Fig. 5

Candidate models of the central bias in ensemble statistics. A A non-uniform pooling scheme—i.e., weighted average, in which central elements are given greater integration weights compared to the rest. This model predicts a fixed bias with no changes due to the overall ensemble uncertainty (exemplified in the bottom plot). B Observers combine the representation of the stimulus at the fovea with the statistics of the elements in the parafovea. As the uncertainty in the ensemble statistics increases, the estimates in the parafovea become noisier while the representation of the central elements is unchanged: the central bias increases in strength (exemplified in the bottom plot)