative/discontinuous changes in behaviour with continuous scaling of a parameter) happen in all kinds of behavioural systems (another classic is the transition from walking to running with increases in speed. The specific 180 -> 0 shift only happens in rhythmic movement tasks.

  5. @Falafulu, definitely an error, I'm not deleting anything and never would unless it was spam. Feel free to try posting it again, if you get an error email it to psychscientists@gmail.com. I'll check your link.

  6. yeh - I was referring to the specific 180 to 0 transition, but your point about the walk to run transition is well taken

  7. It has to be a rhythmic behaviour for relative phase to be defined. Clapping is another good one; people begin to clap in sync, which lowers the volume,so they break it up to get louder. This is why clapping ebbs and flows like that.