Perception during interaction is not based on statistical context

To evaluate the central tendency we plotted the reproduced lengths against the presented length and measured the regression index, defined as the difference in slope between the best linear fit of the data and the identity line. This index varies from 0 (veridical performance) to 1 (complete regression to the mean). In the control "alone" conditions, all subjects showed on average a clear central tendency mechanism (average regression index 0.417 ± 0.028) which decreased significantly when the task was performed with the robot (0.156 ± 0.028, one-tailed, pair-sample t-test, p<;0.01). Part of this common decrease can be explained by the difference in the richness of the two kinds of stimuli: on the one hand just two brief flashes of lights represented the length to be reproduced, while in both robotics conditions the whole arm motion was visible, providing a richer stimulation. Indeed, according to the Bayesian models described in [1,4] the presence of less sensory noise would yield to less regression to the mean. However, the amount of change in regression was different between the two robotic groups.