Misestimation of the perceived shape of disparity-defined surfaces is consistent with disparity-gradient rather than slant bias

Observers overestimate vertical distances relative to horizontal distances. This bias has been interpreted as reflecting the probability distributions of distances found in the natural environment (Howe and Purves, 2002 Proceedings of the National Academy of Sciences of the USA 99 13184 - 13188). We investigated such biases for stimuli in which depth information was available from binocular cues and whether this is best described in terms of properties of the environment (slant) or of retinal images (disparity gradients). By varying the aspect ratio and slant of rectangular planar surfaces defined by sparse random-dot stereograms, we measured the shape that appeared square to the observer, for each slant tested. Results were expressed in terms of effective slant across a range of viewing distances. Effective slants were significantly biased towards intermediate values of around 30°, in the same direction as the typical horizontal - vertical illusion. This value increased with increasing observation distance. We conclude that biases in the perceived shape of disparity-defined surfaces are most parsimoniously accounted for at the level of disparity-gradient processing, prior to a stage of distance scaling.