Saturday 23 June 2012

Individual Variation in the Use of Perceptual Information (Specification IV)

If it is the case that perception requires the use of specifying variables, then there should be no individual variation in what information variables people use. However, as we've already seen, such variation exists: the dynamics of a collision event produces multiple kinematic patterns in the optic array, and people judging the mass ratio of colliding balls use all of these, only one of which actually specifies which ball is heavier. Even with training, people do not always find the specifying variable. 

This is an example of how the mapping between a property in the world (mass ratio) and the optic array can be one-to-many, with consequences for perception. Figure 1 on this post shows us that there is another mapping to investigate, namely the one from a perceptual array to the organism. Can this mapping also be one-to-many?  Withagen & van Wermeskerken (2009) suggest that it can, and that again training does not necessarily help.

Dynamic touch and the perception of length
Various papers had identified individual variation in which information variable people used at any given moment in time (e.g. Jacobs et al, 2000, 2001) but most studies use small n's and average over participants. Withagen & van Wermeskerken (2009) trained 25 people to judge the length of rods they were wielding and examined the individual learning trajectories to get a sense of just how varied individuals are.

The task was length perception by dynamic touch (Turvey, 1996); you present someone with a rod and ask them to hold one end and move the other around, without looking. Experiments using this task suggest we perceive properties such as length via the inertia tensor, which describes the rod's resistance to motion in each dimension. Withagen & van Wermeskerken (2009) had participants wield unseen rods and estimate the maximum reachable distance by winding a surface on a rail out the judged distance. As per Jacobs et al (2000, 2001) the data were expressed as correlations between judged length and the length predicted by each of three candidate variables based in the inertia tensor. The invariant was the ratio M/m, where M is the 1st moment and m is the 0th moment of mass distribution. The two non-specifying variables were the first and third moments (I1 and I3) of the inertia tensor separately. 

Participants were tested 7 times (Baseline, then after each training set) using rods where I1 and I3 were poorly correlated with length while the ratio always specified the length. People were trained, however, on sets of rods where the non-specifying variables had moderate correlations to the length. The test set enabled the authors to disentangle use of the three variables, while the training sets simulate the fact that, in the world, there are often several competing variables with moderate correlations (or so the authors claim).

The results showed wide variation; 5 participants eventually settled on the invariant, 10 started using a non-specifying variable and never switched, and the remaining 10 responded to the feedback but never settled. Participants therefore varied in whether, when and how they responded to the feedback. Withagen & van Wermeskerken (2009) claim that this is evidence that individuals vary in their ability to learn information, as well as in which information they end up using (as might be expected from an evolutionary point of view, if learning capacity is a trait with variation acted on by natural selection; Withagen & Chemero, 2009 and the next post in this series).

Comments on the task; it's not ideal. 
  1. The 'information' variables they test are all dynamic quantities, i.e. they all involve mass. While the inertia tensor does seem to be the dynamic world variable being perceived in these tasks, rather than length per se, there is as yet no kinematic specification of this dynamic entity. Perception trades in kinematic variables, not dynamic ones; this is the perceptual bottleneck. The actual information has not yet been identified, and Withagen & van Wermeskerken are therefore confusing the world with information about the world. Behaviour varies with respect to the information for a property, not the property itself and performance must therefore be evaluated against the information (e.g. coordinated rhythmic movement is organised with respect to relative direction of motion, not relative phase). Without knowing what the information for the inertia tensor is, we cannot yet interpret people's behaviour with respect to it.
  2. Part of the variation in learning may simply be an artifact of their training; people only got feedback when judging rods where I1 and I3 had fairly good correlations to actual length. People may only have made small or rare errors, not enough to drive a search for better information. People are very sensitive to the scope of their learning context (Jacobs et al, 2001 and the previous post).
  3. Small errors in judged distance might also not be noticed because the feedback required participants to watch where they moved the curtain and see how that matched where they had moved the surface to. The rods were 30-121cm long, making this a non-trivial judgment in and of itself.
These problems aside, Withagen & van Wermeskerken did show wide variation in performance; people were clearly not simply doing the same thing (perceiving the specifying variable) using identical perceptual systems. However, given the problems, this paper is very weak evidence about the use of non-specifying information which works against the author's slightly grandiose claims to the contrary. The empirical case supporting the perception of non-specifying variables remains interesting but far from convincing.

Turvey, M. T. (1996). Dynamic touch. American Psychologist, 51(11), 1134-1152. Download

Withagen, R., & van Wermeskerken, M. (2009). Individual differences in learning to perceive length by dynamic touch: Evidence for variation in perceptual learning capacities Perception & Psychophysics, 71 (1), 64-75 DOI: 10.3758/APP.71.1.64


  1. Without knowing what the information for the inertia tensor is, we cannot yet interpret people's behaviour with respect to it.

    I have been having trouble with this for a while now. At the last few Eco Psych meetings, Turvey has been talking about the information for kinesthetic perception as if it exists physically inside the organism. Having probed in Q&A, he means this very seriously. I have not been able to firm up my thoughts about this move. Certainly it is risky. It seemed quite important to the original structure of Gibson's theory that the information be external to the organism. It was, he asserted, one of the major virtues of looking at the ambient optic array, as opposed to the retinal image. Returning to the inertial tensor: One could say that the information is in the resistance the rod gives to certain movements. This seems obviously fine, so long as the resistance is "out there" and not at the internal joints. I just don't know what it would mean for an Ecological Psychologist to assert that the information was at the joints.

    Most at the conferences have seemed oblivious to the potential problem, they seemed to consider the matter fully settled once Turvey had spoken. (Turvey, in contrast, was perfectly willing to discuss the issue at length.)


    P.S. I'm still confused about what you would accept as evidence for use of non-specifying variables. While you may dispute how common findings like Withagen & van Wermeskerken's would be "in nature", surely they show evidence that people do use non-specifying variables... because, well... that is what they show. Isn't it?

    1. Here's what I think about dynamic touch; the inertia tensor is the dynamic property often being specified. That property will produce kinematic effects in the haptic array; something analogous to optic flow. This flow will simply occur in tendons and muscles, rather than in light; other than it's location, I don't think there will be a difference in kind in the flow.

      It is then possible that the inertia tensor will produce different patterns of haptic flow in different tasks, or that there will be multiple patterns all the time and different patterns specify different properties (length, throwability), but are all created in response to how the object resists motion. I don't know enough about the maths to formalise this though; but this is my basic objection to this paper. W&vW say that people should be responding to the dynamic property but they will actually be responding to the kinematic consequences of that dynamic property and there is therefore legitimate room for unexplained variability when your analysis is one step removed like this.

      So the information will be implemented in motions within the organism, and I don't see how this is problematic. Why is the location so key?

    2. P.S. I'm still confused about what you would accept as evidence for use of non-specifying variables. While you may dispute how common findings like Withagen & van Wermeskerken's would be "in nature", surely they show evidence that people do use non-specifying variables... because, well... that is what they show. Isn't it?
      People can indeed be seen to be using a variable that doesn't specify the property in question. BUT given that the best evidence for this is in a judgment task for an odd property (mass ratio in collisions) I'm just as yet unconvinced that this isn't an artefact of the experimental setup. This paper doesn't add to the case for me because of the problems I lay out.

      The empirical approach is the right idea, I'm just unimpressed with the state of the art, if that makes sense.

    3. "Why is the location so key?"

      I am not sure that it is key, but I am sure that Eco-Psych theorists of the past have thought it was key. The question is whether they were correct about location being key, or if there was some other factor that was important, which (in the modality they were focusing on) went along with location.

      To avoid contrasting vision with touch, which is a mess, let's wonder for a second about hearing:

      Is the "information" for hearing in the ambient sound waves, or in the vibrations in the bones of the inner ear?

    4. OK, interesting; I see where you're going.

      For hearing, I think the information is in the sound waves because that's where the structure that was caused by the world is. For haptics, I think the information is in the behaviour of the receptors for the same reason. I have always had the feeling that haptics is interesting because of this.

      The alternative is to think that the information is in the motion of the object according to it's inertia tensor, which may also be viable. But that's again a slightly odd place to put it.

      Interesting thought, though, I see the problem!