Perceptual averaging governs antisaccade endpoint bias.

Antisaccades entail decoupling the spatial relations between stimulus and response and executing a saccade to a target's mirror-symmetrical location. The indirect spatial relations require that a relative target percept supports antisaccade sensorimotor transformations. Here, we sought to identify whether the percept supporting antisaccades results in a respective over- and undershooting bias for the near and far targets within a stimulus-set (i.e., oculomotor range effect hypothesis) or renders an eccentricity-specific bias based on a statistical summary of the individual target percepts in a stimulus-set (i.e., perceptual averaging hypothesis). Antisaccades (and complementary prosaccades) were completed in separate blocks (i.e., proximal and distal) that contained an equal number of target eccentricities, but differed with respect to their magnitudes. The proximal block included eccentricities of 3.0°, 5.5°, 8.0°, 10.5°, and 13.0°, whereas the distal block included eccentricities of 10.5°, 13.0°, 15.5°, 18.0°, and 20.5°. The proximal block showed that antisaccade amplitudes to the central target (8.0°) did not elicit a reliable bias, whereas the block's 'near' (3.0° and 5.5°) and 'far' (10.5° and 13.0°) targets produced an over- and undershooting bias, respectively. Notably, however, the distal block showed a reliable-and large magnitude-undershooting bias for the central target (i.e., 15.5°): a bias that generalized to each target within the block. Taken together, results for the proximal and distal blocks are incompatible with the range effect hypothesis. Instead, results indicate that the visual percept supporting antisaccades is based on the statistical summary of the range of target eccentricities within a stimulus-set (i.e., perceptual averaging). Moreover, perceptual averaging represents a parsimonious basis by which the oculomotor system can specify sensorimotor transformations via non-veridical (i.e., relative) visual information.