Abstract
Ensemble coding is the ability to compute the statistical properties of an ensemble of stimuli. Despite the rapidity and efficiency of ensemble coding, recent work has shown systematic spatial anisotropies by which individual stimuli at different locations are differently weighted. For instance, in size and orientation averaging, stimuli in the central and left visual field contribute more to the estimated mean than stimuli at other retinal locations. Here, we investigated whether the degree of anisotropy relates to the uncertainty about the ensemble of stimuli. In three experiments, we presented an array of 25 oriented Gabor patches centered on the fovea, arranged in a square of 12x12°, and asked participants to reproduce the perceived average orientation. In interleaved sequences of trials, we varied the orientation range of the ensemble (Experiment 1, online; range = 5°, 10° or 15°), the ensemble duration (Experiment 2, online; durations: 0.1, 0.5 or 1 sec), and the relative temporal onset of each Gabor (Experiment 3). Increases in uncertainty —i.e., wider orientation ranges and shorter durations, led to evident deteriorations in the precision of ensemble coding but no evident changes in anisotropy patterns. In particular, the central bias was resilient to changes in the ensemble duration and to the prior entry of stimuli at other locations. We discuss how anisotropies may dominate under conditions where the uncertainty about the upcoming stimulus cannot be anticipated (e.g., in trials with interleaved levels of uncertainty). Under these conditions, spatial anisotropies may reflect a stereotyped compensatory strategy to accumulate more evidence from locations where the resolution of the visual system is higher.