Wednesday 9 December 2015

Quantifying the Affordances for Throwing for Distance and Accuracy

I have a new paper in press at JEP:HPP (Wilson, Weightman, Bingham & Zhu, in presssupplemental material). It is the end result of five years work across two jobs, and it has involved kinematic data collection from expert throwers in Leeds and Wyoming, analysis of that data, then interpretation of that data in the context of detailed simulations we ran in order to identify the affordance property of the target structuring behaviour. This is my first paper on affordances, my first about my current favourite topic of throwing, and probably the heftiest empirical piece I have ever done, so getting it published in my journal of choice is pretty exciting!

I'm going to just lay out the basic framework of the paper here. I will leave the (very many) details to the paper. The paper consists of two experiments, a series of simulations, and a discussion of affordances as dispositional properties of tasks best described at the level of task dynamics. This last bit feeds into the argument in the (mostly philosophical) literature on the nature of affordances; bad news, people who think they are relations - they aren't, and I've got two experiments that back that up!

The topic
This work is done with Arthur Zhu and Geoff Bingham. These two have been studying the affordances of throwing for maximum distance for some years now (Geoff started in 1989 and then he and Arthur picked it up for Arthur's PhD at Indiana). I have blogged a bunch of this work. I came on board when I was at Leeds and ran these studies to expand the work to throwing to hit a target.

Throwing is a fabulous task for two main reasons. First, humans are the only animals to specialise in it, which is fun (and important to my other ongoing project in throwing which I will talk about later). Second, and more importantly, it's a task dynamic that we can study in the lab without scaling down very much. The task of throwing is the production of a projectile motion that satisfies some task demand (maximising distance, intercepting a target). The dynamics of projectile motion are well worked out and are the kind of thing you can simulate usefully. So this is a fairly complicated task I can measure and analyse in great detail and with a lot of control, all while preserving the ecological validity of the task. 

The experiments
The paper consists of two experiments. 

In the first, we tested expert throwers; male baseballers and female softballers in Wyoming with Arthur, and male cricketers in Leeds with me. Participants threw a regulation tennis ball to hit a vertically oriented 4x4ft target placed at one of three distances (5m, 10m and 15m) and one of three heights (target centre at 1m, 1.5m and 2m). We filmed the throwers from the side with a high speed camera, and measured the release angle, speed and height of the ball at release. These are the outputs of the throwing dynamics, and are the parameters for the projectile motion that is the goal of the throw. 

We first just wanted to know what the throwers did under these conditions. What we found was that they scaled their throws to suit the target; release velocity increased with distance, release angle increased with both distance and height. In practical terms, the experts produced throws appropriate to the target location and that produced the straightest trajectory ('throwing a rope'). This all means they were perceiving the current task demands rather than simply producing a single highly practiced action that might fit the bill. 

In the second experiment, we tested male cricketers in Leeds throwing to the same three distances but just the one height (1.5m) with the target oriented either vertically or horizontally. The horizontal orientation forced throwers to adopt a different mode of throwing; not fast and flat as they preferred, but slow and high. We also tested monocular and binocular vision, as a simple initial informational manipulation. 

We replicated the results from Experiment 1 and also found that the throwers readily switched to produce the different throwing mode without being told to (a mode the athletes don't prefer and won't have extensive practice in, again evidence that they are shaping their behaviour to the perceived task demands nicely). All throws were again scaled to the target location. The monocular manipulation only made the throws slightly more conservative (throwers reduced release speeds a little) but equally accurate (they compensated perfectly with increases in release angle). 

The simulations
This is all very well, but what exactly are the task demands that were making them throw the way they did? In ecological psychology, what the task offers are it's affordances, so we set out to quantify a description of the target in affordance terms to see if we could explain why the throws came out the way they did, rather than simply describe them.

To do this, we ran simulations of the projectile motion of a tennis ball towards our target. We varied the release parameters (speed from 0m/s to 45m/s; angle from -30° to 90°; height from 1.5m to 2.7m; ranges selected to match what we saw in the experiments) and also the distance and height of the target specified in the model. For every parameter combination, we simulated the resulting projectile motion and figured out whether it intercepted the target or not. (I had a couple of laptops working non-stop for about 10 days to run these!)

For the vertical targets, the results looked a little like this:
Figure 1. Example affordance subspace for a vertical target at distance = 10m, height = 1.5m, and release height = 2m
For the horizontal targets, the results looked a little like this:
Figure 2. Example affordance subspace for a horizontal target at distance = 10m, height = 1.5m, and release height = 2m
(You should check out the paper and the supplemental PowerPoint containing animations for more detail on how these change with target location and context).

The large blue region codes misses. This task is hard; most of the parameter space leads to misses. The coloured region codes the hits; this subspace is the set of parameter combinations for each target condition that produces a hit. This, we argued, represents the affordance of each target to be hit by a tennis ball undergoing projectile motion. 

We then evaluated the human performance relative to the affordances. We found that the human data was clearly organised with respect to the affordance (our experts chose to throw in a most stable regions of the space, for example). As the target went from vertical to horizontal, the location of this stable region moved (from the large prong on the left to the narrow bit on the bottom of the respective subspaces); human performance did the same. Within the resolution of our simulations, we accounted for why people threw the way they did. 

The bigger picture: affordances are dispositional properties
This experiment and analysis were all derived from a particular conception of affordances, as properties of tasks that must be described at the level of dynamics. The success of the analysis licensed us to take a fairly firm stand on this, so in the Discussion section we spent some time defending this view from the opposing 'affordances are relations' view which has become trendy in more philosophical accounts of affordances (see here and here for me on Chemero on this claim, plus these three posts on the basic debate). 

The simulations only include information about the projectile and the target. The resulting affordance is created by these properties in the context of the task of generating a projectile motion. The affordance is not created by the presence of an observer, as is required by the 'affordances are relations' claim.

Why does this matter? We claim that the affordances we identified are doing work to structure the selection and execution of the throws we recorded. In order to be able to do this work, they have to exist independent of the observer. If the affordance is somehow created by the interaction of the thrower with the target in the task context, then we are left with the question of 'why did they create this affordance?' and we're back with no explanation for the problem of action selection and control.

Far more importantly, if affordances are relations, then neither evolution nor learning can use them for anything. Imagine the novice thrower; they are trying to learn which actions complement the task demands (the affordance) but if those demands are constituted in part by their current ability, those demands can never lead them to do anything different. Learning would have to be driven by something other than affordances, even though affordances are what we want the novice to learn to perceive! The debate is therefore quickly resolved as soon as you try to use affordances to drive either science or behaviour. 

Why, then, do people want affordances to be relations? We got a lot of comments from a reviewer that summed up the reasoning. The basic idea is that actions show individual variation and can fail; the organism is clearly very much in the mix and so this implies a relation is present. This is, of course correct; but the relation is not the affordance. The relation is the act of perceiving the affordance. Perception is the act of placing yourself in a particular relation to a task property like an affordance, and all the 'relational' work that needs to happen happens then. The problem, as always, is that people have read Gibson, got excited about affordances but didn't notice that it was information that does the heavy lifting in his theory.

[A side note: I've realised that people who advocate affordances as relations are all people trying to expand the scope of ecological psychology by expanding the notion of affordances to be social, linguistic, etc. Affordances are cool, but they are a particular kind of thing and can't help expand ecological psychology. Our move is to expand the notion of information in ways that preserves Gibson's key insights but that allows behaviours other than 'mere' perception-action (Golonka, 2015). Information is the thing where the necessary 'relational' work happens.]

This is stage one of an extended research programme on the perception-action mechanisms that underpin the uniquely human skill of throwing for very long distances with great accuracy. There is so much more to do; I have data from a learning study to work on next, a project to examine the dynamics of the thrower in more detail in the pipeline, and we have barely begun to think about the information involved in the perception of this affordance. Happily for me, I will be able to do this work for as long as I want and then some!

Wilson, A. D., Weightman, A., Bingham, G. P., & Zhu, Q. (in press). Using task dynamics to quantify the affordances of throwing for long distance and accuracy. Journal of Experimental Psychology: Human Perception and Performance.  Download (pre-publication version), Supplemental Material


  1. I hope this post will help me win a bar argument I was having this weekend on the afforances as relations vs. affordances as information point--I wasn't eloquent enough to state it as "affordances are information".

    I study social interaction, and so I feel as though there are a lot of moving pieces creating context. I wonder if you might speculate as to how changes over time in the task space could influence things. If people had to throw at increasingly distant targets vs. increasingly close targets, is that a feature of the environment that would change the affordance? (I am trying to generalize this to facial expressions changing from smiling to angry or vice versa)

    1. Affordances are not information. They are, however, the kind of dynamical properties that can create information. They can also change over time; as the target changes distance the affordance subspace would smoothly and continuously go from the 5m one to the 10m one and so on.

      I'm all for my papers helping to win arguments in bars though!