Mel Slater's Presence Blog

Thoughts about research and radical new applications of virtual reality - a place to write freely without the constraints of academic publishing,and have some fun.

22 July, 2012

A Very Long Arm Illusion

Kilteni K, Normand J-M, Sanchez-Vives MV, Slater M (2012) Extending Body Space in Immersive Virtual Reality: A Very Long Arm Illusion. PLoS ONE 7(7): e40867. doi:10.1371/journal.pone.0040867


We believe that our bodies are fixed and unchangeable except through the slow process of growing and ageing. Over recent years there have been research results that defy this common sense view - it seems that the human brain will quickly accept gross changes in the body - incorporating external objects such as a rubber arm into the body representation, and even whole bodies seen in virtual reality.

In this paper we add another dimension to this illusion of body ownership. Using virtual reality we show that a virtual body with one very long arm can be incorporated into body representation. An arm up to three or possibly even four times the length of a person’s real arm can be felt as if it was the person’s own arm. This is notwithstanding the fact that having one such long arm introduces a gross asymmetry in the body. An extended body space (a body with longer limbs occupies more volume than a normal body ), affects also the special space surrounding our body that is called ‘peripersonal space’ -a space that when violated by objects or other people can be experienced as a threat or intimacy, depending on the context.

Virtual Body

In our experiment 50 people experienced virtual reality where they had a ‘virtual body’. They put on a head-mounted display so that all around themselves they saw a virtual world. When they looked down towards where their body should be, they saw a virtual body instead of their real one. They had their dominant hand resting on a table with a special textured material that they could feel with their real hand (Figure 1A, B), but also see their virtual hand touching it (Figure 1C,D).  So as they moved their real hand over the surface of this table they would see the virtual hand doing the same.

In fact for 10 of the participants although their real hand touched the table top, their virtual hand did not (Figure 1E, F)- and we did this to create an inconsistency between what they felt and what they saw. This group always saw their virtual arm at the same length as their real arm. For another group who also saw the virtual arm at the same length as their real one, there was no inconsistency(Figure 1C,D) - the real hand touched the surface of the table, and the virtual hand was seen to do the same. This same consistency was kept for three other groups of 10 people each - but one where the table moved away to double the length of the real arm, and the virtual arm stretched to double its length (Figure 2B), another 10 where the virtual arm stretched to three times the true length (Figure 2C), and another group where it stretched to four times the true length (Figure 2D).

We took three measurements
(a) a questionnaire to assess the subjective illusion that the virtual arm was part of the person’s body.
(b) a pointing task, where the arm that did not grow in length was required to point towards where the other hand was felt to be (with eyes shut) (Figure 3).
(c) response to a threat - a saw fell down towards the virtual hand (Figure 2E, F) - and we measured whether people would move their real hand in an attempt to avoid the attack.


What we found, based on these measures, was that people did have the illusion that the extended hand was their own - based on all three measures. Even when the virtual arm was 4 times the length of the corresponding real arm, still 40-50% of participants showed signs of incorporation of the virtual arm as part of their body representation. We also found that vision alone is a very powerful inducer of the illusion of virtual arm ownership - those who experienced the inconsistent condition where the virtual hand did not touch the table, even though the real hand felt the table top, had a strong illusion of ownership over the virtual arm.

These results show how malleable is our body representation, even incorporating strong asymmetries in the body shape, that do not correspond at all to the average human shape. This type of research will help neuroscientists to understand how the brain represents the body, and ultimately may help people overcome illnesses that are based on body image distortions.


For the first time we used the method of path analysis to analyse the results. The most frequently used tools of analysis (ANOVA, regression) are special cases of the general linear model. This requires a single response variable, and postulates a linear model that relates this response to a number of other variables (for example, representing factors in an experiment, or covariates). It assumes an additive normally distributed error term. However, when there are several variables contributing to a phenomenon, there is likely to be a complex relationship between them that cannot be expressed in a single equation: Y may depend on X1 and X2, but X2 may depend on X1 and X3, and X2 also on X3 and so on. Path analysis allows for the possibility of unravelling multiple associations between variables. For example, consider the situation when X affects Y1 and X also affects Y2. Then this could be an example of so-called spurious correlation between Y1 and Y2, since they both have a common dependent (X). With path analysis it is possible to simply test whether the apparent correlation between Y1 and Y2 is preserved even after allowing for the influence of X. This situation occurs in our study, since there are multiple assessments of what might be the same underlying phenomenon - the sensation of 'ownership' over the virtual arm, or even more basic than that - the actual experimental manipulation itself. These assessments were questionnaire, movement in response to a threat to that arm, and also blind pointing direction towards the virtual hand (do you point to the hand at the end of your elongated virtual arm, or towards your real hand)? Path analysis allowed us to separate out the various influences, and present a simple diagram that summarises the findings. 

In psychology especially it is often very difficult to introduce new analysis techniques - preference is towards statistical techniques that are approximately a century old. Things move on though - a large number of statisticians now prefer the Bayesian interpretation (which renders the idea of 'significance testing' meaningless), and path analysis and its more general counterpart, Structural Equation Modelling are widely used. The freedom to explore more modern approaches is really needed.